Cloud Computing Market

Microsoft’s Azure revenue rose 48% to USD 33.2 billion in the quarter to 30 September 2026 and its commercial remaining performance obligation jumped 73% to USD 677 billion, its results of 28 October 2026 show, while Amazon Web Services (AWS) grew 2025 sales 20% to USD 128.7 billion, Amazon’s fourth quarter release of 5 February 2026 states. The cloud computing market covers public cloud services bought on subscription or by usage: software as a service (SaaS), infrastructure as a service (IaaS), platform as a service (PaaS) and AI and GPU cloud and model services. Private data centres, colocation leases, hardware sold to enterprises and consulting fees are excluded. Douglas Insights sizes it at USD 812.0 billion in 2025, about 58.0 gigawatts (GW) of cloud IT capacity earning an average USD 14.0 million per megawatt (MW) a year, and at USD 2,588.0 billion in 2035, a 12.29% compound annual growth rate (CAGR). The study sits beside our Telecom Cloud Market, Everything as a Service Market and Data Centre Liquid Cooling Market reports; every figure follows the Douglas Insights research methodology.

Which cloud service model earns the most: SaaS, IaaS, PaaS or AI and GPU cloud?

Software as a service earns the most, 46.0% of 2025 cloud computing value or about USD 373.5 billion, Douglas Insights estimates. AI and GPU cloud services are the smallest model, at 10.0%, but grow fastest at 19.5% a year.

SaaS leads because every office suite, customer relationship system and payroll tool now renews as a subscription, and those seats were sold years before AI. Microsoft Cloud, which includes the Microsoft 365 commercial suite as well as Azure, earned USD 63.9 billion in the September 2026 quarter, up 30%, the Microsoft release shows. SaaS grows only 8.6% a year because seat counts in mature markets are close to saturation, and growth now comes from price rises and AI add-ons, a Douglas Insights estimate.

IaaS, USD 211.1 billion, grows 12.8% as enterprises keep moving compute and storage out of their own data centres; Oracle’s cloud infrastructure revenue rose 55% to USD 3.3 billion in its quarter to 31 August 2025, its results of 9 September 2025 show. PaaS, USD 146.2 billion, grows 13.2% on databases, analytics and developer tools. AI and GPU cloud and model services, USD 81.2 billion, reach USD 482.2 billion by 2035. Service model values for 2035 sum to USD 2,543.6 billion, 1.72% below the headline.

Which buyers and deployment types drive cloud computing spend?

Large enterprises buy about 58% of 2025 cloud computing value, roughly USD 471 billion, Douglas Insights estimates; small and mid-sized businesses buy 27% and governments and public bodies 15%.

Large enterprises lead because they run the biggest workloads and sign the multi-year commitments that fill provider backlogs. Their spend grows about 12.8% a year, a Douglas Insights estimate, as they move core systems such as enterprise resource planning and data warehouses and add AI inference on top. Small and mid-sized businesses, about USD 219 billion, buy mostly SaaS through app marketplaces and grow 11.0%. Governments and public bodies, about USD 122 billion, grow 12.5% as sovereign cloud regions open in Europe, the Gulf and Asia.

Deployment matters for who earns the money. Douglas Insights estimates that 82% of 2025 value is pure public cloud, while 18% runs as dedicated or sovereign regions operated by a provider for one customer or one country. Multi-cloud buying is now normal: most large firms use at least two of the three biggest providers, which keeps price competition alive in IaaS while SaaS stays sticky.

By industry, financial services, technology and media, and retail and consumer goods together buy about 52% of 2025 cloud value, Douglas Insights estimates. Banks and insurers spend the most per employee because they run risk models, fraud checks and customer apps that need large, secure compute, while retailers buy elastic capacity for peak seasons. Manufacturing, healthcare and the public sector are the fastest-moving late adopters, each growing above 13% a year in our model, as factory data, medical imaging and citizen services move into hosted platforms. Telecom operators form a special case: they are both buyers of cloud capacity and partners that host edge zones inside their networks, a theme covered in our telecom cloud study.

What drives cloud computing demand: AI workloads, migration or new regions?

Cloud IT capacity rises 14.00% a year in our base case, from about 58.0 GW in 2025 to 215 GW in 2035. Four drivers add 16.50 gross points; restraints remove 2.50.

AI training and inference: 7.50 points

AI is the largest driver, worth 7.50 points. Azure revenue rose 48% to USD 33.2 billion in the September 2026 quarter, from USD 22.4 billion a year earlier, and Microsoft’s additions to property and equipment doubled to USD 39.4 billion, its release of 28 October 2026 shows. Douglas Insights estimates that each MW of AI-ready capacity earns about USD 18 million a year, more than the USD 14.0 million average, because GPU hours sell at a premium.

Enterprise workload migration: 4.50 points

Migration adds 4.50 points. AWS segment sales rose 24% to USD 35.6 billion in the fourth quarter of 2025 and its full-year operating income reached USD 45.6 billion, up from USD 39.8 billion, the Amazon release states, and Amazon expects to invest about USD 200 billion of capital across the company in 2026. Douglas Insights estimates that only about 35% of enterprise workloads ran in public cloud in 2025, leaving a large base still to move.

Asian cloud growth: 2.50 points

Asia adds 2.50 points. Alibaba’s AI cloud and compute services revenue rose 45% to RMB 48.4 billion, or USD 7.14 billion, in the quarter to 30 June 2026, and the segment’s adjusted earnings before interest, tax and amortisation rose 133% to RMB 5.63 billion, its results of 20 August 2026 show. Chinese, Indian and Southeast Asian buyers adopt cloud later than US buyers and so grow faster, a Douglas Insights estimate.

New providers and long contracts: 2.00 points

New capacity from challengers adds the last 2.00 points. Oracle’s total remaining performance obligations rose 359% to USD 455 billion in the quarter to 31 August 2025, and it expected cloud infrastructure revenue to grow 77% to USD 18 billion in fiscal 2026, its 9 September 2025 release states. Summed, 7.50 + 4.50 + 2.50 + 2.00 = 16.50 points; after 2.50 points of restraints the 14.00% capacity leg multiplied by the minus 1.50% revenue-per-MW leg gives the 12.29% revenue rate.

What holds back cloud computing growth: power, capital or sovereignty rules?

Three restraints remove 2.50 points a year from capacity growth: grid power and site limits take 1.20 points, the capital burden of building AI capacity 0.80 points and data sovereignty rules 0.50 points.

Power is the tightest limit. A hyperscale campus now needs hundreds of megawatts, and grid connections in Northern Virginia, Dublin, Amsterdam and Singapore take years to approve, a Douglas Insights estimate. Each GW of delayed capacity defers about USD 14 billion of yearly cloud revenue at the 2025 average, so grid timing, not demand, sets the near-term pace.

Capital is the second limit. Microsoft spent USD 39.4 billion on property and equipment in one quarter, double a year earlier, and Amazon plans about USD 200 billion of 2026 capital spending, the releases above show; Douglas Insights estimates that cloud providers need about USD 11 billion of capital per GW of AI-ready capacity, so higher interest rates or weaker AI returns would slow building. Sovereignty is the third. Customers in the European Union, India and the Gulf increasingly require data to stay in-country, which forces providers to build smaller, costlier regions and slows workload moves while contracts are renegotiated.

Which region buys the most cloud computing, and where does demand grow fastest?

North America buys the most cloud computing, about USD 381.6 billion or 47.0% of 2025 value. The Middle East and Africa grow fastest, at 15.0% a year, and Asia Pacific adds the second-largest increase, reaching USD 704.6 billion by 2035.

North America leads because the largest providers, the largest SaaS vendors and the largest AI model developers are all based there, and US enterprises moved to cloud first. Its value grows 11.6% a year to USD 1,143.7 billion. Europe, USD 178.6 billion, grows 11.2%, slowed by sovereignty rules but supported by new sovereign regions. Asia Pacific, USD 186.8 billion, grows 14.2% as Alibaba’s 45% cloud growth in the June 2026 quarter, reported in its results of 20 August 2026, shows Chinese demand returning.

Latin America, USD 36.5 billion, grows 13.4% as Brazil and Mexico gain new hyperscale regions. The Middle East and Africa, USD 28.4 billion, grow fastest at 15.0% on Saudi and Emirati national AI programmes, a Douglas Insights estimate. Regional values sum to USD 911.9 billion in 2026, 0.02% above the headline, and USD 2,608.2 billion in 2035, 0.78% above it.

Which companies lead cloud computing, and how concentrated is the field?

Microsoft leads cloud computing with about 22% of 2025 value, a Douglas Insights estimate that counts its Microsoft 365 SaaS as well as Azure, ahead of AWS at about 16% and Google at about 7%. The top three hold about 45%.

AWS remains the largest pure infrastructure provider, with 2025 operating income of USD 45.6 billion on USD 128.7 billion of sales, a 35.4% margin, the Amazon release shows. Microsoft leads overall once SaaS is included; its commercial remaining performance obligation of USD 677 billion is the largest disclosed backlog in the industry, the October 2026 release states. Oracle’s total cloud revenue rose 28% to USD 7.2 billion in the quarter to August 2025, its results show.

The rest of the field is wide. SAP, ServiceNow, Workday, Adobe and IBM hold large SaaS and PaaS books, Huawei Cloud, Tencent Cloud and Baidu AI Cloud compete in China, and GPU specialists such as CoreWeave rent AI capacity. Douglas Insights estimates that these and thousands of smaller SaaS vendors share the remaining 44% of value.

What does cloud capacity earn per megawatt, and how fast do cloud prices move?

Cloud capacity earns an average USD 14.0 million per MW of IT load a year in 2025, Douglas Insights estimates, from about USD 8 million for storage-heavy IaaS halls to about USD 18 million for GPU halls, falling 1.50% a year.

The average combines very different services. A megawatt filled with GPUs rented by the hour earns the most, while a megawatt of general compute and storage earns less because list prices for virtual machines and object storage keep drifting down as chips get more efficient. SaaS earns revenue on top of the infrastructure it runs on, so a SaaS vendor’s revenue per MW is higher still, which is why SaaS-heavy regions show higher yields. The average reaches about USD 13.8 million in 2026 and USD 12.0 million in 2035.

Two forces set the minus 1.50% revenue-per-MW leg. Newer chips deliver more compute per watt, so each MW of a newer hall can be priced lower per unit of work while still earning more in total; AI premiums on GPU capacity pull the other way. Douglas Insights estimates that efficiency takes about 3.0 points a year off revenue per MW and the shift to AI and SaaS adds back 1.5 points. Committed-use discounts, typically for one to three years, lock in part of that decline in advance.

Which rules shape cloud computing, from the EU Data Act to sovereignty mandates?

Three layers of rules shape cloud computing: the EU Data Act, applicable from 12 September 2025, which forces easier switching between providers; national data residency rules; and security certification for public sector buyers.

The EU Data Act has applied since 12 September 2025, the European Commission’s policy page states, and it requires cloud providers to help customers switch to another provider or back on-premises and to phase out switching charges. Douglas Insights estimates that easier switching takes about 0.3 points a year off European revenue-per-MW growth as customers use the threat of moving to win discounts, while it encourages multi-cloud designs that lift total spending.

Residency and certification rules sit on top. Governments in the European Union, India, Saudi Arabia and Indonesia require some public and regulated data to stay in-country, which leads providers to build local regions and partner with domestic operators. Public sector buyers also demand security accreditation, such as FedRAMP in the United States, which takes time and money and favours the largest providers. Douglas Insights treats a stricter split of EU cloud markets along national lines as a risk in the slower case, worth about 0.5 points of capacity growth.

What if AI demand slows or accelerates: how large is cloud computing in 2035?

Cloud computing ends 2035 between USD 1,635.0 billion and USD 3,712.3 billion across three cases, around a USD 2,588.0 billion base. AI demand and grid power set most of the spread.

In the slower case AI returns disappoint, providers cut capital plans and GPU prices fall quickly, taking 4.0 points off capacity growth and 1.0 point off revenue per MW. In the faster case AI agents spread through enterprise software and grid connections speed up, adding 3.0 points to capacity and 1.0 point to revenue per MW. One extra point of yearly revenue growth adds about USD 240 billion to 2035 value, Douglas Insights estimates.

Company signals sit above the base case for now. Azure grew 48%, Alibaba’s cloud unit 45% and AWS 24% in their latest reported quarters, the releases above show. Our 12.29% rate sits below these because SaaS, almost half the market, grows far more slowly and because today’s AI surge is assumed to cool as capacity catches up.

Douglas Exclusive: the Cloud Backlog Coverage Index

The Cloud Backlog Coverage Index is a Douglas Insights model built from 8 inputs: 5 company results releases, 2 capital spending disclosures and 1 regulatory page. It divides each provider’s contracted backlog by its yearly cloud revenue.

The finding, a Douglas Insights estimate: Microsoft’s USD 677 billion commercial backlog covers about 2.6 years of its USD 63.9 billion quarterly Microsoft Cloud run rate, while Oracle’s USD 455 billion backlog covered more than 15 years of its then cloud revenue. The industry has sold far more capacity than it has built, so the limit on cloud revenue to 2028 is power and chips, not demand.

The index changes how to read 2027. If backlog keeps growing faster than revenue, Douglas Insights expects capacity, not price, to decide market share, favouring providers with secured power; if large AI customers cut commitments, backlog coverage is likely to fall first at the newest entrants.

How does the methodology turn 58.0 GW into USD 812.0 billion?

Our methodology multiplies about 58.0 GW of cloud IT capacity in 2025 by an average USD 14.0 million of yearly revenue per MW to reach USD 812.0 billion, then grows capacity 14.00% and revenue per MW minus 1.50% a year.

The model covers 4 service models, 3 buyer groups and 5 regions, using 52 data points, 15 sourced and 37 modelled. Capacity comes from provider data centre disclosures and capital spending, converted to IT load; revenue per MW comes from company segment revenue divided by estimated capacity, all Douglas Insights estimates. Arithmetic: 58.0 GW x USD 14.0 million = USD 812.0 billion; USD 812.0 billion x (1.1400 x 0.9850)^10 = USD 2,588.0 billion. Capacity reaches about 66.1 GW in 2026 and 215 GW in 2035.

Two cross-checks agree. First, AWS’s USD 128.7 billion of 2025 sales is 15.8% of our total, close to its estimated 16% share. Second, Microsoft Cloud’s run rate of about USD 256 billion a year in late 2026 sits near 28% of our USD 911.8 billion 2026 total, above its 22% 2025 share, consistent with its faster growth. AI tools help draft the text and automated checks test the arithmetic and links; the Douglas Insights research desk chooses the sources, builds the model and approves publication, as our editorial standards explain.

Sources

1. Microsoft Microsoft fiscal year 2027 first quarter results (2026)
2. Microsoft Microsoft fiscal year 2026 first quarter results (2025)
3. Amazon Amazon fourth quarter 2025 results (2026)
4. Oracle Oracle fiscal year 2026 first quarter financial results (2025)
5. Alibaba Group Alibaba Group June quarter 2026 results (2026)
6. European Commission Data Act (2026)

What counts in the cloud computing study, and what is excluded?

The study counts public cloud SaaS, IaaS, PaaS and AI and GPU cloud and model services, including dedicated and sovereign regions run by providers. Private data centres, colocation leases, servers sold to enterprises, managed services and consulting are excluded.

The study splits value by 4 service models, 3 buyer groups and 5 regions: North America, Asia Pacific, Europe, Latin America and the Middle East and Africa. Values are in US dollars at 2025 average exchange rates, counted as provider revenue from end customers, net of discounts. Readers following cloud infrastructure can compare our Data Centre Liquid Cooling Market study.