AWS vs Microsoft vs Google vs Meta vs Oracle: Hyperscaler Data Center Footprints Compared

Hyperscaler data center footprints are the clearest way to see where the AI build-out is actually happening. This guide compares Amazon Web Services, Microsoft, Google, Meta and Oracle on cloud market share, regions, physical capacity, capital spending, contracted backlog and flagship campuses, using results reported through the end of October 2026. It is part of the Kurums Data Center Economy hub. It pairs naturally with our analysis of the AI capex supercycle and our ranking of the largest colocation providers that lease capacity to these same giants.

Who are the hyperscalers and why do their footprints matter?

Hyperscalers are the handful of companies that build and run data centers at a scale of hundreds of megawatts to gigawatts per campus. For data center purposes they are Amazon, Microsoft, Alphabet, Meta and, more and more, Oracle. Their capex sets the pace for the whole industry.

Synergy Research Group counted 1,360 large hyperscale data centers worldwide at the end of 2025, with almost 800 more in the known pipeline. Hyperscale operators held 48% of all worldwide data center capacity at that point. Synergy expects their share to reach 67% by 2031, while enterprise on-premise capacity falls from 32% to about 19%. In other words, the hyperscalers are absorbing workloads that companies once ran in their own server rooms. On top of that, they are absorbing almost the entire new category of AI training and inference.

Concentration is high even inside this group. Synergy’s Q3 2025 analysis found that AWS, Microsoft and Google alone controlled 58% of global hyperscale capacity, and that the United States hosted about 55% of it.

Nearly 60% of hyperscale capacity sits in facilities the operators own and built themselves. The rest is leased from wholesale developers such as QTS, Vantage, Aligned and Digital Realty. That split explains why hyperscaler capex guidance moves the share prices of landlords, power-equipment makers and chip suppliers at the same time. We cover the latter in our guide to the data center supply chain.

How do AWS, Microsoft and Google compare on cloud market share?

AWS leads with 28% of global cloud infrastructure services in Q2 2026, according to Synergy Research Group. Microsoft holds 20% and Google 15%. Together the three take about 63% of a market growing 43% a year, the fastest pace in eight years.

Synergy put Q2 2026 cloud infrastructure services revenue (IaaS, PaaS and hosted private cloud) at $143.4 billion. Trailing twelve-month revenue reached $500 billion, and the market has doubled in roughly eleven quarters. Generative-AI cloud services grew 165% year over year.

The long-run trend is a slow erosion of Amazon’s lead rather than a collapse. Synergy noted in late 2025 that AWS averaged just under 30% over four quarters, down from just over 32% in 2021, while Microsoft and Google gained. By Q2 2026 Google had moved up to 15% from 13% a year earlier. Its momentum shows in Alphabet’s own numbers: Google Cloud revenue reached $30.7 billion in Q3 2026, up 103% year over year, according to AI Weekly’s summary of the results.

Below the big three, Synergy lists Oracle, CoreWeave, Crusoe, Nebius, Nscale and even model developers like OpenAI and Anthropic among the fastest-growing “tier two” providers. Nine neocloud companies now rank among the top 40 cloud providers. Alibaba keeps growing but more slowly than the market.

Source: Synergy Research Group (Q2 2026); Amazon, Alphabet, Microsoft, Meta and Oracle quarterly results as summarised by Datastudios, AI Weekly, 24/7 Wall St., Let’s Data Science and company filings; AWS and Google Cloud infrastructure pages (accessed November 2026); DCHub (Azure).

How big is the AWS data center footprint?

AWS runs 39 geographic regions with 124 availability zones, and has announced new regions in Saudi Arabia and Chile. Amazon says it doubled AWS power capacity between 2022 and 2025, added 3.8 GW in the twelve months to Q3 2025 and plans to double again by 2027.

Those capacity figures came from Andy Jassy on Amazon’s Q3 2025 earnings call, as reported by DatacenterDynamics. Since then the financial scale has grown sharply. In Q3 2026 AWS revenue reached $46.0 billion, up 39% year over year, with $17.8 billion of operating income, according to a breakdown of Amazon’s October 29 results by Datastudios. Purchases of property and equipment for the quarter were $77.3 billion, more than double the year before. The AWS backlog reached $578 billion.

Amazon raised its 2026 cash capex plan from about $200 billion to roughly $220 billion in July, blaming mainly higher memory prices, and kept it there in October. Management has said that even this budget will not provide enough capacity for 2026 demand, that most 2027 AWS capacity is already reserved and that 2028 bookings are unusually strong.

Flagship AWS campuses

The best-known AWS AI site is Project Rainier, the Trainium2 cluster built largely for Anthropic. When AWS declared it operational in October 2025, Data Center Knowledge reported more than 500,000 Trainium2 chips across several US data centers, including an Indiana campus planned for 30 buildings of about 200,000 square feet each. Amazon’s custom silicon is now a capacity lever in its own right: Datastudios notes that revenue commitments for Trainium and Graviton each rose more than 20-fold year over year in Q3 2026. Northern Virginia remains AWS’s largest cluster. Oregon, Ohio, Indiana, Mississippi and Pennsylvania are other major US hubs (see our deep dive on US data center investment).

How large is Microsoft’s data center build-out?

Microsoft brought 88 new data centers online in fiscal 2026 and added about 1 GW of capacity in each of the last three quarters of that year. Azure grew 48% in the September 2026 quarter, and Microsoft’s commercial backlog reached $677 billion.

DatacenterDynamics reported, from Microsoft’s fiscal year-end call, that 31 of those 88 sites came online in the June 2026 quarter alone, spread across five continents. Management aims to roughly double total capacity within two years. Capex including finance leases was $145.3 billion for fiscal 2026 (July 2025 to June 2026), with about two-thirds going to servers and other IT hardware rather than buildings.

In July Microsoft changed how it accounts for long data center leases, spreading them over 25 years instead of 15. That lowered its calendar-2026 capex figure from about $190 billion to about $175 billion without changing actual spending plans. The company disclosed $329.1 billion of data center leases that had been signed but not yet started, with start dates from fiscal 2027 to 2033. For the September 2026 quarter, 24/7 Wall St. reported property and equipment spending of $39.4 billion, and CFO Amy Hood said demand still exceeds supply.

Fairwater and the ‘AI superfactory’ model

Microsoft’s flagship AI design is Fairwater, a family of tightly networked GPU campuses. It started in Mount Pleasant, Wisconsin, and continued with a site in Atlanta that Microsoft called an AI “superfactory” when it was unveiled in late 2025. The idea is to link several sites over dedicated fiber so they work as one giant training cluster. That makes the geographic footprint less important than the network between sites. Microsoft also relies heavily on leased capacity and on neoclouds: CoreWeave’s filings show Microsoft accounted for about 67% of CoreWeave’s 2025 revenue (see our neocloud guide).

How is Google building out its data center network?

Google Cloud lists 43 regions and 130 zones serving 200+ countries and territories. Alphabet guides to $195–205 billion of 2026 capex, spent $51.4 billion in Q3 2026 alone and has signaled that 2027 spending could top $300 billion.

Alphabet raised its 2026 range twice this year: from $175–185 billion to $180–190 billion in April, then to $195–205 billion in July, according to Yahoo Finance and FactSet-based reporting. On the October call, AI Weekly’s summary says management signaled 2027 capex could exceed $300 billion. Google Cloud’s $30.7 billion quarter and 103% growth explain why investors largely tolerated the jump.

Google’s footprint differs from its rivals’ in one important way: it designs its own AI accelerators, the Tensor Processing Units (TPUs), and builds data halls around them. That gives it more control over cost per unit of compute, and it has started offering TPU capacity to outside AI labs. Like its peers, Google is signing long-term clean power and nuclear agreements to secure electricity for these campuses, a trend we track in our article on nuclear deals for data centers.

What makes Meta’s data center strategy different?

Meta builds hyperscale capacity only for its own apps and AI models; it sells no public cloud. Even so, its 2026 capex range of $135–145 billion puts it in the same league as the cloud providers, and its Hyperion campus in Louisiana is planned to reach 5 GW.

For Q3 2026 Meta reported revenue of $64.6 billion, up 26%, and capex of $34.9 billion including finance-lease principal, according to Let’s Data Science. Free cash flow turned slightly negative for the quarter, at about −$2.1 billion.

Hyperion is the defining project. Meta confirmed in July 2026 that the Richland Parish campus will scale to 5 GW across about 3,200 acres, with the first 2 GW expected by 2030, according to Global Data Center Hub. Satellite and permitting analysis by Epoch AI projects the first phase becoming operational around early 2028, with roughly 1.6 GW of IT power. The financing is as notable as the size: a Blue Owl special-purpose vehicle owns 80% of the joint venture, Meta 20%, and the vehicle issued about $27.3 billion of debt. That structure keeps much of the cost off Meta’s balance sheet (see how data centers are financed). Meta has also partnered with BlackRock’s Global Infrastructure Partners on a roughly 1 GW, $14 billion campus in El Paso.

Why is Oracle now counted among the hyperscalers?

Oracle moved into the hyperscaler tier by signing huge AI infrastructure contracts, above all with OpenAI. Its remaining performance obligations reached $664 billion in August 2026, and cloud infrastructure revenue grew 121% to $7.4 billion in the quarter.

For the quarter ended August 31, 2026 (fiscal Q1 2027), Oracle reported total revenue of $19.3 billion, up 30%, and booked more than $30 billion of new AI cloud contracts. Capex was about $28.5 billion, which pushed free cash flow to roughly −$5.4 billion. The company delivered about 850 MW of data center capacity in that one quarter. Fiscal 2026 capex, according to Oracle’s 10-K, was $55.7 billion. Oracle plans to raise about $40 billion of external funding in fiscal 2027, and it says $75 billion of prepaid and customer-supplied hardware in its large AI contracts reduces the capital it has to raise itself.

Its flagship is the Stargate campus in Abilene, Texas, developed with Crusoe for OpenAI. In March 2026 Oracle said two of eight buildings were fully operational and the rest were progressing, and that leasing for the full 4.5 GW it agreed to build for OpenAI was complete. A planned 600 MW expansion near Abilene was dropped. Oracle’s heavy reliance on a single customer is central to the AI data center bubble debate.

Which hyperscaler has the most physical capacity?

No hyperscaler publishes a total megawatt figure, so rankings rely on third-party estimates. Most estimates put AWS first, Microsoft second and Google third worldwide, with Meta fourth. Oracle is smaller but adding capacity faster than anyone.

Estimates vary widely depending on method. ABI Research, measuring active US IT load in 2025, put AWS at about 2.3 GW, Meta at 1.5 GW, Microsoft at 1.2 GW, Google at about 0.5 GW and Oracle at about 0.47 GW. ABI counted 17.2 GW of active IT load across 2,396 operational US data centers. A Lambda Finance compilation of global operating IT capacity as of Q1 2026 estimated AWS at about 6.2 GW, Microsoft 5.4 GW, Google 4.7 GW and Meta 3.1 GW. Both rankings are third-party estimates rather than company disclosures, and leased capacity is hard to attribute.

The directional message matters more than the exact numbers. The hyperscalers are each adding gigawatts a year, and their pipelines dwarf their installed base. Microsoft alone says it added about 3 GW over three quarters, and Oracle delivered 850 MW in a single quarter. For context on global totals, see how many data centers there are in the world. For the grid implications, see why power is the new bottleneck.

What are the main risks to the hyperscaler build-out?

The main risks are power availability, rising component costs, customer concentration in a few AI labs, and the chance that AI revenue grows more slowly than infrastructure spending. Local opposition and regulation are growing constraints too.

Power is the hardest limit. Every flagship campus above depends on new generation and transmission. That is why Amazon, Microsoft and Oracle are backing on-site gas, fuel cells and long-dated nuclear contracts. Component inflation is the second pressure: Microsoft, Amazon and Meta all raised budgets partly because of higher component and memory prices.

Concentration risk is real but uneven. Microsoft said the roughly $50 billion jump in its backlog in the June quarter came from customers other than the leading AI model makers, which suggests broader demand. Oracle and CoreWeave, by contrast, depend heavily on OpenAI and a few other labs. Finally, cash flow is tightening. Meta and Oracle both reported negative free cash flow in recent quarters, and Microsoft has turned to the bond market. Bulls argue backlogs of $578–677 billion justify the spending. Bears counter that backlog can be renegotiated and that depreciation will weigh on margins for years.

Frequently Asked Questions