OpenAI and Anthropic Capex Dependence: Stocks at Risk
OpenAI and Anthropic compute commitments by partner
| Partner | OpenAI | Anthropic | Source |
|---|---|---|---|
| Microsoft (Azure) | Additional $250 billion of Azure services agreed in the October 2025 restructuring; Microsoft is also OpenAI's largest outside shareholder | Anthropic agreed to buy $30 billion of Azure capacity (November 2025); Microsoft agreed to invest up to $5 billion | Company announcements |
| Oracle (OCI) | Reported contract of about $300 billion over roughly five years, part of the Stargate build-out | No major deal disclosed | Press reports; Oracle backlog disclosures |
| Amazon (AWS) | $38 billion over seven years for NVIDIA GPU capacity (November 2025) | Primary training partner; Project Rainier Trainium cluster; Amazon has invested $8 billion | Company announcements |
| Google Cloud | Some capacity used, no large commitment disclosed | Access to up to one million TPUs, a deal valued in the tens of billions of dollars (October 2025) | Company announcements |
| CoreWeave | Contracts totaling about $22.4 billion | Multi-year agreement to support Claude models; value not disclosed | CoreWeave filings |
| NVIDIA, AMD, Broadcom (chips) | NVIDIA letter of intent for at least 10 GW with up to $100 billion of investment; AMD 6 GW deal with share warrants; Broadcom 10 GW of custom chips | NVIDIA agreed to invest up to $10 billion alongside the Microsoft deal | Company announcements |
OpenAI and Anthropic Capex Dependence: Who Is Building for Them and the Stocks at Risk
OpenAI and Anthropic own very few data centers, yet a large share of this year's AI infrastructure spending is being built to serve them. This page lists the compute commitments each lab has signed, explains why their spending runs far ahead of their revenue, and names the listed companies whose backlogs and balance sheets depend most on those two customers.
OpenAI and Anthropic capex dependence in short
OpenAI and Anthropic do most of their spending through other companies. They sign multi-year contracts to rent computing capacity, and the cloud providers, chip makers and data-center builders then spend their own money, or borrow, to build it. So when people talk about OpenAI capex, most of the steel, chips and power is actually on the balance sheets of Microsoft, Oracle, Amazon, Google, CoreWeave and their lenders.
The dependence runs both ways. The labs depend on these partners for the capacity to train and serve their models. The partners depend on the labs to fill what they build: in late 2025 OpenAI's chief executive put the company's compute commitments at about $1.4 trillion over roughly eight years, and Anthropic signed capacity deals with Google, Microsoft and Amazon in the tens of billions of dollars.
In 2026 the concentration became visible in the partners' own numbers. Oracle reported $664 billion of contracted backlog at the end of August, and analysts estimate that roughly half of it is tied to OpenAI, although Oracle does not break the figure down by customer. CoreWeave reported about $104 billion of backlog, with its three largest customers making up 72% of second-quarter revenue. In July S&P lowered Oracle's credit rating to BBB-, one step above junk, and named OpenAI as a key risk.
OpenAI capex vs revenue: why the gap matters for hyperscalers
Both labs are growing revenue very fast, but their contracted compute spending is far larger than their current sales, and neither is expected to produce positive free cash flow for several years. The difference has to be covered by new funding rounds, a public listing, or by partners carrying the cost on their own books until the labs can pay.
That is why the risk sits with the builders. A cloud provider that signs a long rental contract usually takes on the land, power, chips and leases first, and gets paid over the following years. If a lab slows its plans, renegotiates or cannot raise money, the builder can be left with capacity it financed and cannot easily re-rent. Ratings agencies have flagged this for Oracle, and estimates for total big-tech capex in 2026 now range from about $600 billion to more than $800 billion, with a growing part paid for with debt.
It does not mean the spending is wasted. Demand for AI capacity has outrun supply through 2026, and both labs have other ways to raise cash, including Anthropic's planned listing. What it means is that a few listed companies have become, in effect, lenders to two private customers, and their share prices now move with news about those customers' funding.
Which stocks depend most on OpenAI and Anthropic spending
| Company | How it depends on the two labs | What to watch next |
|---|---|---|
| Oracle (ORCL) | Largest backlog exposure; estimates put about half of its $664 billion backlog on OpenAI | Backlog by customer, debt raises, credit-rating outlook |
| CoreWeave (CRWV) | OpenAI and Microsoft are major customers; three customers made 72% of Q2 2026 revenue | Customer mix in each quarterly report, financing costs |
| Microsoft (MSFT) | Shareholder in OpenAI and supplier to both labs through Azure | Azure growth, capex guidance, OpenAI stake value |
| NVIDIA (NVDA) | Sells the chips behind most lab capacity and has agreed to invest in both labs | Data-center revenue concentration, timing of lab investments |
| Alphabet (GOOGL) | Anthropic shareholder and TPU supplier; also a direct rival to both labs | Google Cloud backlog, TPU external sales |
| Amazon (AMZN) | Anthropic's largest shareholder and main training cloud; also an OpenAI supplier | AWS backlog, Anthropic stake value at listing |
| Broadcom (AVGO) | Building custom AI chips for OpenAI and supplying TPU silicon used by Anthropic | Custom-chip order timing, customer count |
Data in practice: tracking the companies that build AI capacity
The build-out shows up in public records before it shows up in earnings: data-center permits, federal contract awards, lobbying filings, insider trades and congressional trade disclosures. Gemral Edge Feed 1 links these records to the listed companies involved, and every record is tied to the company and ticker that were valid on the day it became public.
Publication lags are applied, so the data can be joined without look-ahead, and each record type ships only after a sampled accuracy check against the original public documents (95% minimum). Data comes only from public sources. Coverage and methods are documented at /edge/data-coverage and /edge/methodology.
OpenAI and Anthropic capex: what to be careful about
Several figures on this page come from press reports and analyst estimates rather than company filings. Oracle and CoreWeave do not name the customers behind their backlog, so any share attributed to OpenAI is an estimate. Contract values are usually maximums over many years and can be cut or delayed.
We found no filing that ties a specific share move to a single lab contract. Treat the company list above as exposure context, not a forecast. This page reflects public information as of November 12, 2026, is for information only and is not investment advice.
Related research
| Page | What it covers |
|---|---|
| CoreWeave Q3 2026 earnings and backlog | CoreWeave's backlog, guidance and customer concentration. |
| Oracle Cloud and the Stargate supercomputer deal | Oracle's contract with OpenAI and what it means for its balance sheet. |
| OpenAI Stargate 5 GW data centers and power | The data centers and power deals behind OpenAI's compute plans. |
| Claude Fable 5.2 and the Anthropic IPO | Anthropic's latest model, pricing and the listed stocks tied to its IPO. |
| Private credit and shadow banking risk | The private lenders now financing a growing share of data-center building. |
| Bittensor TAO and decentralized AI compute (crypto) | The crypto side of AI compute: a token market built around model and compute supply. |
| Crypto Scanner chart patterns | Scan crypto charts for breakout and reversal patterns. |
Access Real-Time Terminal Intelligence & Quantitative Signals
Unlock instant Telegram alerts, full congressional portfolio archives, and algorithmic catalyst radar.
Upgrade to Gemral Edge Pro ($39/mo)Frequently asked questions
What is OpenAI's capex?
OpenAI owns little infrastructure itself, so most of its spending is contracted compute rather than capex on its own books. In late 2025 its chief executive put total compute commitments at about $1.4 trillion over roughly eight years, spread across Microsoft, Oracle, Amazon, CoreWeave and chip makers.
How much has Anthropic committed to compute?
Anthropic's disclosed deals include up to one million Google TPUs (tens of billions of dollars), $30 billion of Microsoft Azure capacity, Amazon's Project Rainier cluster, and a $50 billion plan to build US data centers announced in November 2025.
Which company depends most on OpenAI?
Oracle has the largest visible exposure. It reported $664 billion of backlog at the end of August 2026, and analysts estimate about half is tied to OpenAI. S&P cut Oracle to BBB- in July 2026 and named OpenAI as a key credit risk.
Is OpenAI capex bigger than its revenue?
Yes, by a wide margin. Its contracted compute spending over the coming years is many times its current annual revenue, and the gap is expected to be funded by new capital and by partners carrying costs until revenue catches up.
What happens to these stocks if OpenAI or Anthropic slows spending?
Builders with the most concentrated backlog, such as Oracle and CoreWeave, would be hit first through lower expected revenue and harder financing. Diversified hyperscalers and chip makers would feel it more slowly. Lab funding rounds and listing news are the signals to watch.
Risk Disclaimer
Trading and investing in digital assets, financial instruments, and predictive events involve substantial risk of loss and are not suitable for every investor. The predictive intelligence, probability distributions, historical precedents, and scenario modeling presented on this page are compiled for informational and research purposes only and do not constitute financial, investment, legal, or tax advice. Past performance and statistical precedents do not guarantee future outcomes. Always conduct independent due diligence before committing capital.