Capital & Industrial Strategy

Three interlocking moves this week redefine OpenAI's capital and commercial architecture. First, according to reporting by The Information via Reuters, OpenAI and Microsoft have agreed to cap their revenue-sharing arrangement at $38 billion — a ceiling that limits Microsoft's upside in exchange for, presumably, structural concessions on compute access or equity terms as OpenAI converts to a public benefit corporation. Second, Bloomberg reports Microsoft had originally targeted a $92 billion return on its early investment, giving context to why a renegotiated cap was necessary: at OpenAI's current scale trajectory, the original arrangement would have been prohibitively expensive for the startup. Third, OpenAI has launched a dedicated enterprise unit — the OpenAI Development Company — as a partnership with 19 investment and consultancy firms, backed by $4 billion and valued at $14 billion, according to Axios and Reuters.

The enterprise unit — structured as a majority-controlled OpenAI entity rather than an independent venture — is a direct bid to own the systems integration and consulting layer that firms like Accenture and Deloitte currently occupy. OpenAI's revenue chief told CNBC that enterprise adoption is 'at a tipping point,' and the consulting arm is designed to accelerate that by removing the implementation friction that has slowed Fortune 500 deployments. Collectively, these moves represent OpenAI vertically integrating downward into professional services while simultaneously reducing its structural dependence on Microsoft — a significant strategic repositioning ahead of any eventual public offering.

Cerebras Systems has increased its IPO price range by roughly one-third, now targeting up to $4.8 billion in proceeds, with pricing expected as early as this week, according to Bloomberg and CNBC. The upsizing reflects strong institutional demand and represents the most significant public markets test for an AI chip company outside Nvidia's ecosystem. Cerebras competes on wafer-scale chip architecture, targeting inference workloads where its design offers latency advantages over Nvidia's multi-chip GPU clusters.

The IPO's success or failure will calibrate public market appetite for AI infrastructure equity at a moment when the sector's private valuations are extreme. Cerebras's go-public timing also coincides with TSMC facing a deepening supply squeeze — The Wall Street Journal reports that AI's next phase, particularly inference scaling, structurally advantages TSMC given its monopoly on leading-edge packaging. Any chip company going public now is implicitly also a bet on sustained TSMC capacity allocation, making supply chain positioning a key investor diligence question.

SoftBank founder Masayoshi Son is in active discussions with President Macron to announce a major AI data center project in France, according to Bloomberg. The deal is framed around a planned joint announcement in the coming weeks — it is an announced intention, not a closed deal with committed capital. Separately, The Wall Street Journal reports SoftBank's telecoms arm has launched a battery storage joint venture in Japan with Cosmos Lab and DeltaX, a move that addresses the power reliability constraint that is binding data center expansion in Japan and increasingly across Asia.

These two moves together indicate SoftBank is positioning as an integrated AI infrastructure developer — not merely a financial investor — capable of combining compute, power, and sovereign government relationships. The French initiative follows a pattern established by Microsoft, Google, and Amazon of using large sovereign commitments to secure regulatory goodwill and preferential procurement positioning in European markets. For SoftBank, whose Vision Fund model has faced sustained pressure, owning hard infrastructure rather than venture equity represents a meaningful strategic shift in how it generates and captures AI-era returns.

A Wired analysis makes the case that Nvidia's durable competitive advantage is not its GPU hardware but its CUDA software stack — the developer tooling, libraries, and runtime environment that makes Nvidia silicon the path of least resistance for AI workloads. This framing matters for capital allocation: it implies Nvidia's moat is not vulnerable to hardware commoditisation in the way pure fabless chip companies are, because switching costs are embedded in developer workflows and existing model training infrastructure rather than in silicon specs alone.

Against this backdrop, Reuters reports that Alphabet and Amazon are tapping overseas debt markets — likely Euro-denominated bonds — to fund AI infrastructure capital expenditure. This is strategically significant: both companies have substantial domestic cash generation, so the choice to issue foreign-currency debt suggests they are either managing FX exposure on overseas data center buildouts or optimising cost of capital across their treasury operations. Either way, the scale of AI infrastructure investment now requires active balance sheet management at the largest technology companies, not just operating cash flow.

A senior South Korean policymaker has publicly proposed paying citizens a dividend funded by taxes on AI profits, a proposal that triggered sharp swings in Samsung Electronics and SK Hynix shares, according to Bloomberg. The proposal remains a policy float, not legislation — there are no confirmed tax rates, eligibility criteria, or legislative timeline. Its significance is political: it is the first serious public articulation by a developed economy's government that AI semiconductor profits represent a national resource subject to redistributive claims.

Franklin Templeton's Christy Tan, cited by Bloomberg, framed this as Asian economies signalling 'shared ownership in the digitalization future' — a reading that positions the proposal as strategic communication rather than immediate fiscal policy. For investors in Korean semiconductor equities, the near-term risk is multiple compression if the proposal gains legislative traction; the medium-term risk is capital allocation uncertainty for Samsung and SK Hynix if reinvestment decisions become subject to profit-sharing constraints.