Capital & Industrial Strategy

Three simultaneous moves reveal a coherent strategic pivot at OpenAI. First, an internal memo from OpenAI's new revenue chief — reported by CNBC — explicitly positions the Amazon AWS partnership as a vehicle to reach enterprise clients that Microsoft has, in OpenAI's framing, blocked access to. This is a material escalation: OpenAI is not merely diversifying cloud partnerships but is formally documenting Microsoft as a commercial constraint, which has implications for how enterprise sales teams are being directed and how future licensing structures may be negotiated. The FT reports that some OpenAI investors are already questioning the $852 billion valuation as strategy shifts compound uncertainty about the path to monetisation.

Second, the acquisition of personal finance startup Hiro — confirmed by TechChrunch — indicates OpenAI is building domain-specific financial planning capabilities directly into ChatGPT. This is capability acquisition, not talent acquisition: the target is a product with trained financial reasoning, not just engineers. Third, OpenAI opened its first permanent London office, having previously paused its UK Stargate data center project, per CNBC. The office opening without the compute infrastructure suggests London is being prioritised as a research and policy hub while capital-intensive infrastructure decisions remain contingent on energy and regulatory clarity.

Anthropic has hired Ballard Partners, a lobbying firm with direct ties to the Trump administration, as it contests Pentagon contracts — confirmed by a new public disclosure reported by Bloomberg. Separately, Anthropic co-founder Jack Clark confirmed the company is in direct talks with the Trump administration about its next model, per Reuters. These are not independent events. Anthropic is constructing a Washington presence designed to win government AI procurement — a market that will dwarf commercial enterprise contracts in scale if federal AI adoption accelerates.

The strategic logic is clear: the Trump administration has signalled preference for domestic AI providers with security clearances and cooperative regulatory postures. Anthropic, which built its brand on safety-first positioning, is now translating that into a compliance and trustworthiness narrative for government buyers. The Ballard Partners hire is the operational mechanism; the model talks are the product-level signal. OpenAI's simultaneous attack on Anthropic in internal communications, reported by Axios, suggests both companies see federal procurement as contested territory.

Oracle's confirmed agreement to buy up to 2.8 gigawatts of fuel-cell power from Bloom Energy for AI data centers — reported by Bloomberg — is one of the largest single energy offtake agreements in data center history. The use of fuel-cell technology rather than grid power is significant: it signals that hyperscalers are no longer waiting for grid infrastructure upgrades and are instead contracting directly with alternative energy providers to guarantee compute capacity. This is a structural shift in how AI infrastructure capital is deployed — the investment is moving upstream from GPUs to power generation.

The FT's analysis that the AI build-out is masking the trade impact of Trump tariffs by powering global goods trade in data center components reinforces this picture. Data center construction is generating sustained demand for specialised hardware, cooling systems, and now distributed power generation — creating a secondary capital flow that extends well beyond the frontier model companies themselves. Gulf sovereign wealth funds are also investing heavily in regional AI compute, though Semafor notes that geopolitical instability is testing the durability of those ambitions.

Baillie Gifford's Peter Singlehurst argued on Bloomberg that public market investors have already missed the primary AI growth wave, with the most significant value creation in companies like OpenAI, Anthropic, and SpaceX having occurred while they remained private. This framing matters for capital allocation: if the thesis holds, the coming IPO wave is a liquidity event for early private investors rather than a genuine growth entry point for public market participants. The counterargument — implicit in current valuation levels — is that AI's commercial monetisation is still early, and public market investors are buying into a second, larger phase of growth tied to enterprise deployment rather than model development.

Separately, Vercel's CEO signalled IPO readiness as the company benefits from AI-generated app proliferation, per TechChrunch. Vercel is a structurally interesting IPO candidate because its revenue is a proxy for AI agent deployment volume — the more agents built and hosted, the more infrastructure revenue accrues. Chinese AI startup StepFun is also unwinding its offshore structure to prepare for a domestic IPO, per Reuters — a move driven by Beijing's tightening control over offshore structures for strategic technology companies, not purely commercial IPO logic.