So What Club
Start free
Data centres, cloud and AI compute
Lesson 3 of 3 Math checked Last reviewed 30 September 2026 12 min

Data centres and AI compute: players, trends and cases

Who the players are by region, the 2024 to 2026 build-out and sovereign AI programmes, regulation basics, and typical case prompts.

Industry brief, with a one-minute summary: Data centres, cloud and AI compute

Firm processes and online tests change from year to year and differ by office. Use this to prepare, and confirm the exact current steps on the firm's own careers page.

Key takeaways

  • Data centre cases are usually bets on three things at once: will the demand be there, can you get the power in time, and will the price still cover a very large investment when the capacity arrives?
  • Common traps: Starting a site case with land price or tax instead of power.
  • Cloud demand accelerated.
  • GPU clouds grew very fast, with debt.

Key idea

Data centre cases are usually bets on three things at once: will the demand be there, can you get the power in time, and will the price still cover a very large investment when the capacity arrives?

Examples of data centre, cloud and AI compute players by region (not a ranking)
Examples of data centre, cloud and AI compute players by region (not a ranking)
RegionExamples
Global cloud (hyperscalers)Amazon Web Services, Microsoft Azure, Google Cloud, Oracle; Meta builds for its own use
GPU clouds (neoclouds)CoreWeave, Nebius, Lambda, Crusoe (US and Europe)
Colocation, globalEquinix, Digital Realty, NTT Data, AirTrunk (owned by Blackstone)
EuropeGlobal colocation firms plus Data4 and OpCore (France, owned by Iliad and InfraVia), atNorth (Nordics), Mistral AI compute (France)
Middle EastG42 and Khazna (UAE), Humain (Saudi Arabia, owned by the Public Investment Fund), center3 (stc, Saudi Arabia)
IndiaYotta, CtrlS, Sify, NTT Data, AdaniConneX, Reliance Jio
Southeast AsiaSTT GDC and Keppel (Singapore), Princeton Digital Group, AirTrunk; large campuses in Johor (Malaysia) and Batam (Indonesia) near Singapore

So-what

Synergy Research estimates that in the second quarter of 2026 Amazon had about 28 percent of cloud infrastructure spending, Microsoft about 20 percent and Google about 15 percent, together about two thirds.

Trends from 2024 to 2026 (checked 30 September 2026)

  • Cloud demand accelerated. Synergy Research puts cloud infrastructure spending at about USD 143 billion in the second quarter of 2026, up about 43 percent on a year earlier, the fastest growth in eight years, driven by generative AI.
  • Record capital spending. For 2026, Amazon guided to about USD 200 billion and Alphabet raised its range to about USD 195 to 205 billion (CNBC, July 2026); Microsoft now reports about USD 175 billion including finance leases for calendar 2026, down from about USD 190 billion only because of a lease accounting change (July 2026); Meta guided to USD 130 to 145 billion (Meta, July 2026). Investors keep asking when this will pay back.
  • GPU clouds grew very fast, with debt. CoreWeave reported second-quarter 2026 revenue of about USD 2.6 billion and a revenue backlog (contracted future revenue) of about USD 104 billion, while still reporting a net loss because of interest on its borrowing (CoreWeave).
  • Capacity is scarce and costs are rising. CBRE reports record-low vacancy of about 1.4 percent in primary North American markets in the first half of 2026 and a record 7,481 MW under construction; Cushman & Wakefield reports build costs up about 21 percent per MW since late 2024.
  • Power became the limit. The IEA projects data centre electricity use roughly doubling from about 415 TWh in 2024 to about 945 TWh in 2030. In Ireland, data centres used 23 percent of metered electricity in 2025 (CSO), up from 5 percent in 2015.
  • Sovereign AI build-outs. Governments want AI compute inside their borders: the UAE's Stargate UAE plans a 1 GW cluster with a first 200 MW phase due in 2026 (G42; The National); Saudi Arabia's Humain targets several GW by the 2030s (reported); India's IndiaAI Mission offers more than 38,000 GPUs to startups and researchers at subsidised rates of about INR 65 per GPU hour (PIB); the EU's InvestAI plan includes EUR 20 billion for AI gigafactories (European Commission). Announced plans often change, so check progress.
Regulation basics (general, not legal advice)

Data centres need land zoning, building and environmental permits, and a grid connection approved by the local network operator. Data protection and data residency laws (the EU GDPR, India's Digital Personal Data Protection Act 2023, Saudi Arabia's Personal Data Protection Law) push some data to stay in the country, which creates local demand. The EU Energy Efficiency Directive requires larger data centres to report their energy and water use. The US controls exports of the most advanced AI chips to some countries, and these rules have changed several times since 2022, so check the current rules for any case involving AI chips and the Gulf, India or China.

Typical data centre and AI compute case prompts and how to crack them
Typical data centre and AI compute case prompts and how to crack them
Case promptStructure hintFirst driver to check
Where should we build a 300 MW AI campus?Power (amount, date, price, source), land, water and cooling, fibre, permits, tax, chip export rulesFirm power available and the date the grid can deliver it
Should a telecom company in Southeast Asia enter colocation?Market demand in MW, its land and fibre assets, capex, rivals, anchor tenantsSigned or likely anchor tenants for the first phase
Our GPU cloud is losing money. Why?Price per GPU hour, utilisation, depreciation life, power, interestUtilisation and contract length against the chips' useful life
Should a Gulf government fund a sovereign AI cluster?Strategic goals, demand from local users, cost per GPU hour, chip access, skills, partnersWho will use the compute and at what price
Due diligence on a data centre developerPower pipeline, leased MW and contract terms, tenant quality, build cost and delays, debtShare of the pipeline with secured power and signed leases

So-what

Start every data centre case with power and signed demand. Buildings are the easy part.

Common traps

Starting a site case with land price or tax instead of power. Treating the building (decades) and the chips (a few years) as one asset with one life. Ignoring that colocation customers usually pay for their own electricity, so electricity is not the operator's biggest cost. Assuming today's GPU prices last: prices fell sharply after 2024 on many platforms. Counting announced capacity as if it were built and powered.

Timed math drill

Cloud infrastructure spending was about USD 143 billion in a quarter, up about 43 percent on the same quarter a year earlier. Roughly what was spending a year earlier, in USD billions? Round to the nearest billion.

Related modules: "Semiconductors, electronics, and data centers" covers the chip supply chain; "Power and renewables" covers power purchase agreements and grid economics; "Software and SaaS" covers the companies that buy cloud; "Capacity, supply chain, and footprint" and "Investment and capital project decisions" are the case types these cases most often become.

Check your understanding

Why do governments in the Gulf, India and Europe fund sovereign AI compute?

Check your understanding

A colocation company's electricity bill rose 30 percent. Why might its profit barely change?

Sources for this lesson (23)
My notes on this lesson

0 of 5,000 characters. Saves automatically.

Try the 3 remaining checks and drills above to complete this lesson (0 of 3 done).

Spotted something wrong or out of date? Report a mistake. We check every report and correct the page.