Capital & Industrial Strategy

The UK government formally launched a £500 million sovereign AI fund aimed at reducing dependence on foreign technology and backing homegrown startups, according to Wired. The announcement was timed alongside news that Anthropic has leased office space in London's Knowledge Quarter at Regent's Place sufficient for approximately 800 staff — four times its current UK headcount of 200 — as reported by Wired and confirmed by CNBC and the WSJ. OpenAI has simultaneously announced its first permanent London office.

The strategic logic is explicit: Anthropic's expansion follows its deteriorating relationship with the Pentagon, and the UK is actively exploiting the resulting opening. A UK government minister told the FT that British firms 'should be worried' about Anthropic's model capabilities, framing the sovereign fund as a competitive response rather than pure industrial promotion. The convergence of public capital deployment, inbound lab expansion, and ministerial urgency suggests the UK is executing a coherent strategy to become the primary non-US node for frontier AI development — but the £500 million fund is modest relative to the compute and talent costs involved.

OpenAI is reported by The Information, via Reuters, to commit more than $20 billion to Cerebras chips and receive an equity stake in the chipmaker. This is an unconfirmed report from a single source and should be treated as an announced intention rather than a closed deal. However, if accurate, it would represent a fundamental shift in OpenAI's compute strategy: from pure procurement customer to strategic investor with aligned supply incentives.

The deal's logic is vertical integration under supply constraint. Nvidia dominates AI chip supply and TSMC's capacity remains the binding constraint — the FT reports that nearly 40% of US data centre builds face delays, including projects tied to Microsoft and OpenAI. Cerebras' wafer-scale architecture offers a differentiated alternative for specific inference workloads. An equity stake gives OpenAI influence over Cerebras' roadmap and prioritised allocation, reducing exposure to the Nvidia supply queue. TSMC's 58% profit growth and upward revenue revision, reported by CNBC, confirms the demand environment justifying this kind of long-term supply lock-in.

Anthropic is simultaneously in talks to provide US government agencies with access to its Mythos model — with the White House reportedly preparing to grant federal agency access, per Reuters — and facing federal lawsuits questioning whether it poses a national security risk, as noted by the FT. This contradiction reflects an unresolved tension in US AI policy between security concerns and competitive pressure to deploy frontier capabilities in government operations. The parallel Google-Pentagon classified AI deal discussions, reported by Reuters, suggests multiple frontier labs are competing for classified government contracts simultaneously.

The FT's analysis of Mythos positions frontier model access as a scarcity resource with strategic value analogous to critical infrastructure, arguing that as capabilities advance, access to the technology becomes critically important at a national level. This framing, if adopted by procurement offices, would structurally favour the two or three labs capable of producing genuinely frontier models — concentrating government contract value at the top of the capability distribution and potentially creating winner-take-most dynamics in the most lucrative AI procurement segment.

Factory raised $150 million led by Khosla Ventures at a $1.5 billion valuation to build autonomous coding tools for enterprise clients, as reported by TechCrunch and the WSJ. The round is a confirmed closed deal. Factory is competing directly with Anthropic, OpenAI, and Cursor in agentic coding — a segment where the frontier labs hold a structural advantage through direct model access and ongoing capability improvement.

The same dynamic is playing out in design software: Anthropic's CPO resigned from Figma's board after reports emerged of a competing design tool, per TechCrunch. This is the SaaSpocalypse thesis materialising in real board-level governance conflicts — the largest labs are expanding into application-layer software, using their model superiority and distribution scale to undercut incumbents. For investors in vertical SaaS, the key question is whether any application category is defensible against a frontier lab with the same underlying model capability and a zero-marginal-cost distribution advantage.

American Express is acquiring Hyper, an AI-powered expense management startup backed by Sam Altman, in a move to accelerate its AI capabilities in financial workflows, per Reuters. Terms were not disclosed in available reporting and this should be treated as an announced deal pending confirmation of close. The acquisition is strategically motivated by AmEx's need to embed AI agents into its core expense and corporate card infrastructure — a workflow category where automation can reduce processing costs and improve fraud detection at scale.

Separately, Slash Financial raised $100 million with $300 million in ARR, using AI agents to automate document parsing and dispute processing in banking back-office operations, per Bloomberg. The two deals together signal that financial services AI adoption has moved beyond pilot into revenue-generating deployment, with both incumbents acquiring and challengers scaling — a pattern consistent with a sector in mid-transition rather than early exploration.