a confusing and obscure compute market is perfect for those who benefit from buyers not knowing what theyβre getting. brokers make money because buyers canβt easily check quality on their own.
so if you run a serious neocloud, you should urgently want more transparency. if quality were easier to measure and prove, it will open up opportunities that, today, go to the same providers over and over again
there are only a handful of buyers that truly change where compute money goes (@amazon, @Google, @Meta, @Microsoft, @Oracle). while the number of “neoclouds” keeps growing, all of them combined donβt come close to the 71% of the world’s compute those 5 companies own
and when those 5 companies need more gpus than they can build themselves, they rent from the same, short list of providers (@CoreWeave, @nebiusai, @nscale, @IREN_Ltd, and @LambdaAPI).
plenty of smaller neoclouds say their quality is just as good as the “premium” companies, yet can’t seem to get customers if they charge just as much
people are excited about watching price trends for gpus. for the long tail of neoclouds that are only going to grow, they care less about these price trends. what they want is a chance to compete for the large contracts that keep going to the same few names
The poster’s X post of 2 November 2026, which says five companies own 71% of the world’s compute; the argument is the poster’s opinion.
Amazon, Google, Meta, Microsoft and Oracle held an estimated 71% of the world’s AI compute at the end of 2025, measured in H100-equivalents, up from 63% in the first quarter of 2024, Epoch AI data show.
Google alone held 24.8%, more than Microsoft’s 16.8% and Oracle’s 5.6% combined, by our arithmetic from Epoch’s table. Everyone else in the world shared the remaining 29.3%.
The key insight: Five companies, Amazon, Google, Meta, Microsoft and Oracle, held an estimated 71% of the world’s AI compute at the end of 2025, up from 63% in early 2024, and Google alone held 24.8%, Epoch AI estimates.
What the Epoch Data Show
Epoch, a research group that tracks AI chips, estimates each company’s share of cumulative global AI compute every quarter. At the end of 2025, Google held 24.8%, Microsoft 16.8%, Amazon 12.1%, Meta 11.3% and Oracle 5.6%, according to Epoch.
The five held about 1.5 million H100-equivalents in the first quarter of 2024 and about 14 million by the fourth quarter of 2025, Epoch’s table shows. That is roughly nine times as much in seven quarters, by our arithmetic. Google’s holdings went from about 500,000 to about 5 million, ten times as much.
Microsoft’s share slipped from 18.9% to 16.8% over the period, even as its holdings rose from about 450,000 to about 3.4 million H100-equivalents, the table shows. Amazon’s share rose from 8.4% to 12.1%.
The combined share stopped rising in 2025: it was 71.1% in the second quarter, 71.3% in the third and 70.7% in the fourth, according to Epoch.
Share of global AI compute (H100-equivalents) held by Google, Microsoft, Amazon, Meta and Oracle, Q1 2024 to Q4 2025, per Epoch AI (14 Apr 2026). Combined: 62.5% to 70.7%. The 29.3% rest-of-world share is our arithmetic.
How Epoch Measured It
The figures come from Epoch’s AI Chip Owners dataset, which covers chips made by Nvidia, Google, AMD, Huawei and Amazon, Epoch says. An H100-equivalent converts each chip by its highest 8-bit operations per second, so it does not perfectly reflect real-world performance across chip types, Epoch notes.
Epoch says its data does not capture all global chip ownership, and that first-quarter 2026 data was excluded because it was not yet finalised when it published the insight on 14 April. Many AI labs, including OpenAI and Anthropic, largely depend on these hyperscalers for research and inference compute, Epoch writes.
The Argument on X
On 2 November, Jessie Dong, whose X profile lists compute and AI, used the 71% figure to argue about who wins compute contracts. Dong wrote that “there are only a handful of buyers that truly change where compute money goes” and named Amazon, Google, Meta, Microsoft and Oracle.
When those five need more GPUs than they can build themselves, they rent from a short list of providers, Dong wrote: CoreWeave, Nebius, Nscale, IREN and Lambda. The post says smaller neoclouds claim similar quality but struggle to win customers at the same price, and that more transparency on quality would help them compete for large contracts.
Dong’s point is an opinion. The Epoch figure the post cites is an estimate of chips owned, not of contracts signed.
As we read the table, the concentration grew while the market grew: the five’s holdings rose about ninefold in seven quarters by our arithmetic, and their combined share still edged up.
As we read it, the flat share in the second half of 2025 (71.1%, 71.3%, 70.7%) means everyone else, neoclouds and labs included, grew about as fast as the five over those quarters, which is the backdrop to Jessie Dong’s argument about who wins the large contracts.
Epoch AI, 14 April 2026
“Many AI labs (including OpenAI and Anthropic) largely depend on these hyperscalers for access to R&D and inference compute.”
Three Implications
FOR NEOCLOUDS Jessie Dong argues that more transparency on quality would let smaller providers compete for large contracts that, in Dong’s view, go to a short list today.
FOR AI LABS Epoch writes that many labs, including OpenAI and Anthropic, largely depend on the five hyperscalers for research and inference compute.
FOR READERS OF THE NUMBERS Epoch counts chips owned in H100-equivalents and says its data does not capture all global ownership; 2026 figures were not finalised when it published.
Business Pill · Neoclouds Explained
Business Pill: what a neocloud is and how it differs from a hyperscaler.
What Is Not Established
Epoch’s figure counts chips owned, not chips in use or capacity rented out, and it does not capture all global ownership, by Epoch’s own account. Epoch had not finalised data for 2026 when it published, so the shares for this year are not known. Dong’s list of five preferred providers is an opinion; we have not seen contract-level data that ranks them.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Amazon, Google, Meta, Microsoft and Oracle held an estimated 71% of global AI compute at the end of 2025, up from 63% in early 2024, Epoch AI estimates, with Google alone at 24.8%.
A note on sourcing. Shares and H100-equivalent counts are from Epoch AI’s data insight of 14 April 2026 and its quarterly table; the multiples, the 29.3% rest-of-world share and the Google versus Microsoft-plus-Oracle comparison are our arithmetic. Jessie Dong’s argument is from a post on X of 2 November 2026 and is an opinion.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
