OpenAI × Microsoft: Unlocking fresh capital without losing the keys

OpenAI is renegotiating its 2019 partnership with Microsoft so it can pivot from a capped-profit LLC to a Public Benefit Corporation (PBC) and, ultimately, float. In return, Microsoft wants longer-dated access to frontier models beyond the current 2030 cut-off. The discussion sits at the intersection of capital-intensity, governance risk and competitive tension. Below is why the outcome matters – and where I think the trade-offs land.

Why the cap no longer fits

OpenAI’s model-training bill has exploded. Project “Stargate” alone carries a reported US $500 bn capex budget over four years – that is roughly 3× the combined FY-24 capex of Alphabet, Meta and Microsoft. A capped-profit structure (investors’ IRR stops after they earn 100× their money) was always philosophically neat; it is now a straight-jacket when your marginal GPU is priced like a small hydro-plant. Converting to a PBC removes the cap, letting OpenAI tap SoftBank’s rumoured US $40 bn round at a US $260 bn valuation

NBC 5 Dallas-Fort Worth and tee up an IPO exit.

Back-of-the-envelope: at US $12.7 bn forecast FY-25 revenue, that implies a 20.5× forward P/S multiple – rich, but not absurd versus Nvidia (c. 23×) given OpenAI’s 200 %+ CAGR. For investors the wager is that AGI monetisation arrives sooner than multiple compression.

Microsoft’s chessboard

Satya Nadella’s original US $13 bn cheque bought:

• priority access to GPT-series models,

• 49 % of a profit pool (subject to the cap), and

• exclusivity to run those models on Azure until 2030.

OpenAI now wants to reopen the cap table; Microsoft wants to reopen model access. Sources suggest Redmond will surrender a slice of equity in exchange for perpetual or at least 2040-vintage rights to next-gen models. From Microsoft’s lens this is a classic option trade: dilute today, but lock in quasi-perpetual AI tolls inside Microsoft 365, Windows and Copilot.

The UK CMA has already ruled the pairing does not constitute a merger under local competition law, so regulatory risk looks low on this side of the Atlantic – a subtle win for Microsoft headquarters in Reading, not Seattle.

Governance – the soft under-belly

OpenAI pledges to grant the non-profit board a “golden share” and director-nomination rights post-conversion. Critics (Elon Musk, Page Hedley, assorted academics) argue that this is governance theatre: control without cash flow rarely survives first contact with fiduciary duties. Columbia Law’s Dorothy Lund bluntly notes that mission-driven companies “walk a dangerous line once the growth‐capital habit kicks in”. I agree – but pragmatically, mission purity does not pay a US $500 bn AWS bill. My take is that the PBC gives regulators a lever without denying investors economic upside. Expect dual-class, sunset provisions – the Shopify playbook writ AI.

Capital-intensity and competitive moats

With a 20× P/S, the market is already pricing in quasi-monopoly economics. The moat is not algorithms (open-source is sprinting) but:

• Supply-side – control of scarce compute via Stargate and Azure.

• Demand-side – enterprise lock-in through Microsoft’s distribution rails.

• Regulatory – heavy-weight lobbying to shape AI safety standards in ways that favour incumbents.

If Microsoft extracts perpetual access, that moat steepens; if negotiations stumble, OpenAI risks losing not only hyperscale-compute but also half a billion daily active distribution end-points. My base case: a middle path where Microsoft secures rolling five-year renewal options tied to a defined minimum-spend on Azure. That keeps cash flowing into OpenAI while preventing permanent vendor lock-in – a structure private-equity readers will recognise as a staple financing with call rights.

What to watch next

Catalyst	Timing	Directional read-through
Definitive term-sheet with Microsoft	Within Q2 CY-25	Equity % Microsoft concedes vs extension length tells us who blinked
SoftBank + others close US $40 bn round	H2 CY-25	Implied up-round signals IPO window; anything below US $260 bn hints at exuberance cooling
Delaware AG sign-off on PBC conversion	2025-Q3	Conditional approval likely; onerous reporting covenants would raise compliance cost
S-1 filing (NYSE/Nasdaq/LSE?)	2026 earliest	London is lobbying hard but dual-listing seems probable; watch FCA reforms on PBCs

MY VIEW

• Strategic fit: Microsoft cannot walk away – its product road-map is now twinned with GPT. That asymmetry hands OpenAI negotiating leverage.

• Valuation: 20× sales looks lofty, but if management believe AGI unlocks non-linear revenue (think synthetic labour markets) then DCF tails fatten fast.

• Risk: Governance drift is real. The bigger threat, however, is technological obsolescence if a rival open-source model matches GPT-5 at one-tenth the cost-to-serve.

• Trade idea: For public-markets investors, a Microsoft overweight is still the cleaner proxy – you get AI upside plus diversified FCF, without single-asset risk. For private-capital allocators, any OpenAI pre-IPO allocation must be underwritten with a low-teens IRR hurdle and exit flexibility given mounting regulatory fog.

In short, these negotiations matter less as a will-they-won’t-they drama and more as a bellwether for how capital markets will price the next wave of AI infrastructure. If OpenAI can square mission, money and models, it will set a template UK unicorns can copy. If it can’t, the lesson will be brutal: in deep-tech, charitable intent is no substitute for deep pockets.