AWS and OpenAI’s $38 Billion Cloud Deal Didn’t Stay $38 Billion. Four Months Later It Was $138 Billion — Plus a $50 Billion Stake for Amazon.
OpenAI’s own blog announced it Nov. 3, 2025, in a single sentence: “OpenAI and AWS have entered a multi-year, $38 billion partnership to scale advanced AI workloads.” Under the deal, OpenAI got immediate access to hundreds of thousands of Nvidia GB200 and GB300 GPUs clustered on new Amazon EC2 UltraServers, with room to scale to tens of millions of CPUs. “Scaling frontier AI requires massive, reliable compute,” said OpenAI CEO Sam Altman. Amazon stock closed at a record high that day, adding roughly $140 billion in market value in a single session.
Four months later, on Feb. 27, 2026, the number changed. As part of a $110 billion funding round that valued OpenAI at $730 billion, Amazon agreed to invest $50 billion directly into OpenAI — and AWS and OpenAI expanded the original compute agreement by another $100 billion, bringing the combined commitment to roughly $138 billion. OpenAI also committed to consuming 2 gigawatts of AWS’s own Trainium chips — its first major bet on hardware that isn’t Nvidia’s.
The arrangement only became possible because Microsoft’s six-year run as OpenAI’s exclusive cloud provider expired Oct. 28, 2025 — days before the first AWS announcement. It nearly came apart again in the spring, when Microsoft disputed AWS’s claim to exclusivity over OpenAI’s new enterprise platform, before the two companies signed a revised agreement in April 2026 that let OpenAI’s products go live on AWS the very next day.
- $38B — the original AWS multi-year compute commitment announced Nov. 3, 2025 · Source: OpenAI, AWS
- $138B — the combined AWS compute commitment after the Feb. 2026 expansion added another $100 billion · Source: Amazon
- $50B — Amazon's direct equity investment into OpenAI — $15B upfront, $35B conditional · Source: aboutamazon.com
- 2 GW — of AWS Trainium3/Trainium4 chip capacity OpenAI has committed to consume · Source: Data Center Dynamics
- $140B — the single-day gain in Amazon's market value on announcement day, Nov. 3, 2025 · Source: CNBC
The core of the Nov. 3, 2025 announcement is infrastructure, not equity. Under the seven-year agreement, OpenAI began running ChatGPT inference and model-training workloads on AWS immediately, tapping clusters of Nvidia’s newest GB200 and GB300 chips housed in Amazon’s new EC2 P6e UltraServers — machines that link up to 72 GPUs over NVLink into a single 360-petaflop unit with 13.4 terabytes of shared memory, according to AWS’s own technical documentation. AWS said its clusters can already support more than 500,000 chips working together, with the full $38 billion in capacity targeted for deployment by the end of 2026 and room to expand into 2027 and beyond.
AWS CEO Matt Garman framed the deal as proof of AWS’s scale: “The breadth and immediate availability of optimized compute demonstrates why AWS is uniquely positioned to support OpenAI’s vast AI workloads.” Amazon CEO Andy Jassy posted a more specific version the same day, writing that the partnership would “provide our industry-leading infrastructure for them to run and scale ChatGPT inference, training, and agentic AI workloads,” adding that it let OpenAI “leverage our unusual experience running large-scale AI infrastructure securely.”
The timing was not an accident. Microsoft’s exclusive hold on OpenAI’s cloud infrastructure — in place since 2019 — formally expired Oct. 28, 2025, less than a week before the AWS deal was announced, and came shortly after OpenAI completed its conversion to a for-profit structure. OpenAI didn’t abandon Microsoft: the same week, it separately committed to purchase an incremental $250 billion of Azure services, per CNBC. AWS was the diversification play, not a replacement.
Investors treated it as a verdict on AWS specifically. Amazon shares rose 4.83 percent to close at a record high on announcement day, part of a two-day, 14 percent run — the stock’s best two-day stretch since November 2022, per CNBC. Not every analyst read it as pure upside. Gartner’s Arun Chandrasekaran told CNBC the deal reflected the “rising capital intensity of AI” and further concentrated compute in the hands of a few hyperscalers, while Deep Analysis founder Alan Pelz-Sharpe was blunter: “For OpenAI, the deal means more access to compute power, though at a hefty price.”
Sam Altman announced the February expansion partly on video — he was in an Oakland courtroom that day for the trial over Elon Musk’s lawsuit against OpenAI. The substance: Amazon’s $50 billion investment (an initial $15 billion tranche, with $35 billion more contingent on conditions the companies haven’t fully disclosed) arrived alongside AWS becoming the exclusive third-party cloud distributor for Frontier, OpenAI’s new enterprise agent platform, and a jointly built “Stateful Runtime Environment” on Amazon Bedrock that lets AI agents retain memory and context across tools. “Combining OpenAI’s models with Amazon’s infrastructure and global reach helps us put powerful AI into the hands of businesses and users at real scale,” Altman said in the official release.
Microsoft's six-year cloud exclusivity over OpenAI formally expires under renegotiated terms.
AWS and OpenAI announce the $38 billion, seven-year compute deal. Amazon stock closes at a record high.
AWS launches Trainium3 chips and P6e-GB300 UltraServers at re:Invent.
OpenAI closes a $110B round at a $730B valuation. AWS-OpenAI compute deal expands by $100B; Amazon invests $50B.
Microsoft disputes AWS's exclusivity claim over OpenAI's Frontier platform; reportedly weighs legal action.
OpenAI and Microsoft sign a revised agreement: nonexclusive license through 2032, capped revenue share.
OpenAI's models, Codex, and Bedrock Managed Agents go live on AWS.
European Commission reaches a preliminary position that AWS and Azure should be designated DMA “gatekeepers.”
The most consequential line item is the 2-gigawatt Trainium commitment — OpenAI’s first large-scale bet on chips that aren’t Nvidia’s, covering both the current Trainium3 and the Trainium4 chip due in 2027. Gartner’s Jason Wong called the arrangement significant: “This move clearly puts AWS as a cloud platform back in conversation for AI,” he told CIO Dive, while Forrester’s Lee Sustar linked the Trainium volume to Amazon’s broader effort to build a chip ecosystem that doesn’t depend entirely on Nvidia.
New multi-year, strategic partnership with @OpenAI will provide our industry-leading infrastructure for them to run and scale ChatGPT inference, training, and agentic AI workloads. Allows OpenAI to leverage our unusual experience running large-scale AI infrastructure securely…
The arrangement almost unraveled before it launched. Microsoft disputed AWS’s exclusivity claim over Frontier in March 2026, and the Financial Times reported the company was weighing legal action. The two sides settled it in April: Microsoft’s license became nonexclusive through 2032 (replacing an open-ended clause tied to OpenAI declaring artificial general intelligence), OpenAI could sell products on any cloud, and Microsoft stopped collecting revenue share from OpenAI in exchange for a capped payment schedule through 2030. A day later, OpenAI’s models, its Codex coding tool, and a new Bedrock Managed Agents product went live on AWS. “This is the beginning of a deeper collaboration between AWS and OpenAI,” Amazon said.
Excited about our new strategic partnership with @OpenAI. Developers and companies of all kinds are eager to run services powered by OpenAI models on AWS, and our unique collaboration will provide a stateful runtime environment for them that's powered by OpenAI's frontier…
The AWS deal is one piece of a much larger pattern: OpenAI has now committed to roughly $1.25 trillion in infrastructure spending across seven vendors, and Amazon isn’t even the largest of them. Critics call this web of hyperscalers investing equity in OpenAI on the condition it spends the money back on their own infrastructure a “circular deal” — one that inflates the AI sector’s apparent value without adding real outside capital. OpenAI CFO Sarah Friar has acknowledged the risk directly, warning that if revenue growth doesn’t keep pace, the company could struggle to pay for the contracts it has already signed.
Neither the FTC nor the Justice Department’s Antitrust Division has filed a complaint against the AWS-OpenAI arrangement specifically. But an FTC staff study has already flagged rising switching costs and concentrated compute access as risks in cloud-AI partnerships generally, and DOJ has examined whether vertical integration across compute, capital, and model deployment could let incumbents foreclose competitors, per Mogin Law LLP’s analysis of the regulatory posture.
The European Commission has gone further: it opened formal Digital Markets Act investigations into AWS and Azure in November 2025 and, in June 2026, reached a preliminary position that both should be designated “gatekeepers.” The Commission’s own language ties the finding directly to AI demand: “AI is significantly increasing the demand for cloud-related services, and AWS and Azure appear to retain a large proportion of this increased demand within their respective ecosystems.” AWS and Azure together account for more than 65 percent of EU cloud revenue.
“Anthropic and OpenAI are dangerous and unsustainable companies that shouldn't IPO. The AI bubble is a con and retail investors are the marks. AI doesn't have ROI, it's nothing like AWS/Uber, and it's got no post-bubble recovery story.”
Ed Zitron, Better Offline — on X, June 2026
Went on Bloomberg - Anthropic and OpenAI are dangerous and unsustainable companies that shouldn't IPO. The AI bubble is a con and retail investors are the marks. AI doesn't have ROI, it's nothing like AWS/Uber, and it's got no post-bubble recovery story.
A $38 billion cloud deal became $138 billion in compute plus a $50 billion equity stake in four months, and it’s a fraction of the roughly $1.25 trillion OpenAI has now committed across seven infrastructure vendors. AWS gained a marquee customer, a hedge against Nvidia dependence in Trainium, and exclusive distribution rights on OpenAI’s new agent platform. What no one involved has fully answered is Sarah Friar’s own question: whether OpenAI’s revenue will grow fast enough to pay for what it has already signed.
