Satya Nadella’s Multibillion-Dollar Question: How Much OpenAI Is Too Much?

I have bought a few things in my time that nobody could explain to me, and I have usually managed to explain them to myself afterward, which is not the same as being right. When Microsoft began investing in OpenAI, the arrangement puzzled almost everyone, including a good many men who were paid to be puzzled professionally. A company whose business was selling software licences was putting billions into a research lab with a non-profit parent, no revenue, and a stated mission that had nothing to do with selling software.

A decade later the logic looks considerably clearer. Microsoft bought itself a position at the centre of the most important technology transition in a generation, and — this is the part that earns the hat-tip — it did so without having to build a frontier lab from scratch.

What the partnership actually buys

• Preferred access to models that Microsoft can embed across its entire product line.
• A revenue share on the models it resells through its cloud.
• A seat at the table in the field’s most consequential decisions.
• Recruiting appeal: engineers join to work near the frontier.

The uncomfortable parts

The trouble with investing in a partner rather than owning it is that a partner is a separate animal with legs of its own, and it may turn around and walk in a direction you did not choose. When OpenAI’s governance crisis unfolded, Microsoft found itself with enormous exposure and no formal control — a fine position for a spectator, a poor one for a passenger who has already paid for the trip. The subsequent restructuring of the relationship — more equity, a clearer commercial framework, a seat on the board — was a response to exactly that experience.

There is also the concentration question. Microsoft’s cloud revenue growth is now substantially tied to AI workloads, which are themselves substantially tied to one partner’s models. Investors have started asking how much of the company’s valuation rests on a relationship it does not fully control.

Nadella’s framing

Nadella has consistently described the strategy as platform-building rather than product-building: own the layers below and above, partner where someone else is faster, and make sure every one of your products gets better because of it. It is the same playbook he used for open source and for cloud, and it has a good track record.

The open question is whether the platform logic survives contact with a partner that increasingly wants to be a platform itself. That is the tension every large company in this field is now managing, and Microsoft is simply the furthest along.

What to watch

Whether Microsoft’s internal model effort closes the gap with its partner; whether the commercial terms survive the next renegotiation; and whether regulators in Europe and the United States decide that the relationship is close enough to count as a merger in all but name.

The infrastructure bill

The partnership’s cost is not only the investment. Microsoft has committed to building and operating an enormous amount of capacity to serve the resulting workloads, which converts an equity bet into a decade-long capital programme with its own risks.

If demand holds, the infrastructure is a moat: nobody else can serve that volume on short notice. If demand disappoints, the company is left with facilities justified by a forecast nobody can now verify — and I have seen a good many warehouses built on a forecast.

The talent question

A subtler strategic problem is that the company’s own research organisation has to compete for people against a partner that can offer frontier work and startup equity. Nadella has addressed this by funding internal model development and by building a research organisation with its own distinct identity.

Whether that produces a genuine second capability or a permanent junior partner is the question that determines how much leverage Microsoft has at the next negotiation. On that bench rests the price of the next handshake.