Capital & Industrial Strategy

OpenAI is approaching its anticipated IPO with annualised revenue exceeding $40 billion, roughly doubling its trajectory from late 2025, driven by ChatGPT subscriber growth, enterprise contract expansion, and its Codex coding assistant gaining commercial traction, according to Bloomberg and Semafor. The more strategically significant data point came from CFO Sarah Friar's investor communication: enterprise revenue has now surpassed consumer, a threshold that transforms OpenAI's investor pitch from a consumer-app story into a durable B2B revenue model less vulnerable to churn cycles and platform risk.

Against that backdrop, the sudden departure of revenue chief Denise Dresser — coming during a week of broader C-suite turbulence — is being flagged by analysts as a material IPO risk, per CNBC. The concern is not one departure in isolation but the pattern: repeated senior exits signal cultural or governance dysfunction that pre-IPO institutional investors will demand explanations for. The compute cost overhang remains structural — OpenAI's gross margins are constrained by inference infrastructure spend, and that tension between top-line acceleration and bottom-line pressure will be the central question when the S-1 lands.

Anthropic reported to prospective investors that Q2 2026 revenue grew at least 14-fold year-on-year, per Bloomberg, and is anchoring its IPO valuation to a $190–200 billion 2028 revenue forecast, according to Reuters. That forecast implies a valuation multiple anchoring exercise — by setting a 2028 revenue target, Anthropic is inviting investors to back-solve a current valuation using forward multiples, a technique that requires sustained hypergrowth and leaves little margin for execution slippage. Simultaneously, Anthropic is reportedly in talks to acquire data startup Decart for approximately $6 billion, according to Bloomberg/Glasswing Ventures commentary. The strategic rationale is data quality: as Rudina Seseri of Glasswing Ventures noted, frontier AI companies are increasingly bottlenecked not by model architecture but by proprietary training data, and a Decart acquisition would be an attempt to build a defensible data asset ahead of listing.

The Decart deal, if confirmed, would be a significant pre-IPO capital allocation signal — spending $6 billion on a data acquisition while simultaneously pitching growth forecasts to IPO investors suggests Anthropic believes its current trajectory requires inorganic reinforcement. This deal remains reported as in-talks only; no confirmed terms or regulatory filings have been disclosed.

Nvidia disclosed in its Q2 filing a $21 billion stake in SpaceX — acquired via its investment in xAI — and a $30 billion stake in Intel, per Bloomberg and CNBC. The SpaceX stake is particularly notable in context: following Elon Musk's announced exclusive arrangement to kit out SpaceX data centres with Nvidia hardware, per the Financial Times, the equity stake functions as a demand-creation mechanism — Nvidia effectively cross-subsidises its chip sales by taking equity in the customer, aligning incentives and locking in long-term compute procurement. The Intel stake is strategically distinct: at $30 billion it represents a significant bet on a competitor-turned-partner, likely reflecting Nvidia's interest in Intel's foundry capacity and packaging technology as it diversifies supply chain risk.

Simultaneously, the Wall Street Journal reported that Nvidia is scaling back its $250 billion funding guarantee for OpenAI's Ohio data centre — a commitment that had attracted investor concern about balance sheet risk exposure. The scaling back suggests Nvidia's board is applying discipline to a strategy that, taken to its logical extreme, would have Nvidia functioning as a quasi-development finance institution for the AI ecosystem. The tension is clear: equity stakes and financing guarantees drive chip demand, but they also concentrate risk and raise questions about whether Nvidia's reported earnings reflect durable customer demand or internally engineered consumption.

The US government is preparing to formally instruct partner nations that they must choose alignment in the AI competition with China, per Reuters. This is a qualitative escalation from existing export control frameworks: rather than restricting specific components, the US is signalling a comprehensive alignment demand that would affect access to American AI models, cloud infrastructure, chip supply chains, and potentially defence-adjacent AI procurement. For countries currently hedging — including in Southeast Asia, the Gulf, and parts of Europe — this forces a binary choice with significant economic consequences either way.

The timing intersects with Taiwan's projected fastest economic growth in four decades driven by AI-related semiconductor demand, per Reuters. Taiwan's economic trajectory is now structurally dependent on AI infrastructure demand — making geopolitical alignment not merely a diplomatic question but an economic survival calculation for Taipei. The US 'pick sides' doctrine, if implemented as reported, would formalise what has been an informal pressure campaign and create explicit market access conditionality.