Microsoft Invests US$10 Billion in OpenAI — The Largest AI Tie-Up, and Its Subsequent Strains

On 23 January 2023, Microsoft formally announced an expanded investment in OpenAI. The amount was not disclosed, but reporting cited a multi-year, multi-stage commitment in the order of US$10 billion. Combined with the 2019 investment of US$1 billion and a 2021 follow-on, it became the largest AI-related investment in industry history.

It came just 54 days after the public launch of ChatGPT on 30 November 2022.

A Three-Pillar Deal

The press release was short. The actual agreement was a three-layer structure.

1. Capital and profit-share. Microsoft invested in OpenAI's for-profit subsidiary (OpenAI Global, LLC) and would receive distributions under the company's "capped profit" structure. Reporting indicated that, until its investment was recouped, Microsoft would take up to 75% of OpenAI's profits, with a further share (reported as 49%) retained thereafter.

2. Azure exclusivity. All of OpenAI's training and inference workloads would run on Microsoft Azure—an exclusive arrangement. Azure data centres stacked with NVIDIA H100 / H200 GPUs became the physical substrate for GPT-4, GPT-4 Turbo, and the o1 line.

3. Product integration. GPT-4 would be deployed laterally across Microsoft's flagship products. The "new Bing" of February 2023 (Bing Chat, later renamed Copilot) was the first; Microsoft 365 Copilot, GitHub Copilot X, Windows Copilot, and Dynamics 365 Copilot all followed within the same year.

Why Microsoft Bet So Heavily

Since becoming CEO in 2014, Satya Nadella had reshaped Microsoft into a "cloud and subscriptions" company. A third pillar—AI—was now to be acquired not by in-house build but through capital partnership.

Google had DeepMind; Meta had FAIR; Amazon had Alexa AI. Microsoft Research existed but had not reached the frontier of large language models. By effectively bringing OpenAI inside its perimeter, Microsoft put itself overnight on par with—or ahead of—Google.

At the Bing Chat launch in February 2023, Nadella declared he wanted to "make Google dance". After twenty years as a distant second in search, Microsoft was going on the offensive, weaponised by generative AI.

The November 2023 Boardroom Coup

Microsoft's control was, however, not absolute—a fact the world learned on Friday evening, 17 November 2023.

OpenAI's nonprofit board abruptly dismissed CEO Sam Altman. Microsoft, reportedly, was given minutes of notice. The governance of a strategic partner Microsoft had wagered ten billion dollars on was operating under a structure in which Microsoft had no formal say.

Nadella spent the weekend manoeuvring, eventually announcing that Microsoft would hire Altman and Greg Brockman directly. Within forty-eight hours, 95% of OpenAI's staff had signed a letter threatening to follow them to Microsoft. By 22 November, Altman was reinstated as CEO.

The coup was contained, but the lesson for Microsoft was unambiguous: "we do not own OpenAI."

Regulators Move In

US FTC. In January 2024, the Federal Trade Commission issued 6(b) orders to all major firms involved in generative-AI partnerships, including Microsoft and OpenAI. In November 2024 this escalated into a formal civil investigative demand (CID) seeking Microsoft's data from 2016 through 2025. The focus: whether the structure had been engineered to evade merger review while creating effective control.

UK CMA. Beginning December 2023, the Competition and Markets Authority examined "whether Microsoft exercises material influence over OpenAI". After multiple reviews through 2024 and 2025, it concluded the arrangement did not constitute a relevant merger situation.

European Commission. The EU competition directorate ran a parallel preliminary inquiry and pushed for modifications to the integration agreement.

January 2025 — The End of Hard Exclusivity

On 21 January 2025—the same day the Trump White House announced the Stargate Project (OpenAI + Oracle + SoftBank, up to US$500 billion over four years)—Microsoft and OpenAI publicly revised their agreement.

The key changes:

• Azure exclusivity was downgraded to a right of first refusal. OpenAI could now use other clouds when Azure could not supply capacity.
• OpenAI began partnerships with Oracle, SoftBank, and later AWS.
• Microsoft retained access to OpenAI's technology, but no longer the monopoly.

In effect, Microsoft retreated from "owning the OpenAI relationship" to "being one of OpenAI's largest shareholders and customers". By April 2026, the original Azure exclusivity expired entirely, and OpenAI began offering its models on AWS Bedrock.

The Frenemy Era

From 2025 onward, "frenemy" became the standard descriptor for the relationship. Microsoft pivoted Copilot to incorporate not just OpenAI's GPT line but its own in-house MAI-1 small models, Anthropic's Claude, and open-weights models in parallel. OpenAI, for its part, built up an independent enterprise sales force and targeted growth that did not depend on Microsoft's customer base.

The ten-billion-dollar wager bought Microsoft AI leadership in 2023–2024, but failed to "enclose" OpenAI. That is the present-day verdict on the partnership.

What It Demonstrated

The Microsoft–OpenAI deal pioneered a new M&A pattern: a hyperscaler absorbing a fast-growing AI startup through capital rather than outright acquisition. Yet it also showed how a structure built on a nonprofit board, capped profits, and exclusivity clauses can create governance instability that conventional M&A would never generate.

The biggest beneficiary of the post-ChatGPT AI boom was not OpenAI itself but Microsoft's market capitalisation—roughly doubling from early 2023 to mid-2024, briefly overtaking Apple to become the world's most valuable company. By that measure, US$10 billion was a vanishingly small price.

And from that deal flowed Stargate, the regulatory siege, the end of hard exclusivity—the geopolitics of the generative-AI industry that begins here.