Resources for teachers
Cloud and infrastructure in the classroom: a teacher's guide
Everything a teacher needs to bring cloud and infrastructure to life: games you can run tomorrow, from the arithmetic of 'nines' to a data hall that loses its power, a rack you have to fill without overheating it, three true stories read from the documents the companies and investigators wrote, and the apprenticeships that lead in, at a desk and on site. Every figure is sourced and every game says what is made up.
For teachers. Every figure is sourced and was checked on 5 November 2026. Every shop, hall, server, price and failure rate in the games is made up for the lesson, and each game says so. The walk round the building and the role plays use things in your own building: choose the route in advance with your site manager or IT technician, keep students out of plant rooms, electrical cupboards, roofs and anywhere that is not open to them, and do not let students handle batteries, generators or live electrical equipment.
The activities
Thirteen activities, balanced between the desk and the plant room
There are thirteen activities. Three need no devices at all, and ten are interactive on this page. Five start at a desk with a sum, four start in the building where the real power, cooling and cabling are, and four suit both, so a teacher with a student who likes to design the system and a teacher with one who likes to keep the hardware running both have something to run on Monday. Timings are our own suggestions from how long each takes to run, not measurements. The teacher pack has the full notes.
Works with no devicesStarts on site
Where is the cloud?
Trace one saved photo from a phone to the building where a copy sits, then walk your own building to find the power, the back-up, the cooling and the cables that every data hall needs too.
25 to 35 minutes
Needs a screenStarts at a desk
How many nines?
99% sounds like nearly perfect. Work out how long a year of 1% down really is, then keep the error budget of a made-up shop through a bad month.
20 to 30 minutes
Needs a screenStarts at a desk
Spread it across two
A made-up shop has three layers and one building. Add spare machines and a second building to reach 99.99%, and find out why more machines in one building stop helping.
25 to 35 minutes
Needs a screenStarts on site
Lights out
The mains power fails in a made-up hall. A battery has minutes, a generator has to start and take over, and the fuel has to last. Choose the set-up and see whether the hall stays up.
20 to 30 minutes
Needs a screenStarts at a desk
Black Friday
A made-up shop has a quiet night, an evening peak and a sudden spike. Meet it with a fixed number of servers or let the number follow demand, and compare what turns away and what it costs.
25 to 35 minutes
Needs a screenStarts on site
Keep it cool
Everything a data hall draws that does not reach the computers is a cost. Work out what a point of PUE is worth in energy, carbon and money, then find out how hot is too hot for the equipment.
25 to 35 minutes
Needs a screenStarts at a desk
By hand or by pipeline
Twenty steps, each with a one in a hundred chance of a slip, is not a one in a hundred chance. See how small chances add up, and what a script changes and what it does not.
20 to 25 minutes
Needs a screenStarts on site
Rack and stack
A rack has 42 units of space and 8 kW of power it may draw. Fill it from a made-up shelf to get as much done as you can, then see the best possible answer.
25 to 35 minutes
Needs a screenStarts at a desk
Rent by the hour or by the year
A made-up machine costs 10p an hour or a lower price for promising it all month. Which is cheaper depends on how much you use it, and the crossing point is a share of the month.
20 to 25 minutes
Works with no devicesSuits both
Who looks after what?
When a company rents computers from a cloud provider, some jobs stay with the provider and some stay with the customer. Sort six jobs into the right pile, using what the National Cyber Security Centre says.
15 to 20 minutes
Works with no devicesSuits both
Three true stories
A fire in a French data centre, a cooling failure in a London heatwave and a bug that emptied one record at the heart of a huge cloud. Read what each company and each investigator says, and only that.
30 to 40 minutes
Needs a screenSuits both
Eight kinds of job
Cloud and infrastructure is eight areas of work, from designing systems to keeping the cables and cooling running. See the jobs in each, the pay with its source and the routes in.
20 to 25 minutes
Needs a screenSuits both
Three ways in
A university degree, an apprenticeship at a desk and an apprenticeship on site, side by side, with the same information for each from the people who set the standards.
20 to 25 minutes
Download the teacher pack. Timings, step-by-step instructions, discussion questions, answers and sources for every activity, as a PDF or an editable Word file you can adapt.
Why it matters
The case for teaching it
We could just say that cloud and infrastructure matter and leave it there. But you are the one standing in front of students who will check, so here is what the evidence says, including the awkward parts, because the awkward parts are where the good lessons live.
It is a big, physical thing. The Department for Energy Security and Net Zero estimates that data centres used 4.5 TWh, which is 2% of the electricity consumed from the grid in Great Britain in 2024, and that around three quarters of it was in the South East and London. The same article says there is no universally agreed definition of a data centre, so counts and totals in other sources can differ, and they do: the National Energy System Operator estimates demand at 7.6 TWh from 2.4 GW of connected facilities, where DESNZ's meter-based estimate for 2023 is 4.1 TWh. The government classed UK data centres as critical national infrastructure in September 2024.
The demand for more is the story of the next few years, but be careful with the numbers. The International Energy Agency says data centres used around 1.5% of the world's electricity in 2024 and that consumption could reach around 945 TWh by 2030 in its base case. Ofgem says around 73 GW of data centre projects are in the grid connection queue, and NESO says that pipeline is likely to significantly overstate demand. A request for a connection is not a data centre.
It is also a market with a few very large names. The Competition and Markets Authority found that Microsoft and AWS each had between 30 and 40% of infrastructure as a service in 2024, across the UK and the European Economic Area, and that competition is not working well. Ofcom had said in 2023 that the two had a combined 70 to 80% in the UK in 2022. Those two figures measure different things in different years, so do not put them side by side.
Are there jobs? The honest answer is that we found no government figure for cloud or data centre vacancies. The Uptime Institute's 2025 survey of data centre operators worldwide found that 46% report difficulty finding qualified candidates, and that shortfalls in electrical and mechanical trades continue to be significant. That is an industry survey and not a UK one. The government's AI Growth Zones paper projects more than ten thousand jobs from its programme, which is a projection and not a count, and the Data Centre Covenant expects developers to invest in paid apprenticeships and training places without giving a number.
One more for the staff room. The Department for Science, Innovation and Technology's 2026 business data survey found that 31% of businesses that handle digitised data say they use a public cloud provider to store and process it. The question is about data storage and covers sole traders as well as big firms, so it is not the same as 31% of businesses using any cloud service. And the Technology Code of Practice tells government technology projects to consider public cloud first, which is guidance for government and not a rule for everyone.
Closer to home
Your school or college is a customer too
Every school and college depends on infrastructure that someone else may run: the network, the learning platform, the email, the timetable, the lunch payment system. A good first question for your IT technician is which of those run on computers in the building and which on someone else's. It makes a better teaching resource than any case study, because the answers are real and students can see who is responsible.
On who is responsible, the National Cyber Security Centre says there is a joint responsibility for the security and availability of data and workloads in a cloud service, shared between the provider and the customer. It says the provider will almost always be responsible for the availability and physical protections of its servers, and that the customer is always responsible for setting up the services it chooses securely and for deciding what data to store in them. When things go wrong it is not rare: the Department for Science, Innovation and Technology's survey found that 20% of businesses that store data away from their premises said they had been affected by a server or cloud outage in the last 12 months.
A student who likes the hands-on side should know it comes with rules. The Health and Safety Executive says people working on electrical equipment must be competent to do so, and those who cannot show it should not work unless supervised by someone who can. Regulation 14 of the Electricity at Work Regulations 1989 says nobody should work on or so near a live conductor that danger may arise unless, among other tests, it is unreasonable for it to be dead. The HSE also says falls from height are one of the biggest causes of workplace deaths and major injuries. None of that is specific to data centres, and the HSE has no page on data centres that we found.
Starter: where is the cloud?
Before you say the word cloud, ask students where the last photo they saved is. Give them two minutes to answer. Most will say on their phone. Some will say in the cloud, and a few will say they have no idea. All three are fine answers.
That is the whole opportunity of this subject. The cloud is not a cloud. It is somebody's computers in a building that needs electricity, a way of getting rid of the heat, and cables in and out, and somebody's job is to keep all of that working. Some of those jobs are done at a desk and some are done on site with a hard hat on. This page is built to show both, and neither comes first.
Where is the cloud?
Trace one saved photo from a phone to the building where a copy sits, then walk your own building to find the power, the back-up, the cooling and the cables that every data hall needs too.
25 to 35 minutesPaper and pens, and a short walk led by your site manager or IT technician.
What this is
A short trace and a short walk, with no devices. First you follow one saved photo from a phone to the building where a copy sits. Then you walk your own building with the site manager or the IT technician and find the things every data hall needs too.
What to do
- In pairs, think of the last photo you saved or sent. Write every place it passed through on the way, and every place a copy might be kept.
- Share your lists. Add anything missing: a mast or a router, cables, and a building full of computers called a data centre.
- Walk your own building with your teacher and the site manager or the IT technician. Find where electricity comes in, anything that gives back-up power, how rooms are heated and cooled, and where the network cables come in. Look, ask and write. Stay where you are shown.
- Back in the room, write who looks after each of those things, and what would stop working if one of them failed.
What you should notice
The cloud is other people's computers in real buildings, and those buildings need what yours needs: power that does not go off, air that stays the right temperature and cables in and out. Somebody looks after each one.
How to run it, questions to ask, and what to watch for
Running it
- Ask students to think of the last photo they saved or sent. In pairs they list every place it passed through on the way, from the phone to the other person's phone, and every place a copy might be kept.
- Share the lists on the board. Add what is missing: the mast or the router, the cables, a building full of computers called a data centre, and the electricity and cooling that building needs.
- Show the page's figures for where UK data centres use their electricity. Ask why they might be in some places and not others.
- Walk your own building, led by the site manager or the IT technician. Find where electricity comes in, anything that gives back-up power, how rooms are heated and cooled, and where the network cables come in. Look, ask and write. Photograph only where you are allowed to.
Questions to ask
- Which parts of the photo's trip are buildings and cables, and which are software?
- If one of those failed, what would stop working?
- Who looks after each part in your school or college, and what happens when it goes wrong?
Watch for
- Students who think the cloud is not a place. It is computers in buildings that someone owns and powers.
- Plant rooms, electrical cupboards, roofs and anything locked. Choose the route in advance with the site team and keep students out of anywhere that is not open to them.
- Counts of UK data centres. The sources disagree because they count different things, and the page shows the range and says why.
Students will be able to: say that the cloud is other people's computers in real buildings, and name what those buildings need. name where electricity comes into your own building, what gives back-up power and how the rooms are kept comfortable. say who looks after each of those things in your building.
Lab: how many nines?
How many nines?
99% sounds like nearly perfect. Work out how long a year of 1% down really is, then keep the error budget of a made-up shop through a bad month.
20 to 30 minutesA projector for the whole class, or one device per pair.
What this is
A calculator for how much downtime a target allows, and then a month in the life of a made-up online shop. You see how a promise such as 99.9% turns into minutes, and how an error budget gets used up.
What to do
- Under Pick an availability target, choose 99%. Read how long the service may be down in a day, a month and a year. Then choose 99.9% and 99.99%.
- Before you choose 99.99%, guess its allowance for a year. The answer is 52 minutes 34 seconds.
- In part two, read the shop's budget. Tick A bad update goes out and is rolled back, The main database fails and the spare takes over and The shop's payment provider has an outage, and read how much is left.
- Now tick A security certificate expires and nobody notices. Read what happens to the bar.
What you should notice
At 99% a year of downtime is 3 days 15 hours. At 99.9% it is 8 hours 45 minutes. Each extra nine cuts the allowance to a tenth. With the first three incidents the shop has 2 minutes 42 seconds to spare in a month, and the fourth takes it 35 minutes 18 seconds over.
Part one: pick a target and see how much downtime it allows. Part two: a made-up online shop has a bad month. How many of its troubles can it afford?
Part one: what a target allows
| In a | The service may be down for no more than |
|---|---|
| Day | 1 minute 26 seconds |
| Month of 30 days | 43 minutes 12 seconds |
| Year of 365 days | 8 hours 45 minutes |
Each extra nine cuts the allowance to a tenth. The jump from 99% to 99.9% sounds small and takes a year of allowed downtime from days to hours.
Part two: the error budget of a made-up shop
The shop promises 99.9% over a 30 day month. That leaves 43 minutes 12 seconds of trouble it can afford. Tick the things that happened this month.
Nothing has gone wrong yet, so the whole budget is still there.
How to run it, questions to ask, and what to watch for
Running it
- Show the page on a projector. Start at 90% and ask whether a service that works 90% of the time is good. Step up through 99%, 99.9% and 99.99%, reading out the time allowed in a year each time.
- Before you move from 99.9% to 99.99%, ask the class to predict the new allowance. Then click.
- Move to part two. Read out the shop's target and its budget. Have the class vote on each incident before you tick it.
- Ask what a team should do when the budget is used up, and what it can do while some is left.
Questions to ask
- Why would a company not promise 100%?
- Would a customer on a poor mobile signal notice the difference between 99.99% and 99.999%?
- Who decides how much downtime is acceptable, the engineers or the business?
Watch for
- Students who read 99% as nearly perfect. A year of 3 days 15 hours offline is a long time.
- Sources round these figures differently. Google's table gives 52.6 minutes a year for 99.99%, and the page works out 52 minutes 34 seconds from a 365 day year. Both are right to their own rounding.
- Google's error budgets run over a quarter, and the page uses a 30 day month so the sums are easy to check. A Tier from the Uptime Institute is not a promise of nines, because Uptime says Tiers do not certify an availability percentage.
Students will be able to: work out how much downtime a target such as 99.9% allows in a day, a month and a year. explain why each extra nine is ten times harder to promise. say what an error budget is and what a team does when it has used it up.
Lab: spread it across two
Spread it across two
A made-up shop has three layers and one building. Add spare machines and a second building to reach 99.99%, and find out why more machines in one building stop helping.
25 to 35 minutesA projector for the whole class, or one device per pair.
What this is
A made-up shop with three layers, a web layer, an application layer and a database, and one building. You add spare machines and a second building and watch the availability, the downtime and the cost. All the numbers are made up.
What to do
- Read the starting figures: one machine in each layer, all in one building. The shop works 96.9329% of the time, which is about 11 days 4 hours down in a year.
- Use Add a copy of Web servers, Add a copy of Application servers and Add a copy of Database. Find the layer that is marked as the weakest and see what a second copy does.
- Give every layer two copies, still in one building, under Where does it all sit? Read the figure: 99.87%. Then add more copies and see that it hardly moves.
- Choose A full copy in a second building. Find the cheapest set-up that reaches 99.99%.
What you should notice
More machines in one building cannot beat the building: with every layer doubled the shop reaches 99.87%, about 11 hours 23 minutes down in a year. A second building changes the answer, and the cheapest way to 99.99% in this game is two copies of every layer in both buildings, which costs 42 units and gives 99.9998%.
A made-up shop has three layers. Every machine works 99% of the time and the building works 99.9% of the time. Add spares and see what happens. The goal: reach 99.99% for as little cost as you can.
Web servers
1 copy, working 99% of the time as a group. This is the weakest layer.
1Application servers
1 copy, working 99% of the time as a group
1Database
1 copy, working 99% of the time as a group
1
Your shop works
96.9329%
of the time, so it may be down for about 11 days 4 hours a year.
Cost
8 units
Not yet at 99.99%. Try the layer that is the weakest, or ask whether the building is the limit.
How to run it, questions to ask, and what to watch for
Running it
- Start with one machine in each of the three layers in one building. Read out the availability and the downtime a year.
- Ask which single change would help most before anyone clicks. Then let pairs try. The goal is 99.99% for the least cost, and the page shows the weakest layer.
- Move to a second building. Ask why it changes the answer when more machines in the first building did not.
- Close with the note at the foot of the game: the sums assume the failures are independent. Ask the class to name a cause that two machines in one building would share.
Questions to ask
- What did the third copy of a layer buy compared with the second?
- What could take out both buildings at once?
- Is the cheapest way to 99.99% also the way you would choose for a hospital?
Watch for
- Students adding copies everywhere. The weakest layer is the one to fix first, and the page marks it.
- The numbers are made up. They are not real machines or real buildings.
- The formulas are the ones AWS prints, with its own condition: independent. The London heat story and the AWS story are both cases where the parts shared a cause.
Students will be able to: work out the availability of a chain of parts that must all work, and of copies where any one will do. say why the weakest layer is the one to fix first. explain that spare machines only help if they do not fail for the same reason.
Game: lights out
Lights out
The mains power fails in a made-up hall. A battery has minutes, a generator has to start and take over, and the fuel has to last. Choose the set-up and see whether the hall stays up.
20 to 30 minutesA projector or one device per pair. Cards and a stopwatch for the role play version.
What this is
A made-up data hall that draws 200 kW loses its mains power. A battery carries the load first, a generator has to start and take over, and its fuel has to last until the power comes back. You choose each and see whether the hall stays up.
What to do
- Leave the settings as they are and read the result for a cut of 1 hour. Under Battery size the battery lasts 4 minutes 3 seconds, the generator starts in 30 seconds and the hall stays up.
- Change The cut lasts to 3 minutes, then back to 1 hour, then to 6 hours and 24 hours. Read what ends each one.
- Change Generator takes to start to 10 minutes and read the result. Then change it back and tick The generator fails to start.
- Change Fuel in the tank lasts to 2 hours and set The cut lasts to 6 hours. Find the minute the hall goes dark.
What you should notice
A battery buys minutes and nothing more. A generator has to start before the battery is empty and must have fuel for the rest of the cut. Each one is a way for the hall to go dark.
A made-up hall draws 200 kW. The mains power fails. Set up the battery and the generator, then choose how long the cut lasts and see whether the hall stays up.
The battery lasts
4 minutes 3 seconds
15 kWh at 90% usable, over 200 kW
Result
The hall stays up
The battery held, the generator started in time and had the fuel to last. The hall stayed up.
How to run it, questions to ask, and what to watch for
Running it
- Ask students to list, in order, what keeps a data hall running when the mains power fails. Then show the page and the chain: mains, battery, generator.
- Set up the hall with a battery that lasts a few minutes and a generator that starts in seconds. Ask them to predict what happens in a 3 minute cut, a 1 hour cut and a 6 hour cut before you choose each.
- Tick the box for a generator that fails to start, and run the 1 hour cut again. Ask what the hall needed that it did not have.
- No devices? Run it as a role play with a stopwatch. One student is the battery and holds a card with the minutes it lasts, one is the generator and may only stand when its start time is up, and the rest hold cards for the servers. Say when the mains fails and see who is still standing.
Questions to ask
- Why not just a very big battery?
- What would you test, and how often, to be sure the generator will start?
- What does a hall do about the fuel for a cut of two days?
Watch for
- The battery, generator and times are made up. Vendors quote typical ranges, and the page says the vendor said it.
- Students treating a generator as magic. It has to start, take the load and have fuel. Each is a way to fail.
- Do not let students handle real batteries, generators or live electrical equipment as part of this activity.
Students will be able to: work out how long a battery lasts from its size and the load it carries. say what a battery and a generator each do, and why a hall needs both. explain that a back-up has to start in time and last long enough, and that it is only a plan if it has been tested.
Game: Black Friday
Black Friday
A made-up shop has a quiet night, an evening peak and a sudden spike. Meet it with a fixed number of servers or let the number follow demand, and compare what turns away and what it costs.
25 to 35 minutesA projector for the whole class, or one device per pair.
What this is
A made-up shop has a day of demand: a quiet night, an evening peak and, if you switch it on, a sudden spike. You choose how many servers to run, either a fixed number or a number that follows demand, and see the share of requests turned away and the bill.
What to do
- With A fixed number of servers chosen and 8 servers, read the share turned away and the bill for the day.
- Tick Add a sudden spike at 10:00. Move Servers, all day up until the share turned away is 1% or less. Read the bill.
- Choose Add and remove servers as demand changes. Set Never more than to 30 and A new server takes to No delay, and read the result.
- Change A new server takes to 30 minutes, then 1 hour. Then lower Keep servers busy and read what a lower figure buys and costs.
What you should notice
A fixed fleet of 30 servers turns nothing away on the spike day and costs £72, because most of the servers sit idle for most of it. An elastic fleet with no delay costs £19 and turns nothing away. When a new server takes 30 minutes to start, the same plan turns away about 6.8% of requests: being late costs customers.
A made-up shop has a day of demand. Each server can answer 50 requests a second. Choose how to meet the day. The goal: turn away no more than 1% of requests for as little money as you can.
Requests a second the shop receives What the servers can answer
Requests turned away
0%
Most servers at once
8
Bill for the day
£19
192 server-hours at £0.10 each (made up)
Inside the goal at £19. Is there a cheaper way?
How to run it, questions to ask, and what to watch for
Running it
- Show the day's demand and ask when the shop is busiest and when it is quiet. Ask how many servers they would buy for the whole day.
- Let pairs choose a fixed number of servers and read the share of requests turned away and the bill.
- Switch to the elastic plan. Ask what happens if a new server takes an hour to start, then add the spike.
- Close by asking what a team gives up for a lower bill, and who pays when a shop turns customers away.
Questions to ask
- When is a fixed number of servers the right choice?
- What would you want to measure before trusting an automatic plan?
- What does 'pay for what you use' mean for a business with an unpredictable day?
Watch for
- NIST says capabilities can be provisioned and released elastically 'in some cases automatically'. Do not tell students the cloud always scales itself.
- The traffic, the servers and the price are made up. A real service has other limits that this game leaves out.
- Students aiming for 0% turned away at any cost. The goal in the game is 1% or less for as little money as they can.
Students will be able to: say what elastic means, and what it costs to be slow to start. compare a fixed fleet and an elastic one on requests turned away and on the bill. explain why a fixed number of servers is either wasted for most of the day or too few at the peak.
Game: keep it cool
Keep it cool
Everything a data hall draws that does not reach the computers is a cost. Work out what a point of PUE is worth in energy, carbon and money, then find out how hot is too hot for the equipment.
25 to 35 minutesA projector or one device per pair. Cards with the hall's figures for the paper version.
What this is
Part one: how many more kilowatts a made-up data hall needs, on top of its computers, for cooling and everything else, and what that comes to in energy, carbon and money over a year. Part two: how hot is too hot for the equipment.
What to do
- Under How much do the computers in the hall draw? choose 1,000 kW. Press the button for the average in a 2025 survey of operators and read what everything else draws.
- Press the button for what the government expects of new builds. Read how the figures change, and the saving in the sentence below.
- Move the PUE slider to 1.00. Say why a real hall cannot get there.
- In part two, move Air temperature in the hall up from 22 degrees C. Find where it leaves the recommended range, and where it leaves the range every class of equipment is designed for.
What you should notice
At 1,000 kW of computing and a PUE of 1.54, everything else draws 540 kW, which is 4,730 MWh a year and about 619 tonnes of CO2e at the 2026 UK grid figure. At 1.3 the saving is 2,102 MWh a year. The hall and the price are made up, and PUE leaves out things such as the water a site uses.
Part one: a made-up hall full of computers. Every kilowatt that reaches the computers needs more power for cooling and everything else, and the ratio has a name: power usage effectiveness, or PUE. Part two: how hot is too hot for the equipment?
Part one: what everything else costs
For a hall whose computers draw 1,000 kW, moving from the survey average (1.54) to the new-build expectation (1.3) saves 2,102,400 kWh a year, which is 2,102 MWh.
Whole site draws
1,540 kW
1,000 kW of it reaches the computers
Everything else draws
540 kW
Over a year
4,730 MWh, about 619 tonnes of CO2e, and £1,182,600 at a made-up 25p a kWh.
Part two: how hot is too hot?
- Recommended: 18 to 27 degrees C
- Class A1 equipment is designed to work between 15 and 32 degrees C
- Class A2 equipment is designed to work between 10 and 35 degrees C
- Class A3 equipment is designed to work between 5 and 40 degrees C
- Class A4 equipment is designed to work between 5 and 45 degrees C
At 22 degrees C
Inside the recommended range
The equipment is in the range recommended for reliability and efficiency.
How to run it, questions to ask, and what to watch for
Running it
- Ask what a building full of computers needs besides power. Take a few answers, then read out what the draft National Policy Statement says about the cooling systems that keep data centres from overheating.
- Show part one. Start at the average PUE from the survey. Ask what a kilowatt of computing needs besides, then move the slider to the new-build expectation and read out the saving.
- Move to part two and slide the temperature up through the recommended range and past it. Ask what the London heat story tells us about a hall built for cooler summers.
- No devices? Give pairs a card: 400 kW of computing and a whole site that draws 600 kW. They work out the PUE, then what the site would draw at 1.3, and the difference in a year.
Questions to ask
- Why is a PUE of 1.0 impossible?
- Who gains when a hall lowers its PUE, the operator, the customer or the grid?
- Why does the page say a 2026 grid figure is not a live reading?
Watch for
- The hall and the price are made up. The PUE figures, the grid figure and the temperature ranges are from sources, and each is listed under the game.
- PUE leaves out things, such as the water a site uses, so a low PUE is not the same as a green site. The page says so and so should you.
- The government's 1.3 is a draft expectation for new builds in England, not a rule for existing halls.
Students will be able to: work out PUE from the power the whole site draws and the power the computers draw. turn a power in kW into energy over a year in kWh, and into carbon using a stated grid figure. say what the recommended temperature range for equipment is, and what happens beyond it.
Lab: by hand or by pipeline
By hand or by pipeline
Twenty steps, each with a one in a hundred chance of a slip, is not a one in a hundred chance. See how small chances add up, and what a script changes and what it does not.
20 to 25 minutesA projector or one device per pair.
What this is
A made-up team releases its software over and over. By hand, a person follows a list of steps and any one of them can slip. A script does the same steps the same way, but a mistake in the script is in every release. You compare the two.
What to do
- Leave Steps in a release at 20 and Chance of slipping on one step, by hand at 1%. Read how often a release goes right by hand and by script.
- Move Steps in a release up to 40 and the chance of slipping to 2%. Read what happened to the figure by hand.
- Move Steps in a release down to 5 and the chance of slipping down to 0.5%. Then move Chance the script has a mistake that spoils a release up to 5%, and read what the page says about the two.
- Change Releases a week and read how many releases a year go wrong by hand and by script.
What you should notice
A 1% chance of a slip on each of 20 steps leaves only 81.8% of releases going right first time. At 40 steps and 2% it falls to 44.6%. A script is not automatically better: at 5 steps and 0.5%, doing it by hand goes right 97.5% of the time, and a script with a 5% chance of a mistake loses. At the starting figures a script with a 2% chance of a mistake means about 5 of 260 releases a year go wrong, against about 47 by hand.
A made-up team releases its software over and over. By hand, every step is a chance to slip. With a script, the steps are the same every time, but a mistake in the script is in every release. Move the sliders and compare.
By hand
81.8%
of releases go right first time
About 47 of 260 releases a year go wrong.
A person spends about 130 hours a year on the steps.
By script
98.0%
of releases go right first time
About 5 of 260 releases a year go wrong.
A person writes and checks the script once, then it runs by itself.
With these numbers, the script goes wrong less often. A small chance of a slip on each step adds up fast over many steps.
How to run it, questions to ask, and what to watch for
Running it
- Ask students for a task with a list of steps, such as making a sandwich or logging in to a game on a new device. How often do they miss a step?
- Show the page. Set 20 steps and a 1% chance of a slip, and ask what chance there is that all 20 go right. Take guesses, then read the answer.
- Raise the steps to 40 and the chance of a slip to 2%. Ask what changed.
- Now compare the script. Ask when a script is a better bet than a person, and when it is not, using the sliders.
Questions to ask
- Why is a list of 20 careful steps still likely to go wrong somewhere?
- If a script has a mistake in it, who finds it, and how quickly?
- What would you check before you trusted a script with the whole shop?
Watch for
- Students who say automation is always safer. The game lets the script lose, on purpose.
- The numbers are made up, and the sums assume each step goes wrong independently of the others.
- A student who has not met compound probability. Ask what the chance is of two heads in a row and build up from there.
Students will be able to: work out the chance that every step of a list goes right when each has a small chance of going wrong. say what automation changes, and why a script can fail every release at once. explain why a team tests its automation before it trusts it.
Game: rack and stack
Rack and stack
A rack has 42 units of space and 8 kW of power it may draw. Fill it from a made-up shelf to get as much done as you can, then see the best possible answer.
25 to 35 minutesA projector or one device per pair. Squared paper and scissors for the paper version.
What this is
A made-up rack has 42 units of space and may draw 8 kW. You fill it from a made-up shelf to get as much work done as you can without going over either limit, then you see the best possible answer.
What to do
- Use Put a Large server in to add large servers. Watch the Space and Power and heat bars as you go.
- Keep going until you hit a limit. Say whether it was space or power, and what is left over.
- Take things out and try another mix. Use Put a GPU server in and Put a Storage shelf in.
- Press Show the best possible answer and compare it with yours.
What you should notice
In this game the best rack does 84 units of work a day using 2 GPU servers and 8 storage shelves. It uses 24 of the 42 units of space and all 8 kW of power. A rack can be full by power with space to spare, and a rack that is full of space may have power left over.
A made-up rack has 42 rack units of space (1866.9 mm of height) and may draw 8 kW. In this game, the power a rack draws is also the heat the room has to take away. Fill it to get as much work done as you can without going over either limit.
Network switch
1 unit high, draws 0.15 kW, does 2 units of work a day. 2 on the shelf.
0Small server
1 unit high, draws 0.35 kW, does 2 units of work a day. 14 on the shelf.
0Large server
2 units high, draws 0.8 kW, does 7 units of work a day. 8 on the shelf.
0Storage shelf
2 units high, draws 0.25 kW, does 3 units of work a day. 8 on the shelf.
0GPU server
4 units high, draws 3 kW, does 30 units of work a day. 3 on the shelf.
0
Work done a day
0
Inside both limits.
In the rack: nothing yet.
How to run it, questions to ask, and what to watch for
Running it
- Show the made-up shelf and the two limits, space and power. Ask which they think will run out first.
- Let pairs fill the rack. They can see the space and the power gauge as they go. The goal is the most work done without going over either limit.
- Ask a pair that fills the space and has power left to say what they could do about it, and a pair that runs out of power with space left.
- Show the best possible answer and ask what the best rack does that most of the class did not.
- No devices? Give each pair a strip of squared paper 42 squares tall and cut-out blocks for each piece of kit, with its power written on it. They keep a running total of power on paper.
Questions to ask
- Why do real halls care about power per rack as much as space?
- What would you buy to use the space that is left when the power runs out?
- In the game, the power a rack draws counts as the heat the room must take away. Why would a hall care about the heat in each rack?
Watch for
- The rack limits, the kit and the work done are made up. A rack unit is a fixed height, but real racks and equipment vary.
- Students who treat the best possible answer as the only right answer. Several racks may tie, and the page says it keeps the first it finds.
- Heavy equipment. Weight is a real limit too, and the game leaves it out.
Students will be able to: say that a rack is limited by its space and by its power at the same time. add up the space and power of a set of equipment and say whether it fits. compare their own answer with the best possible and say what made the difference.
Lab: rent by the hour or by the year
Rent by the hour or by the year
A made-up machine costs 10p an hour or a lower price for promising it all month. Which is cheaper depends on how much you use it, and the crossing point is a share of the month.
20 to 25 minutesA projector or one device per pair.
What this is
One made-up machine costs 10p an hour if you rent it by the hour, or a lower rate if you promise to keep it for the whole month. You pick what the machine is for and see both bills and the number of hours at which they are the same.
What to do
- Under What is the machine for? choose A test machine for developers. Read both bills.
- Choose An online shop that never closes. Read both bills again.
- Move Discount for promising to keep it all month between 10% and 60%, and watch the crossing point move.
- Find the workload where the answer changes as you move the discount.
What you should notice
At 40% off, promising the whole month costs £44, the same as renting for 438 hours, which is 60% of the month. The test machine used for 176 hours costs £18 by the hour, so renting wins. The shop that never closes costs £73 by the hour, so the promise wins. Which is cheaper depends on how much you use the machine.
A made-up machine costs £0.10 an hour to rent by the hour. Or you can promise to keep it for the whole month and pay less per hour. Which is cheaper depends on how much you use it.
A month is about 730 hours. With 40% off, promising the whole month costs the same as renting by the hour for 438 hours, which is 60% of the month. Use it for more hours than that and the promise is cheaper. Use it for fewer and renting by the hour is.
Rent by the hour
£18
176 hours at £0.10
Promise the whole month
£44
730 hours at 40% off
Renting by the hour is cheaper by £26.
How to run it, questions to ask, and what to watch for
Running it
- Show the four made-up workloads and ask students to guess which is cheaper to rent by the hour and which to commit to.
- Pick the test machine for developers and read both bills. Then pick the shop that never closes and read them again.
- Move the discount and watch the crossing point move. Ask what share of the month the machine would have to be busy for a 40% discount to pay off.
- Ask who in a company would decide this and what they would need to know before they did.
Questions to ask
- Why do providers offer a lower price for a promise?
- What is the risk of promising a month and then not needing the machine?
- Why might a company have someone whose job is to watch the cloud bill?
Watch for
- The price, the discount and the workloads are made up, and no real provider's prices are used. Real price lists change, so send students to the provider's own page for the day.
- A month is taken as 730 hours, which is 8,760 hours in a year divided by 12.
- The FinOps Foundation says FinOps is not just about saving money. Do not let the lesson become a race to the cheapest bill.
Students will be able to: work out a bill for renting by the hour and for committing for a month. find the number of hours at which the two cost the same. say who in a company decides between the two, and what they need to know first.
True stories
Three true stories about infrastructure
Each story below comes from a document you can open, written by the company or the investigators, and each says only what that document says. Where the documents disagree, we show both. Read them first and decide how to handle them: these are real events that cost real people real work.
A fire in France. In the night of 9 to 10 March 2021 a fire started in the rooms that housed the batteries and the uninterruptible power supply at OVH's data centre in Strasbourg, says the French government's industrial accident investigation bureau, the BEA-RI. The building, called SBG2, was destroyed and nobody was hurt. The bureau says its detection systems worked perfectly, and that what went wrong came after: no automatic extinguishing system, one weak water supply, and generators that started automatically even though the building was on fire, so the site was only made electrically safe around 02h30. It says the cause of the failure in the power equipment was not established when it reported. OVH says in its reply that it is rolling out automatic extinguishing and putting battery rooms apart from the servers. Question for the class: which of the safety systems in this story was designed to help, and which got in the way?
A heatwave in London. On 19 July 2022, Google's own incident report says, a simultaneous failure of multiple, redundant cooling systems in one of the data centres hosting its London zone, combined with extraordinarily high outside temperatures, meant the site could not keep a safe operating temperature. Google powered down part of it on purpose to prevent an even longer outage. It counts 4 hours 8 minutes of cooling failure, 18 hours 23 minutes of impact on cloud services and 35 hours 15 minutes for the slowest recoveries, and says some regional services were hit because it inadvertently changed traffic routing. The Uptime Institute wrote that the heatwave approached or passed the design limits for data centres built before 2019, and that most data centres probably got through it without a major incident. Question for the class: the sums in the spare machines game assume failures are independent. Was this independent?
A bug in an automatic system. On 19 and 20 October 2025 Amazon Web Services' own account says a latent race condition in the automation that manages the DNS records for its DynamoDB database service in its North Virginia region left one record empty, and the automation could not repair it. Anything that depended on that database had trouble, and AWS says its engineers restored the records by 2:25 AM and that clearing the knock-on effects took until 2:20 PM that afternoon. It says servers that were already running stayed healthy, and what failed was starting new ones and reaching the database. It does not say who or what to blame, and it does not mention an attack. Question for the class: what depended on one empty record, and what would you have done to make it depend on less?
Three true stories
A fire in a French data centre, a cooling failure in a London heatwave and a bug that emptied one record at the heart of a huge cloud. Read what each company and each investigator says, and only that.
30 to 40 minutesPaper. A printout of the page, or a projector. No devices needed.
What this is
Three real events, each read from the document written by the company or the investigators, with a timeline and a plain list of what is safe to say and what is not. Nothing here has been added to what the documents say.
What to do
- Split into three groups, one for each story. Read the timeline and the lines on what is safe and not safe to say.
- In your group write two sentences on what happened, one on what changed afterwards, and one on something you would not say and why.
- Tell your story to the other groups. Listen for the parts that were physical and the parts that were software.
- As a class, decide which story shows spare parts that failed for a shared reason.
What you should notice
In France, the investigators say a fire started in the rooms with the batteries and the UPS, and that the cause was not established. In London, Google says several redundant cooling systems failed at the same time in a heatwave. At AWS, the company says an empty record in an automatic system caused the trouble. In the second and third, the spare parts failed for a shared reason.
How to run it, questions to ask, and what to watch for
Running it
- Split the class into three groups, one for each story. Give each group the story's timeline and the lines marked safe to say and not safe to say.
- Each group prepares two sentences on what happened, one on what changed afterwards and one on what they would not say, and why.
- Each group tells its story to the others. Ask the listeners to say which parts were physical and which were software.
- Finish by asking which story shows spare parts that failed for a shared reason.
Questions to ask
- What did each company or investigator say, and what did they not say?
- Which of these could happen in a school or college?
- What would you want to be true of a company's report on its own failure?
Watch for
- Blame. The documents say what they say, and the page attributes each statement. Do not add a cause the document does not give.
- The OVH fire has no official cause in the report, and its times differ between OVH and the investigators. Say both.
- The London cooling story names Google only because Google wrote the report. No other company is named.
- Treat these as real events that frightened real people and cost them real work.
Students will be able to: say in two sentences what happened in each of the three stories, using only what the documents say. name one thing in each story that was a physical failure and one that was a software failure. say which of the three shows spare parts that failed for a shared reason.
Careers
Eight kinds of cloud and infrastructure job
Eight kinds of job
Cloud and infrastructure is eight areas of work, from designing systems to keeping the cables and cooling running. See the jobs in each, the pay with its source and the routes in.
20 to 25 minutesA projector or one device per pair.
What this is
Eight areas of cloud and infrastructure work from the TechSkills Career Matrix. For each one you can see the jobs, the pay with its own source, who it tends to suit and the routes in.
What to do
- Pick an area you know nothing about, such as Solutions Architecture. Read Jobs in this area and Best suited to.
- Open Pay, with its source for two areas, one that is done at a desk and one that is done on site, and compare.
- Find a route into an area that is not a degree, and note its level and length.
- Find one area that has no apprenticeship route of its own, and read what the page says about the job title cloud engineer.
What you should notice
Cloud and infrastructure is eight areas, some done at a desk and some on site. Pay ranges are a guide and come from the source shown. No apprenticeship standard is called cloud engineer, so the page names the standards where those job titles appear.
8 areas of work in cloud and infrastructure, from the TechSkills Career Matrix. Pick one to see what the work is, the jobs in it, what it suits and what it pays, with the source for the pay. A hard hat mark means there is a hands-on route into that area as well as a desk-based one.
Area 01
Cloud Engineering
Designing and managing virtual environments on AWS, Azure, or GCP.
Designing and managing virtual environments on AWS, Azure, or GCP.
The desk or degree route
- Jobs in this area
- Cloud Engineer, Cloud Architect, Platform Engineer, FinOps Analyst
- Best suited to
- Analytical minds who enjoy building complex, logical systems in the cloud.
- Pay, with its source
- £35,000 to £117,500+ (median junior Cloud Engineer to 90th-percentile Cloud Architect, IT Jobs Watch)
- Industries that hire for it
- Financial Services, Retail & E-commerce, Technology & Software, Media & Entertainment, Government & Public Sector
Pay is quoted as the Career Matrix gives it, with its own source and date: use it with that, and not as a promise. See the full Cloud and Infrastructure careers page.
How to run it, questions to ask, and what to watch for
Running it
- Show the eight areas and ask students to pick one they know nothing about and one they might try.
- Open the pay panel for two areas that differ in how they are done, one at a desk and one on site. Read out the pay and its source.
- Ask students to find a job in an area that does not sound like a computer job.
Questions to ask
- Which of these jobs would you not have known existed?
- Which needs the most hands-on work, and which the most analysis?
- Which would you try first, and what would you need to find out?
Watch for
- Students assuming cloud means only desk work. Data centre operations is on site and hands-on.
- Pay ranges are a guide from the sources shown, and they vary by employer and place.
- Any area can be reached by more than one route. Nothing here is a ceiling.
Students will be able to: name at least three areas of cloud and infrastructure work and a job in each. say which areas are done at a desk, which on site, and which both. say what the sources say about pay, and what they do not.
A student who searches for a cloud engineer apprenticeship will not find one. Skills England has no current standard with cloud engineer, data centre technician or site reliability engineer in its title. Those titles appear inside other standards instead: the Information communications technician standard lists data centre support technician and cloud technician as typical job titles, and the DevOps engineer standard lists site reliability engineer and platform engineer. The Career Matrix on this page shows each of them under the area of work it fits.
The government's own role framework is the same. The Digital and Data Profession Capability Framework has infrastructure engineer, infrastructure operations engineer, DevOps engineer and architect roles, and no site reliability engineer, cloud engineer or platform engineer role.
Routes in
Three ways in, side by side
Three kinds of route into cloud and infrastructure are worth putting next to each other, and none comes first on this page. One starts with study. Two are apprenticeships, which are jobs with training in them: one mostly at a desk and one mostly on site. Cloud and infrastructure is one of the families with a hard hat route in the Career Matrix, Data Centre Operations, so the columns are balanced between the grad cap and the hard hat.
A plain university degree route is missing from the first column on purpose: we did not find an official source that describes one for these jobs, and we would rather say so than guess. What we did find is sourced, and the first column shows it: a T Level, and a Level 7 apprenticeship that Skills England says is likely to need a degree. The T Level that fits best is Digital Support and Security, whose Digital infrastructure specialism is aligned to the Information communications technician standards.
The desk column is DevOps engineer at Level 4, IT solutions technician at Level 3, which started a new version on 2 November 2026, and the Level 6 integrated degree apprenticeship in digital and technology solutions, which is not specific to cloud work. The site column is the Level 2 foundation apprenticeship, the Information communications technician standard and the cooling standard, and it says plainly what is missing: no standard covers the back-up power a data centre depends on, and the cooling standard names data centres once, in a list of places the work is done.
Several of these are being replaced for new starts: DevOps engineer on 18 December 2026 and Information communications technician on 28 January 2027. Check the live Skills England page before giving a student a version. Pay for each area of work, with its source, is in the Career Matrix above.
Three ways in
A university degree, an apprenticeship at a desk and an apprenticeship on site, side by side, with the same information for each from the people who set the standards.
20 to 25 minutesA projector or one device per pair.
What this is
Three ways in side by side: a study route, an apprenticeship at a desk and an apprenticeship on site, with the same information for each, taken from Skills England and the government's T Level pages.
What to do
- Read the three headings and say what the columns have in common.
- Take one column each in a group. List three questions a student would ask, then find the answers on the page.
- Open a standard from your column. Read its level and length, then the line that says what the length leaves out.
- Find the sentence that says what none of the standards covers.
What you should notice
The three columns have equal weight and a student can move between them. No standard is called cloud engineer or data centre technician, and none covers a data centre's back-up power. Several standards are being replaced for new starts on dates the page gives.
Study first
Grad cap: the study route
Often suits someone who wants to study first and train after, or who already holds a degree and wants to move towards design and architecture.
T Level in Digital Support and Security
Level 3, 2 years
A course taken after GCSEs with an industry placement of at least 315 hours. Its Digital infrastructure specialism is aligned to the Information communications technician standards.
Digital and technology solutions specialist (integrated degree)
ST0482, Level 7, typically 18 months
A specialist route with an enterprise systems architecture option. Skills England says entry is likely to include a degree at 2.1 or higher in a relevant subject, though some employers will accept other qualifications or experience.
Where it can lead: The T Level's Digital infrastructure specialism is aligned to the Information communications technician occupations, which is the Level 3 standard in the site column. The Level 7 standard is for people who already hold a degree, and government funding for it has been restricted since 1 January 2026.
Earn and learn at a desk
The desk-based apprenticeship route
Often suits someone who would rather learn the work inside an employer from the start, mostly at a screen, and who is comfortable with a longer route in exchange for being employed while they train.
ST0505, Level 3, typically 18 months
Developing, implementing and maintaining complete IT solutions such as networks, operating systems and applications.
ST0825, Level 4, typically 24 months
Enabling organisations to get valuable working software out in front of active users and improving the quality of digital services.
Digital and technology solutions professional (integrated degree)
ST0119, Level 6, typically 48 months
A degree built into the job. It is not specific to cloud work: its options include network engineering professional, IT consultant professional and cyber security professional.
Where it can lead: The IT solutions technician page lists example progression routes including network engineer and DevOps engineer. The DevOps engineer standard lists infrastructure engineer, platform engineer and site reliability engineer among its typical job titles.
Earn and learn on site
Hard hat: the on-site apprenticeship route
Often suits someone who wants to keep real hardware running, who is comfortable with shift-based, site-based work and who would rather learn it hands-on from the start.
Hardware, network and infrastructure foundation apprenticeship
FA0004, Level 2, typically 8 months to the gateway
A Level 2 start, normally for people aged 16 to 21 at the start, whose job titles include cloud technician and data centre support technician.
Information communications technician
ST0973, Level 3, typically 18 months
Delivering efficient operation and control of IT and telecommunications infrastructure, on premises or provisioned as cloud services.
Refrigeration, air conditioning and heat pump engineering technician
ST0322, Level 3, typically 36 months
Installing, maintaining, servicing, commissioning and decommissioning the cooling systems found in refrigeration, air conditioning and heat pumps. Data centres are one of many places the work is done.
Where it can lead: The Information communications technician standard lists data centre support technician and cloud technician among its job titles. No standard we found covers a data centre's backup power, such as batteries and generators, and the electrician and building services standards we read do not mention data centres.
How to run it, questions to ask, and what to watch for
Running it
- Show the three columns together and read the headings. Ask students to say what the columns have in common before you compare them.
- Ask groups to take one column each and list three questions someone choosing it would want answered. Then check the page for the answers.
- Open a standard on the page and read the level and length together. Ask what the number of months leaves out.
- Close by asking what a student who picked one route and wanted to change would do.
Questions to ask
- What does each route cost a student, and what does it pay them?
- Who is each route likely to suit, and who might it leave out?
- Why might a job called cloud technician not have an apprenticeship with that name?
Watch for
- Treating any column as the better one. They have equal weight, and a student who picks one is not locked in.
- Entry requirements change each year and by employer. Send students to the standard, the employer and the course for the current ones.
- Several standards on the page are being replaced for new starts on dates the page gives. Check the live page before giving a student a version.
- No standard covers a data centre's backup power. The hands-on route here covers IT and communications field work, cooling and a foundation route, and the page says so.
Students will be able to: compare three routes on level, length and who pays. say what each route tends to suit, as a tendency and not a rule. explain that no apprenticeship standard is called cloud engineer or data centre technician, and say where those jobs appear instead.
Students will hear about vendor certificates. The two beginner ones the Matrix names are AWS Certified Cloud Practitioner and Microsoft Azure Fundamentals. AWS says its exam is designed for people new to cloud who may not have an IT background. Both are issued by the companies that run the clouds, and the fees and exam details can change, so send students to the issuer's page.
For teachers
The Translator: jargon in plain English
Cloud and infrastructure has a lot of acronyms and tool names. Type a word to see what it means, a comparison from school life you can use at once, and where the comparison stops working.
26 terms
Availability
Also called: uptime
- In plain English
- The share of the time a service is working when people try to use it, usually written as a percentage such as 99.9%.
- Think of it like this
- The share of lessons in a year in which the room was open and the projector worked.Where that stops working: A percentage hides how the time is spread. Ten minutes down every day feels very different from one long outage.
- At work
- Site Reliability Engineers and Observability Engineers measure it and work to keep it high.
Availability zone
Also called: region, zone
- In plain English
- A separate location inside a cloud provider's region, designed so that the failure of one does not take down the others.
- Think of it like this
- Two classrooms in different corridors, so that a problem in one does not close the other.Where that stops working: Two classrooms still share the same building and the same main supply, and zones can share more than they seem to.
- At work
- Cloud Architects and Solutions Architects choose how many to use.
Cloud
Also called: cloud computing
- In plain English
- Computing services, such as storage, software and processing, that you use over the internet from someone else's computers and pay for by how much you use.
- Think of it like this
- Borrowing a minibus from the council instead of buying one: you ask when you need it, use it and hand it back, and someone else keeps it running.Where that stops working: A minibus is one vehicle you can see. The cloud is a very large number of computers shared between many customers, and you never see which ones are yours.
- At work
- Cloud Engineers and Platform Engineers design and run what a company builds on it.
Colocation
Also called: colo
- In plain English
- A data centre where a company rents space, power and cooling for its own computers, rather than building its own or renting computers from a cloud provider.
- Think of it like this
- Renting a stall in a market hall: your goods, their building.Where that stops working: A stall holder can walk away with the stall. Moving a rack of servers between buildings is a project.
- At work
- DC Ops Managers run the sites that offer it.
Data centre
Also called: data hall, server room
- In plain English
- A building, or part of one, full of computers, with the power, cooling and network connections they need. Sources count them differently, because there is no universally agreed definition.
- Think of it like this
- The school kitchen. It is a room built round the things that make the meal happen: the cooker, the extraction, the water and the delivery door.Where that stops working: A kitchen serves one school. A data centre may hold computers for thousands of customers who never meet.
- At work
- Data Centre Techs and DC Ops Managers keep it running, and Hardware Engineers install the equipment.
DevOps
- In plain English
- A way of working in which the people who build software and the people who run it work as one team, using tools to release changes often and safely.
- Think of it like this
- The same people who write the school play also set up the stage and run the lights on the night, instead of passing it over a wall.Where that stops working: A play is performed once. Software is changed and released again and again, so the tools matter more.
- At work
- DevOps Engineers and Build Engineers do it.
Elasticity
Also called: autoscaling, scaling
- In plain English
- The ability to add computing capacity when demand rises and take it away when it falls, so you pay for what you need at the time.
- Think of it like this
- Opening extra tills in the canteen at lunchtime and closing them again at two.Where that stops working: Opening a till takes a moment. In the cloud a new server also takes time to start, and a sudden rush can arrive before it is ready.
- At work
- Cloud Engineers and DevOps Engineers set it up.
Error budget
- In plain English
- The amount of failure a service is allowed under its target. If the target is 99.9% over a month, the budget is the 0.1% of that month that may go wrong.
- Think of it like this
- A set number of late marks a student may have before a letter goes home.Where that stops working: A late mark is the student's fault. An error budget is spent by anything that breaks, including the team's own changes.
- At work
- Site Reliability Engineers and Incident Managers use it to decide when to slow down releases.
FinOps
- In plain English
- The practice of managing what the cloud costs, bringing engineering, finance and the business together to decide where spending is worth it.
- Think of it like this
- A department deciding together how to spend a budget so that the money buys the best result and not just the cheapest.Where that stops working: A department budget is fixed once a year. Cloud spending changes by the hour, so the watching never stops.
- At work
- FinOps Analysts do it.
Generator
Also called: standby generator, engine generator
- In plain English
- An engine that makes electricity from fuel, started when the mains fails so the equipment can keep running for much longer than a battery can.
- Think of it like this
- A caravan's portable power unit, used in a field when there is no plug.Where that stops working: A portable unit is switched on by a person. A data centre's generator has to start and take the load by itself, in seconds, and has to be tested to be trusted.
- At work
- Data Centre Techs and DC Ops Managers keep them ready.
Infrastructure as code
- In plain English
- Describing the computers, networks and settings a system needs in files that a tool can read, so the same set-up can be built again exactly.
- Think of it like this
- A recipe, written down so that anyone can make the same dish, rather than one cook doing it from memory.Where that stops working: A recipe still needs a cook to check it. A file with a mistake builds the mistake exactly, every time.
- At work
- Cloud Engineers and DevOps Engineers write it.
kW and kWh
Also called: kilowatt, kilowatt-hour
- In plain English
- A kilowatt is a rate of using energy. A kilowatt-hour is the energy used by 1 kW running for one hour, which is what a bill counts.
- Think of it like this
- Speed and distance. 60 miles an hour is a rate, and driving at it for an hour gives 60 miles.Where that stops working: A car's speed changes. A data hall's load is closer to steady, which is why the maths is simple.
- At work
- DC Ops Managers and FinOps Analysts work in both.
Nines
Also called: three nines, five nines
- In plain English
- A way of saying availability by counting the nines: 99.9% is three nines. Each extra nine cuts the time a service may be down to a tenth.
- Think of it like this
- Attendance. Going from 90% to 99% attendance sounds small, and over a year it is the difference between missing weeks and missing days.Where that stops working: Attendance is counted in days. Availability is counted in minutes, and the sums get small very fast.
- At work
- Site Reliability Engineers set targets in nines and report against them.
Pipeline
Also called: CI/CD, continuous integration and delivery
- In plain English
- A series of automatic steps that takes a change to software, tests it, and, if it passes, releases it.
- Think of it like this
- A homework routine that always runs in the same order: write it, check it, hand it in.Where that stops working: A routine you follow yourself can be skipped. A pipeline cannot, but a mistake in it is made every time.
- At work
- DevOps Engineers and Automation Leads build and look after them.
PUE
Also called: power usage effectiveness
- In plain English
- The total power a data centre draws divided by the power its computers draw. A PUE of 1.5 means that for every 1 kW reaching the computers, 0.5 kW more goes on cooling and everything else.
- Think of it like this
- The total of the school's electricity bill compared with what the classrooms' computers used. The rest goes on lights, heating and the kitchen.Where that stops working: PUE leaves things out, such as the water a site uses and how efficient the computers themselves are, so a low PUE is not the same as a green site.
- At work
- DC Ops Managers and Sustainable Tech Leads report and improve it.
Rack
Also called: server rack
- In plain English
- A metal frame, a standard width, that holds computers and network equipment one above another, with power and cables at the back.
- Think of it like this
- A bookcase for computers, where every shelf is the same width and each item takes a whole number of shelves.Where that stops working: A bookcase has no electricity to supply and no heat to take away. A rack has both, and they are often the real limit.
- At work
- Data Centre Techs install and cable them.
Rack unit
Also called: U, 1U
- In plain English
- The standard measure of height in a rack. One unit is 44.45 mm, and equipment is made in whole numbers of units.
- Think of it like this
- A line on lined paper: everything has to take up a whole number of lines.Where that stops working: Lined paper has no weight limit. A rack has a limit on weight as well as space and power.
- At work
- Hardware Engineers and Data Centre Techs plan what goes in which unit.
Redundancy
Also called: spares, failover
- In plain English
- Having more than one of something that matters, so that when one fails another takes over.
- Think of it like this
- A spare projector bulb, a second teacher who knows the register system and a back-up key to the music room.Where that stops working: A spare only helps if it does not break for the same reason. Two bulbs on one blown fuse are one fault.
- At work
- Cloud Architects and Solutions Architects design it in.
Server
- In plain English
- A computer whose job is to provide something to other computers, such as web pages, files or a database, when they ask.
- Think of it like this
- The person in the school office who answers every phone call and hands out whatever was asked for: the timetable, the form, the key.Where that stops working: A real server answers many requests at the same moment and never needs a break, which the office cannot do.
- At work
- Systems Administrators and Storage Engineers look after them.
Shared responsibility
Also called: shared responsibility model
- In plain English
- The idea that a cloud provider looks after some things and the customer looks after others, and that the split depends on the kind of service.
- Think of it like this
- A school trip: the coach company keeps the coach safe, the school decides who goes and the teacher counts heads.Where that stops working: Nobody on a trip can swap the jobs. In the cloud the customer can choose how much to hand over, but never everything.
- At work
- Cloud Security Engineers and Cloud Governance Managers keep the split clear.
Single point of failure
- In plain English
- One part that, if it fails, stops the whole thing, because nothing else can do its job.
- Think of it like this
- The one member of staff who knows the password to the timetable.Where that stops working: It is easy to spot in a drawing and hard to spot in real life, because many things share a hidden dependency such as one power feed or one DNS record.
- At work
- Solutions Architects and Site Reliability Engineers look for them.
SLA
Also called: service level agreement
- In plain English
- A written promise from a provider to a customer about how well a service will work, and what the customer gets if it does not, which is usually a credit on a future bill.
- Think of it like this
- The school's promise to parents that the office will answer within a day, and what it will do if it does not.Where that stops working: Real agreements have conditions. A headline figure applies only if you set things up the way the agreement says.
- At work
- Compliance Analysts and Cloud Governance Managers read and check them.
SLO
Also called: service level objective
- In plain English
- A target a team sets for itself for how well its service should work, such as 99.9% of requests answered. It is usually a little stricter than the promise it keeps.
- Think of it like this
- A team's own target for the term, set a bit above what the teacher has asked for.Where that stops working: A target has no penalty attached. Missing it is a signal to the team, not a payment to a customer.
- At work
- Site Reliability Engineers set and watch them.
Tier
Also called: Uptime Institute Tier
- In plain English
- A grade from I to IV that the Uptime Institute gives to a data centre's design, based on how much redundancy it has and whether parts can be worked on without shutting it down.
- Think of it like this
- A star rating for a restaurant's kitchen, covering the equipment and how it is run.Where that stops working: A star rating is a promise of quality. A Tier is not a promise of availability: Uptime says its Tier awards do not certify any availability percentage.
- At work
- DC Ops Managers and Solutions Architects use them to choose and describe sites.
UPS
Also called: uninterruptible power supply
- In plain English
- A battery-backed power supply that keeps equipment running for a short time when the mains fails, and cleans up dips and spikes.
- Think of it like this
- The torch in the staff room that lasts long enough to get everybody to the exit.Where that stops working: A torch is for walking out. A UPS is meant to bridge the gap until the generator takes over, not to run the building.
- At work
- Data Centre Techs service them.
Virtual machine
Also called: VM, virtualisation
- In plain English
- A computer that exists as software, running on a real computer that can host several of them, each acting as if it were separate.
- Think of it like this
- A hall that is used for assembly, an exam and a concert on different days: one building, several separate uses.Where that stops working: The three uses in the hall happen in turn. Virtual machines run at the same time and share the real computer's power between them.
- At work
- Virtualisation Leads and Systems Administrators manage them.
For teachers
The Curriculum Translator
Which parts of the curriculum in England each activity supports, using only the Department for Education's own words, the exam boards' specifications and the plans the government has published. We say plainly where the curriculum is thin on this subject, and what the government has said it will change. The words data centre appear in none of them.
Show the curriculum links for
| Where it comes from | What the curriculum says | Activities that support it |
|---|---|---|
| Computing, national curriculum, secondary | “understand the hardware and software components that make up computer systems, and how they communicate with one another and with other systems”National curriculum in England: computing programmes of study | Where is the cloud?, Who looks after what? |
| Computing, national curriculum, primary | “understand computer networks, including the internet”National curriculum in England: computing programmes of study | Where is the cloud? |
| GCSE computer science, Department for Education subject content | “main and contemporary secondary storage and ways of storing data on devices”GCSE subject content for computer science | Where is the cloud?, Who looks after what? |
| GCSE computer science, Department for Education subject content | “the ethical, legal and environmental impacts of digital technology on wider society”GCSE subject content for computer science | Keep it cool, Three true stories |
| Department for Education response on the GCSE content, March 2025 | “which may inhibit coverage of cloud computing within awarding organisations' specifications”GCSE Computer Science subject content: government consultation response (PDF) | Where is the cloud?, Black Friday, Rent by the hour or by the year |
| AQA GCSE Computer Science 8525, section 3.4.5 | “Explain the term cloud storage.”AQA GCSE Computer Science 8525 specification, version 1.3 | Where is the cloud?, Who looks after what? |
| AQA GCSE Computer Science 8525, section 3.8 | “AQA lists cloud storage among the areas that its exam questions on the ethical, legal and environmental impacts of digital technology will be taken from.”AQA GCSE Computer Science 8525 specification, version 1.3 | Keep it cool, Three true stories |
| OCR GCSE Computer Science J277, section 1.3.1 | “The Cloud: remote service provision (e.g. storage, software, processing)”OCR GCSE (9-1) Computer Science J277 specification, version 3.1 | Where is the cloud?, Black Friday, Rent by the hour or by the year |
| OCR GCSE Computer Science J277, section 1.3.1 | “Concept of servers providing services”OCR GCSE (9-1) Computer Science J277 specification, version 3.1 | Spread it across two, Who looks after what? |
| Mathematics, national curriculum, secondary | “calculate the probability of independent and dependent combined events, including using tree diagrams and other representations, and know the underlying assumptions”National curriculum in England: mathematics programmes of study | How many nines?, Spread it across two, By hand or by pipeline |
| Science, national curriculum, secondary physics | “power as the rate of transfer of energy”National curriculum in England: science programmes of study | Lights out, Keep it cool, Rack and stack |
| Science, national curriculum, secondary physics | “comparing power ratings of appliances in watts (W, kW)”National curriculum in England: science programmes of study | Keep it cool, Rack and stack |
| Science, national curriculum, secondary physics | “comparing amounts of energy transferred (J, kJ, kW hour)”National curriculum in England: science programmes of study | Keep it cool, Lights out |
| Science, national curriculum, secondary physics | “domestic fuel bills, fuel use and costs”National curriculum in England: science programmes of study | Keep it cool |
| Science, national curriculum, secondary physics | “conservation of energy in a closed system, dissipation”National curriculum in England: science programmes of study | Rack and stack, Keep it cool, Lights out |
| Science, national curriculum, secondary physics | “calculating energy efficiency for any energy transfers”National curriculum in England: science programmes of study | Keep it cool |
| Curriculum for Wales, Science and Technology | “Computation is the foundation for our digital world.”Science and Technology: statements of what matters | Where is the cloud?, Who looks after what? |
| Curriculum for Excellence, Technologies | “business, computing science, digital literacy, food, textiles, craft, design, engineering, and graphics”Technologies in Curriculum for Excellence | Eight kinds of job |
| Planned, first teaching from 2028 | “publish a revised national curriculum in 2027 for first teaching in 2028”Government response to the Curriculum and Assessment Review (PDF, 61 pages) | Where is the cloud? |
| Planned, GCSEs updated for first teaching from 2029 | “We will replace the computer science GCSE with a GCSE in computing”Government response to the Curriculum and Assessment Review (PDF, 61 pages) | Where is the cloud? |
Be aware of the limits
- The words cloud, server, virtual and data centre appear nowhere in the national curriculum computing programmes of study or in the Department for Education's GCSE computer science content. The honest link at secondary level is the statement about how computer systems communicate with one another and with other systems, and the GCSE content on environmental impacts.
- The statement about computer networks and the internet sits in the primary programme of study. The secondary programmes do not name networks or the internet.
- The exam boards go further than the Department. AQA's specification names cloud storage, and OCR's names the cloud and servers. Neither names a data centre, and neither names virtualisation as content. Check your own board's current specification.
- The Department removed the phrase inside computers from the GCSE content because it may inhibit coverage of cloud computing. That change applies to GCSE exams from summer 2027.
- The sums in these games sit in mathematics and science, not in computing. The mathematics row is the wording on combined events and knowing the underlying assumptions, which is the assumption the spare machines game keeps coming back to: that the parts fail independently. The science rows are power, energy transfer, dissipation and efficiency. We have not mapped the electricity and thermal physics in more detail.
- The two rows from Wales and Scotland are one line each, to show that computing and technology are in those curricula too. In Scotland we could read only the general description of the technologies area and not the experiences and outcomes, and Northern Ireland is not covered.
- The last two rows are a plan, not a curriculum. The government's response to the curriculum and assessment review, published in November 2025, says it will replace the computer science GCSE with a GCSE in computing, publish a revised national curriculum in 2027 for first teaching in 2028, and update GCSEs for first teaching from 2029. Nothing has been published yet, and the response says only that it will explore a new level 3 qualification in data science and AI.
- We read the Department for Education's own documents and the current specifications from two exam boards. Check your own board's current specification, and for England the version your school currently follows.
Straight answers
Questions students, parents and colleagues ask
Do you need a degree to work in cloud and infrastructure?
For most of the apprenticeship standards the Career Matrix maps, no. The DevOps engineer standard, ST0825, names A levels, a Level 3 apprenticeship or another relevant qualification, relevant experience and an aptitude test as the likely entry, and lists no degree. The IT solutions technician and Information communications technician standards state no entry requirements of their own on the Skills England pages we read.
A degree matters in two places. The Level 6 integrated degree apprenticeship in digital and technology solutions, ST0119, has a degree built into it, and the Level 7 specialist standard, ST0482, says entry is likely to include a degree at 2.1 or higher in a relevant subject. We found no official source that says how many people in cloud and infrastructure jobs hold a degree, so we do not give a figure.
Is there a cloud engineer apprenticeship?
No. Skills England has no current standard with cloud engineer, data centre technician or site reliability engineer in its title. Those job titles appear inside other standards: data centre support technician and cloud technician in the Information communications technician standard, ST0973, and site reliability engineer and platform engineer in the DevOps engineer standard, ST0825. The Network engineer standard, ST0127, lists network and cloud engineer.
The government's own role framework also has no site reliability engineer, cloud engineer or platform engineer role. Its closest are infrastructure engineer, infrastructure operations engineer and DevOps engineer.
What is the hands-on, hard hat side of cloud and infrastructure?
It is the physical side: installing and keeping running the servers, cooling, cabling and backup power in a data centre. The Career Matrix calls the area Data Centre Operations and marks it as a site-based trade.
The sourced routes in are a Level 2 foundation apprenticeship, FA0004, normally for people aged 16 to 21 at the start, the Level 3 Information communications technician standard, ST0973, and the Level 3 refrigeration, air conditioning and heat pump standard, ST0322, which names data centres once in a list of places the work is done. We found no standard that covers a data centre's backup power, such as batteries and generators, and the electrician and building services standards we read do not mention data centres. Work on electrical equipment carries rules: the HSE says people must be competent to do it.
Is there a shortage of cloud and data centre workers?
We found no government figure for cloud or data centre vacancies, so we do not say there is a shortage. What we did find is partial. The Uptime Institute's 2025 survey of data centre operators worldwide found that 46% report difficulty finding qualified candidates, and that shortfalls in electrical and mechanical trades continue to be significant. It is an industry survey, not a UK one.
The government's AI Growth Zones paper projects more than ten thousand jobs from its programme. That is a projection, not a count. The Data Centre Covenant expects developers to invest in paid apprenticeships and training places, and gives no number.
What does 99.9% availability actually mean?
It means the service may be down for up to 0.1% of the time. Over a year that is about 8.76 hours, in Google's table, which assumes no planned downtime. Each extra nine cuts the allowance to a tenth: 99% allows 3.65 days a year, 99.99% allows 52.6 minutes and 99.999% allows 5.26 minutes.
It is a promise about the average, not about when. Ten minutes every day and one long outage can both come to the same figure. And the Uptime Institute says that a data centre's Tier does not certify an availability percentage, so a Tier is not a promise of nines.
Does the cloud use a lot of electricity?
It uses a lot in total, and the figures depend on who is counting. The Department for Energy Security and Net Zero estimates that data centres used 4.5 TWh, which is 2% of the electricity consumed from the grid in Great Britain in 2024, and that most of it was in the South East and London. The National Energy System Operator's estimate for 2023 is higher, 7.6 TWh, because it counts differently.
Globally, the International Energy Agency says data centres used around 1.5% of the world's electricity in 2024 and projects around 945 TWh by 2030 in its base case. For the future in Great Britain, be careful: around 73 GW of data centre projects are in the grid connection queue, and the National Energy System Operator says that pipeline is likely to significantly overstate demand.
How much water do data centres use?
Nobody can say for the whole country. The government's water resources framework says it is experiencing barriers in gaining information about data centre water consumption, and says data centres can require a large amount of water for cooling. The draft National Policy Statement says newer closed, waterless cooling can reduce or remove the need for water.
The only survey figures we found are from techUK's voluntary survey of 73 mainly commercial sites in England: 51% used waterless cooling systems and 64% used less than 10,000 cubic metres of water a year. techUK says it is not a definitive snapshot.
Is the cloud green?
It depends on what you measure. Power usage effectiveness, PUE, is the total power a site draws divided by what its computers draw. Uptime's 2025 survey found a weighted average of 1.54, and the government expects new data centres to be designed to achieve 1.3 or better.
PUE leaves out things such as the water a site uses and how efficient the computers are. And the carbon in each kilowatt-hour depends on the grid: the government's 2026 factor for UK electricity is 0.13096 kg of CO2e per kWh, which is a reporting average and not a live reading.
Who owns the cloud?
A few very large companies supply most of it. The Competition and Markets Authority found that Microsoft and AWS each had between 30 and 40% of infrastructure as a service in 2024, across the UK and the European Economic Area, with Google lower, and that less than 1% of customers switch provider each year. Its final decision was that competition is not working well in cloud services.
Ofcom's earlier figure, a combined 70 to 80% for the two in the UK in 2022, measures a different thing in a different year, so do not set them side by side. In March 2026 the CMA said it would launch an investigation into Microsoft's business software ecosystem.
What happens when a data centre fails?
It depends on how, and the three stories on this page are three different ways. In France in 2021, the investigators say a fire started in the rooms with the batteries and the uninterruptible power supply and destroyed a building, and that the cause was not established when they reported. In London in 2022, Google says several cooling systems failed together in a heatwave. At AWS in 2025, the company says a bug in the automatic system that manages DNS records left one record empty.
In each, the company or the investigators say what changed afterwards. We show only what the documents say, and we do not add a cause that they do not give.
Who is responsible for security in the cloud?
Both sides, and the split depends on the kind of service. The National Cyber Security Centre says there is a joint responsibility for the security and availability of data and workloads in a cloud service, shared between the provider and the customer.
It says the provider will almost always be responsible for the availability and physical protections of its servers, and that the customer is always responsible for setting up the services it chooses securely and for deciding what data to store in them. AWS and Microsoft each describe their own model in their own words, and those are the companies' descriptions, not independent ones.
Do students need a vendor certificate?
Employers name them, and the two beginner ones are AWS Certified Cloud Practitioner and Microsoft Azure Fundamentals. AWS says its exam is designed for people new to cloud who may not have an IT background, and lists the exam as 90 minutes with 65 questions. Microsoft lists Azure Fundamentals at beginner level with a 45 minute exam.
Passing either one does not promise a job. Fees and exam details change, so send students to the issuer's own page.
Where does this fit in the curriculum?
Honestly, in a few places and thinly. The words cloud, server, virtual and data centre appear nowhere in the national curriculum computing programmes of study or in the Department for Education's GCSE computer science content. AQA's specification asks students to explain the term cloud storage, and OCR's lists the cloud and servers providing services. Neither names a data centre.
The government has said it will replace the computer science GCSE with a GCSE in computing and publish a revised national curriculum in 2027 for first teaching in 2028. That is a plan, and nothing has been published yet.
Are data centres part of the country's critical infrastructure?
In the UK, yes. On 12 September 2024 the Technology Secretary announced that UK data centres are classed as Critical National Infrastructure, the first such designation in almost a decade. The government said it allows it to support the sector in the event of critical incidents.
That describes the government's intent at the time of the announcement, and it is not an audit of what has happened since.
Coming from employers
From the industry
Resources shared by employers
The first employer resources are on their way. If you work for an employer with a TechSkills portal account, you can share a resource from your portal and it will appear here once it has been reviewed.
Articles and stories
Industry Insights
Beyond the Silos: Why the UK's 2035 Economic Ambitions Require an Open Infrastructure
techUK argues the UK's digital infrastructure is at a genuine turning point: keep building isolated, incompatible platforms and risk 'platform sprawl', or commit to open, standardised architecture that keeps skills and systems portable across cloud, government and industry.
Industry Insights
Welcome to techUK's 2026 Tech & Innovation Focus Week
techUK's 2026 Tech & Innovation Focus Week brought together member organisations across robotics, agentic AI, photonics, quantum and tech convergence, alongside a new photonics report and upcoming events including November's Tech and Innovation Summit.
Industry Insights
Event Round-Up: Future Careers in Emerging Tech
A round-up of techUK's Future Careers in Emerging Tech event, where panellists made the case that routes into AI, quantum, cybersecurity and semiconductors run through far more than traditional STEM degrees - and that soft skills matter just as much as technical ones.
Industry Insights
How Is Cloud Powering the Future of Work?
A techUK Cloud Week panel on what cloud adoption actually demands of a workforce: not just new tools, but a genuine shift in how organisations train, retrain and measure the people using them.
Over the coming months this is where employers will add their own articles, videos, lesson ideas and day-in-the-life stories for teachers, from cloud, infrastructure and data centre teams. Nothing appears here until the TechSkills team has read it and published it, and every piece names the organisation that shared it.
Where to next
Careers guides for Cloud and Infrastructure
Every area of work in Cloud and Infrastructure, the routes in, pay and hiring, and the skills employers ask for, with sources and dates, are in our Cloud and Infrastructure guides. There is one for teachers and one for careers advisers.
Show your working
Sources and how we checked them
Each source below was opened and read on 5 November 2026, and every figure on this page is the one the source gives, with the condition the source puts on it. Where a source warns about its own figures, as the government's draft policy, the Competition and Markets Authority and the Uptime Institute do, we pass the warning on. Where the sources disagree, as they do on the number of data centres and how much electricity they use, we show both.
- Cloud and Infrastructure Career MatrixTechSkills. Maintained by TechSkills, apprenticeship standards last checked 5 November 2026.Used for: The eight areas of cloud and infrastructure work, the jobs in them, pay with its own source, and the apprenticeship standards the Matrix maps.
- Apprenticeship search (full listing, 2086 entries)Skills England. Live listing, read 5 November 2026.Used for: The gap check: every non-retired standard checked by title and route
- Installation and maintenance electrician (ST0152)Skills England. Summary page read 5 November 2026.Used for: Checking whether it names data centres.
- Building services engineering service and maintenance engineer (ST0061)Skills England. Summary page read 5 November 2026.Used for: Checking whether it names data centres.
- Maintenance and operations engineering technician (ST0154)Skills England. Summary page read 5 November 2026.Used for: Checking whether it names data centres.
- IT solutions technician (ST0505) v2.0, Level 3Skills England. Date updated 02/11/2026; approved for delivery 18 February 2019; v2.0 earliest start 02/11/2026.Used for: Current version, level, duration, funding, purpose, job titles, progression routes
- IT solutions technician (ST0505) v1.1Skills England. Status Retired; Date updated 02/11/2026; latest start 01/11/2026.Used for: Shows the version the Matrix currently links to is now retired
- DevOps engineer (ST0825) v1.1, Level 4Skills England. Date updated 17/06/2026; approved for delivery 12 March 2020.Used for: Level, duration, funding, status, duties, job titles, entry wording, and the notice of a revised version from 18 December 2026
- DevOps engineer (ST0825) revised version awaiting implementationSkills England. Page updated 17 June 2026; its status line reads Standard in development.Used for: The revised version: replaces v1.1 for new starts from 18/12/2026, funding band £17,000, extra job titles (cyber operations, modernisation operations, network operations engineer)
- Information communications technician (ST0973) v1.2, Level 3Skills England. Date updated 30/06/2026; approved for delivery 4 May 2021.Used for: Level, duration, funding, options, purpose, job titles incl. data centre support technician, revision notice
- Refrigeration air conditioning and heat pump engineering technician (ST0322) v1.3, Level 3Skills England. Date updated 08/12/2025; approved for delivery 10 September 2015.Used for: Level, duration, funding, purpose, job titles, mandatory qualifications, data centre mention
- Network engineer (ST0127) v2.0, Level 4Skills England. Date updated 02/11/2026; approved for delivery 12 November 2014.Used for: Current network engineer version; job titles include "Network and cloud engineer" and "Infrastructure engineer"
- Digital support technician (ST0120) v2.0, Level 3Skills England. Date updated 03/11/2026; v2.0 earliest start 29/10/2026.Used for: Adjacent standard not in the cloud-infra block
- Hardware, network and infrastructure foundation apprenticeship (FA0004) v1.0, Level 2Skills England. Approved for delivery 29/04/2025.Used for: Level 2 entry route that names cloud technician and data centre support technician
- Digital and technology solutions professional (integrated degree) (ST0119) v1.2, Level 6Skills England. Date updated 01/09/2023; approved for delivery 26 March 2015.Used for: Degree route with "network engineering professional" option
- Digital and technology solutions specialist (integrated degree) (ST0482) v1.0, Level 7Skills England. Date updated 14/01/2022; approved for delivery 7 August 2018; page notes it is being revised.Used for: Level 7 route with architecture and operations specialist options; Level 7 funding rule
- Cyber security technician (ST0865) v1.1, Level 3Skills England. Date updated 08/09/2026; approved for delivery 13 May 2020.Used for: Closest "security technician" standard, for the Cloud Security domain
- Introduction of T LevelsDepartment for Education, GOV.UK. Updated 26 January 2026.Used for: T Level length, placement hours, classroom hours
- Occupational specialism: Digital infrastructure (TL0007a)Skills England. Approval date 17/12/2020; available from 01/09/2020.Used for: T Level specialism, aligned standards, placement minimum
- Digital Support and SecurityT Levels (DfE). Page carries no date.Used for: Student-facing description, core content, four specialisms
- National curriculum in England: computing programmes of studyDepartment for Education. Published 11 September 2013, with no later update shown.Used for: The wording of the computing programmes of study for secondary school
- National curriculum in England: mathematics programmes of studyDepartment for Education. Updated 28 September 2021.Used for: The wording on the probability of combined events, including the underlying assumptions
- National curriculum in England: science programmes of studyDepartment for Education. Updated 6 May 2015.Used for: The wording on power, energy transfer, dissipation and efficiency in the secondary science programme
- GCSE subject content for computer scienceDepartment for Education. Updated 31 March 2025.Used for: National GCSE content wording
- GCSE Computer Science subject content: government consultation response (PDF)Department for Education. 31 March 2025.Used for: The five removals, the cloud computing reason, and the 2027 start
- AQA GCSE Computer Science 8525 specification, version 1.3AQA. Version 1.3, 16 June 2025; GCSE exams June 2027 onwards.Used for: Cloud storage lines in 3.4.5 and 3.8; absence of server, data centre
- AQA GCSE Computer Science 8525 specification, version 1.2 (exams June 2022 to 2026)AQA. Version 1.2, 29 November 2022.Used for: Shows the cloud storage lines are the same in the earlier specification
- OCR GCSE (9-1) Computer Science J277 specification, version 3.1OCR. Version 3.1, May 2026.Used for: Section 1.3.1 lines on servers, DNS, the Cloud
- Curriculum and Assessment Review Final Report: government responseDepartment for Education. Published 5 November 2025, last updated 10 December 2025.Used for: The plan for the computing curriculum and GCSE, with dates
- Government response to the Curriculum and Assessment Review (PDF, 61 pages)Department for Education. Published 5 November 2025.Used for: The exact wording on computing and the 2027, 2028, 2029 dates
- Technologies in Curriculum for ExcellenceEducation Scotland. Last updated 31/10/2025.Used for: Scotland: what the technologies area covers (general only)
- Science and Technology: statements of what mattersHwb (Welsh Government). Page date not shown in text; the area page shows first published 30 April 2019, last updated 28 January 2020.Used for: Wales: the "Computation" statement
- Infrastructure engineer, Government Digital and Data Profession Capability FrameworkGovernment Digital Service, GOV.UK. Published 7 January 2020, last updated 28 August 2026.Used for: Role summary, five levels, IT infrastructure skill levels
- Infrastructure operations engineer, same frameworkGovernment Digital Service. Published 7 January 2020, last updated 30 November 2024.Used for: Role summary
- Development operations (DevOps) engineer, same frameworkGovernment Digital Service. Published 7 January 2020, last updated 29 August 2025.Used for: Role summary, nine levels
- Technical architect, same frameworkGovernment Digital Service. Last updated 28 August 2026.Used for: Role summary, five levels
- Solution architect, same frameworkGovernment Digital Service. Last updated 28 August 2026.Used for: Role summary, five levels
- Government Digital and Data Profession Capability Framework homeGovernment Digital Service. Current page, retrieved 2026-11-05.Used for: The full role list (no site reliability engineer or cloud engineer role listed)
- AWS Certified Cloud PractitionerAmazon Web Services. Page carries no date, read 5 November 2026.Used for: Exam length, format, fee, who it is for
- Microsoft Certified: Azure FundamentalsMicrosoft Learn. Last Updated 07/20/2026.Used for: Whether offered, exam time, what is assessed
- Data centres to be given massive boost and protections from cyber criminals and IT blackoutsDepartment for Science, Innovation and Technology (DSIT), GOV.UK press release. Published 12 September 2024.Used for: The CNI announcement and what DSIT said it means.
- Energy Trends: June 2026, special feature article - Data centre electricity consumption in Great Britain, 2020 to 2024Department for Energy Security and Net Zero (DESNZ). Official Statistics in Development. Published 30 June 2026.Used for: Best official figure for GB data centre electricity use and the count of data centres in operation (Table 1 of the workbook).
- Estimate of Data Centre Capacity: Great Britain 2024Department for Business, Innovation, Science and Trade and DSIT (GOV.UK, reproduces answer to Parliamentary Question HL5239). Published 1 May 2025; the underlying answer was first published 11 March 2025; capacity is estimated as of autumn 2024.Used for: Official capacity estimate in megawatts and its definition.
- Draft National Policy Statement for Data Centres (document title: Initial Proposal for a National Policy Statement (NPS) for Data Centres)Department for Business, Innovation, Science and Trade. Applies to England. Published 28 October 2026; consultation closes 8 January 2027.Used for: Government's current count of data centres, 6 GW target, water and PUE expectations (draft, not final policy).
- Future Energy Scenarios: Pathways to Net Zero (V.5), section "Commercial | Data centres", page 110National Energy System Operator (NESO). November 2025.Used for: NESO baseline for GB data centre demand and its definition.
- Curate - Demand Connections Reform (consultation)Ofgem. Publication date 29 July 2026; response deadline 16 September 2026.Used for: Size of the grid connection queue and the data centre share of it.
- Electricity Markets Advisory Council, 22 September 2026 (slide deck, "Data centres" slide)NESO. 22 September 2026.Used for: NESO's own warning that the data centre connection pipeline overstates likely demand.
- Energy and AI, Executive summaryInternational Energy Agency (IEA). Published 10 April 2025.Used for: Global data centre electricity figures and projections.
- AI Opportunities Action PlanDSIT, GOV.UK. Published 13 January 2025.Used for: Origin of the AI Growth Zones idea.
- Delivering AI Growth ZonesDSIT, GOV.UK policy paper. Published 13 November 2025.Used for: Grid connection as the main blocker; government jobs projection.
- The Data Centre CovenantHM Treasury, GOV.UK policy paper. Published 28 October 2026.Used for: Government tests for data centres (PUE 1.3, water, jobs and apprenticeships, 6 GW).
- National Framework for Water Resources 2025, chapter 9.4 "Water for data centres and artificial intelligence"Government policy paper on GOV.UK (Applies to England and Wales). Updated 27 August 2026.Used for: Government statement on data centre water for cooling and the data gap.
- New report sheds light on commercial data centres water usage in England (press release for "Understanding Data Centre Water Use in England")techUK, with a quote from the Environment Agency's Deputy Director for Water Resources. 18 August 2025.Used for: Only survey-based water figures found for English data centres. Voluntary survey, see qualifier.
- Cloud services market investigation, Summary of final decisionCompetition and Markets Authority (CMA). 31 July 2025.Used for: Market shares, spend, switching rate, decision.
- Cloud services market investigation (case page)CMA, GOV.UK. Published 5 October 2023, last updated 31 July 2025; case closed 31 July 2025.Used for: Outcome and SMS recommendation. The page's "next steps" wording ("decision ... in the first quarter of 2026") is out of date against the March 2026 press release.
- CMA announces package of actions on business software and cloud servicesCMA, GOV.UK press release. Published 31 March 2026.Used for: What happened after the July 2025 decision.
- Cloud services market study, Final reportOfcom. Published 5 October 2023.Used for: 2022 market shares and the referral to the CMA.
- The Technology Code of PracticeGovernment Digital Service, GOV.UK. Published 14 July 2021, last updated 29 October 2026.Used for: One-line cloud first statement (point 5).
- UK Business Data Survey 2026 (2025 to 2026 fieldwork)DSIT, Official Statistics (fieldwork by Ipsos). Published 18 June 2026.Used for: Share of UK businesses using a public cloud provider, with base and survey scope.
- Cloud security shared responsibility modelNational Cyber Security Centre (NCSC). Published 10 May 2022, Reviewed 7 June 2023, Version 2.1.Used for: UK government guidance on who is responsible for what.
- Shared Responsibility ModelAmazon Web Services (vendor's own published model). Page carries no date.Used for: "Security of the cloud" versus "in the cloud" in AWS's wording.
- Shared responsibility in the cloudMicrosoft Learn (Azure). Last updated 24 August 2026.Used for: Microsoft's wording of the same idea.
- Uptime Institute Global Data Center Survey 2025 (Keynote Report 180)Uptime Institute. July 2025; survey run online April to May 2025.Used for: Average PUE, PUE definition wording, outage and staffing findings.
- Data Centres programmetechUK. Page carries no date.Used for: techUK's own statement on key worker status for the data centre workforce.
- Uptime Institute Tier Classification SystemUptime Institute. Page carries no date.Used for: Tier I to IV descriptions in Uptime's own words.
- Publicity Kit (Tier Certification)Uptime Institute. Page carries no date.Used for: Uptime's statement that Tier awards do not certify an availability percentage.
- Embracing Risk (Chapter 3, Site Reliability Engineering book)Google (O'Reilly), sre.google. 2017 (copyright Google, Inc.).Used for: nines, allowed downtime formula, error budget idea in Google's own words
- Appendix A, Availability Table (Site Reliability Engineering book)Google (O'Reilly), sre.google. 2017 book, with no separate page date.Used for: allowed unavailability per year/quarter/month for 90% to 99.999%
- Availability, AWS Well-Architected Framework, Reliability PillarAmazon Web Services. Page carries no date, read 5 November 2026.Used for: series (hard dependency) and parallel (redundant) availability formulas, with the independence wording
- Define reliability and recovery targets (Azure Well-Architected Framework, Reliability)Microsoft Learn. Last updated 20 August 2024.Used for: composite SLO is a product of the parts; SLA only covers the conditions in the agreement. (The older "business-metrics" composite-SLA URL redirects here.)
- Amazon Compute Service Level AgreementAmazon Web Services. Last updated 25 May 2022.Used for: region-level 99.99% and instance-level 99.5% commitments with service credit tiers
- Compute Engine Service Level Agreement (SLA)Google Cloud. Current version, last modified 14 October 2026.Used for: multi-zone and single-instance monthly uptime commitments and financial credit tiers
- Availability options for Azure Virtual MachinesMicrosoft Learn. Last updated 5 December 2025.Used for: Microsoft's own statement that availability sets are for the 99.95% Azure SLA (the Azure VM SLA document itself could not be read, see Part 4)
- Service Level Objectives (Chapter 4, Site Reliability Engineering book)Google (O'Reilly), sre.google. 2017 book.Used for: "nines" naming and the point that SLOs are not 100%
- The NIST Definition of Cloud Computing, Special Publication 800-145National Institute of Standards and Technology (US Department of Commerce). September 2011.Used for: five essential characteristics, three service models, four deployment models, rapid elasticity, measured service
- ISO/IEC 30134-2:2026 Information technology - Data centres key performance indicators - Part 2: Power usage effectiveness (PUE)ISO catalogue. Edition 2, January 2026.Used for: PUE is a standard KPI. The catalogue pages do not print the formula.
- The Electricity at Work Regulations 1989, regulation 14 (Work on or near live conductors), as madelegislation.gov.uk. Statutory instrument 1989 No. 635, 1989.Used for: Statutory wording on live working.
- Work on electrical equipment, machinery or installationsHealth and Safety Executive (HSE). Updated 15 April 2025.Used for: HSE competence wording for electrical work.
- Working at height (INDG401)HSE. Updated 7 August 2025.Used for: HSE statement on falls from height.
- 19-inch rackWikipedia (large neutral source; it lists EIA-310 and IEC 60297 as the formal standards). Last edited 12 October 2026.Used for: 19 inch width, 1U = 1.75 in (44.45 mm), 42U as a common height, post spacing
- 2021 Equipment Thermal Guidelines for Data Processing Environments, ASHRAE TC 9.9 Reference CardASHRAE (Technical Committee 9.9). 2021 and 2024, Fifth Edition of Thermal Guidelines for Data Processing Environments.Used for: recommended 18 to 27 degrees C for classes A1 to A4 and the allowable ranges. NOTE: the card says it may not be quoted or reproduced without ASHRAE permission, so link to it and state the figure in our own words; do not paste the table.
- Best Practices Guide for Energy-Efficient Data Center DesignUS Department of Energy (Federal Energy Management Program). Revised July 2024.Used for: what a UPS does (government wording)
- Tier Classification SystemUptime Institute. Page carries no date, read 5 November 2026.Used for: Tier I requirements name a UPS and an engine generator
- Design Considerations of UPS Systems in Data CentersSchneider Electric (a UPS and generator vendor; vendor blog, not a standards body). 20 July 2020.Used for: typical generator start and transfer time and UPS runtime ranges (vendor view only)
- Key terms explained: Non-Domestic RHIOfgem (the GB energy regulator). Page carries no date, read 5 November 2026.Used for: kilowatt-hour and kilowatt definitions (the page is about renewable heat, the definitions are general)
- Greenhouse gas reporting: conversion factors 2026 (full set spreadsheet, UK electricity and Transmission and distribution sheets)Department for Energy Security and Net Zero (DESNZ), GOV.UK. Published 11 June 2026, last updated 31 July 2026.Used for: UK grid average electricity factor in kg CO2e per kWh for 2026 (generation basis) plus the T&D factor
- 2026 Government greenhouse gas conversion factors for company reporting: Methodology paperDESNZ, GOV.UK. part of the 11 June 2026 publication.Used for: the 2026 method change for the electricity factor
- Amazon EC2 On-Demand PricingAmazon Web Services. Page carries no date, read 5 November 2026.Used for: billing by the hour or second, no long-term commitment (no price used)
- Amazon EC2 Reserved InstancesAmazon Web Services. Page carries no date, read 5 November 2026.Used for: reserved/committed discount idea in AWS's words (percentages on the page are undated, do not print them)
- What is FinOps?FinOps Foundation. Updated November 2026.Used for: the Foundation's definition of FinOps
- Rapport d'enquete sur l'incendie au sein du centre de stockage de donnees OVH situe a Strasbourg (67) le 10 mars 2021 (report no. MTE-BEARI-2022-005, in French)BEA-RI, the French government's industrial-accident investigation bureau (CGEDD). Report dated 24 May 2022.Used for: official investigation: timeline, scenario, contributing factors, recommendations
- Strasbourg datacentre: latest information (press release, updated)OVHcloud. Press release of 18 March 2021, updated to 23 April 2021.Used for: OVH's own counts (servers destroyed, services affected), timings and rebuild decision
- OVH letter to the BEA-RI with its observations and follow-up to the recommendations (in French)OVH (published on the BEA-RI site). Letter dated 25 July 2022.Used for: what OVH says it changed after the report
- Multiple Cloud products experiencing elevated error rates, latencies or service unavailability in europe-west2 (Incident report)Google Cloud Service Health. Incident report posted 29 July 2022; the incident ran from 19 July 2022 to 20 July 2022, US/Pacific time.Used for: Google's own account of the London cooling failure
- Extreme heat stress-tests European data centers, againUptime Institute Journal (Daniel Bizo, Research Director). 27 July 2022.Used for: independent industry comment on the July 2022 heatwave
- POSTnote 762, What are data centres and how sustainable are they?UK Parliament, Parliamentary Office of Science and Technology. 16 March 2026.Used for: UK Parliament summary of the July 2022 London cooling failures; data centres designated critical national infrastructure in September 2024
- Summary of the Amazon DynamoDB Service Disruption in the Northern Virginia (US-EAST-1) RegionAmazon Web Services (post-event summary). Event on 19 and 20 October 2025; the page carries no publication date.Used for: AWS's own account of the October 2025 outage: causes, times, services, changes
This page is general information to help with teaching. It is not legal, safety or careers advice, and for England the curriculum links should be checked against the version your school currently follows.