Private Docs Show OpenAI Called Microsoft a 'Concentrated Risk'

OpenAI publicly describes its relationship with Microsoft as a strategic partnership. In a private investor document, it describes the same relationship as a concentrated risk.

CNBC reported on March 23 that it had viewed a confidential memorandum circulated to investors in OpenAI's $110 billion funding round, which closed in February 2026. The document reads like an IPO prospectus — risk factors, financial projections, forward-looking statements — but it is not an S-1 filed with the SEC. It is a confidential information memorandum, the kind of document a company shares with sophisticated investors before a major capital raise, and it came with strings: OpenAI declined to comment on its contents.

The Microsoft disclosure is the part that will get attention. According to CNBC, the document states that Microsoft, the software giant that has invested roughly $13 billion in OpenAI since 2019, is responsible for a "substantial portion" of OpenAI's financing and compute capacity. That language — "substantial portion" — is the kind of phrasing lawyers attach to dependencies a company can't quantify without embarrassing itself. It doesn't say 30 percent. It doesn't say 70 percent. It says substantial.

The dependency is structural, not incidental. The $110 billion round that this memo accompanied — reported by CNBC in February 2026 — brought in $50 billion from Amazon, $30 billion from Nvidia, and $30 billion from SoftBank at a $730 billion pre-money valuation. Under the terms of a renegotiated partnership announced in October 2025, Microsoft's Azure cloud remains OpenAI's exclusive host for stateless API traffic until the companies agree that artificial general intelligence has been achieved. OpenAI's version of that announcement confirmed the same terms: Azure exclusivity through the AGI threshold, a $250 billion Azure compute purchase commitment from OpenAI, and an IP and revenue-share arrangement that both sides said was unchanged from the original deal. A joint statement issued in February 2026 after the funding round closed reiterated those terms again — Azure stays exclusive, IP rights unchanged, revenue share unchanged.

The messaging from both companies is that the relationship is healthy and deepening. The investor document offers a more clinical view: if the partnership deteriorates, or if Azure can't deliver at the scale OpenAI needs, the consequences are severe.

That concern about compute scale has real-world precedent. According to Built In's analysis of OpenAI's cloud arrangements, when OpenAI launched GPT-4.5 in February 2025, the company had to stagger the rollout because it had run out of GPUs. That episode motivated OpenAI's subsequent diversification — compute contracts with Oracle, Google Cloud, and CoreWeave — but none of those arrangements replace Azure as the primary infrastructure backbone. The investor document names TSMC, the Taiwan-based semiconductor manufacturer that fabricates most of the world's advanced AI chips, as a separate supply chain risk tied to geopolitical tension in the Taiwan Strait. OpenAI doesn't control that variable either.

The compute spend figures in the document are also worth a second look. CNBC's reporting cites $665 billion in committed infrastructure spend through 2030. In February 2026, CNBC had separately reported that OpenAI had revised its compute ambitions downward, from a prior target of $1.4 trillion to approximately $600 billion by 2030 — the same story that put OpenAI's 2025 revenue at $13.1 billion and weekly active users at 900 million. The investor document's $665 billion figure is about $65 billion higher than that public estimate. The discrepancy isn't enormous at this scale, but it means the document is presenting a more aggressive spending commitment than OpenAI was signaling publicly a month earlier. That's worth flagging.

The litigation section of the document names at least 14 user harm lawsuits and identifies Elon Musk's ongoing case against OpenAI as a material risk. TechCrunch reported in January 2026 that the Musk suit will go to trial — U.S. District Judge Yvonne Gonzalez Rogers found sufficient evidence that OpenAI leaders made assurances about maintaining the nonprofit structure to let the case proceed to a jury. That creates genuine uncertainty for the company at exactly the moment it's trying to complete a conversion from nonprofit to public benefit corporation.

Speaking of which: the investor document apparently flags the PBC structure itself as an unusual governance form that may constrain what OpenAI can do. That's a company telling its own investors that its legal structure is a risk factor. The February 2026 joint statement from OpenAI and Microsoft didn't address governance structure at all.

One notable absence in the risk factors: Sam Altman, OpenAI's chief executive, who was fired by the company's board in November 2023 and reinstated five days later after a near-total employee revolt. Key-person risk is standard S-1 boilerplate. The fact that the investor document omits Altman from the risk factors suggests the company either believes the November 2023 episode is resolved, or decided that naming the CEO as a risk in a private fundraising document was bad optics. When the actual S-1 lands — assuming it does — that language will need to appear.

NBC News reported on one of the user harm lawsuits in detail: the case filed by Matt and Maria Raine in August 2025, alleging that ChatGPT contributed to the death of their teenage son Adam. OpenAI denied responsibility. The investor document treats cases like this as a category — 14 lawsuits, not a list of names — which is how you present litigation risk to investors without making it feel specific.

For investors trying to size the Microsoft dependency, the practical question is what "substantial portion" means in dollar terms. OpenAI's 2025 revenue was $13.1 billion. The company is spending far more than that on compute. If Microsoft is providing a substantial portion of that compute through Azure credits, preferred pricing, or direct subsidy, the financial relationship is less a partnership and more a condition of existence. That's not necessarily alarming — plenty of large technology companies are structurally dependent on cloud providers — but it's exactly the kind of risk that needs to be quantified before a public offering, not gestured at with vague language in a private memo.

The document is a preview of the S-1, not the S-1 itself. The SEC will want numbers.