Anthropic spent the first quarter of 2026 mapping a list of nine candidate cities for its first non-San Francisco research presence and quietly cut the list to two finalists by mid-April. Neither of them is London — a result that has surprised every European recruiter we've spoken to and several of Anthropic's own UK applicants. The two cities still in the running, per three people involved on the company's people-side team and one external real-estate advisor, are Tokyo and Zurich. The decision is expected before the end of Q2.
That London is out reframes the entire frontier-lab geography map. Google DeepMind is headquartered in King's Cross. Stability AI built around Shoreditch. ElevenLabs operates out of London and New York. The assumption inside the policy press for the last eighteen months has been that any second Anthropic office would land in the same corridor — partly for talent flow, partly for proximity to the UK AI Safety Institute, partly because Dario Amodei and Jared Kaplan have personal academic ties to Cambridge. The opposite has happened. London made the long list. It did not make the short list.
The reason, on every reading we can defend, is sovereign-AI policy alignment, not talent supply. Anthropic's senior leadership has publicly emphasised that constitutional AI requires regulatory partners willing to co-develop safety frameworks ahead of deployment. Tokyo offers a METI-coordinated industrial policy stance that maps almost cleanly onto Anthropic's own messaging on responsible scaling. Zurich offers neutral-jurisdiction credibility and a policy environment that has, since 2024, been rebuilding itself as an AI-governance brand around ETH Zurich and the Swiss Federal Council's digital ministry. Both carry a tax profile that materially beats London after the UK's 2025 corporate-rate adjustment.
What Anthropic is optimising for, by our cut, is the ability to sit with a national regulator at a single table and negotiate model-deployment frameworks for an entire region. The lab does not need a second talent pool of 4,000 ML engineers. It needs a credentialed jurisdiction it can point to when the EU AI Act, Japan's new model-disclosure rules, and Singapore's IMDA framework all simultaneously demand a local responsible entity. That is a policy decision dressed as a real-estate decision.
The headcount math reinforces it. Anthropic's San Francisco footprint sits around 1,400 employees as of April, up from roughly 850 at the same point in 2025. The international hub mandate, per the same internal sources, is not 1,000 hires — it is 120 to 180 over eighteen months, weighted heavily toward policy, alignment research, and enterprise GTM rather than core pre-training. That is a mission outpost, not a satellite engineering org. London at 1,000 ML engineers is the wrong instrument for a 150-person mission outpost.
For competitive context: OpenAI's London office, opened in 2023 under Diane Yoon's people leadership, now sits north of 200 staff and has tilted toward go-to-market and customer success. Google DeepMind's UK headcount remains the densest pure-research cluster outside North America. Mistral has effectively claimed Paris as the European frontier-lab seat. Meta FAIR retains a Paris presence. London is contested. Tokyo and Zurich are open ground.
The UK government will read this as a setback. Prime Minister Keir Starmer's administration has staked considerable political capital on London becoming the post-Brexit AI-safety capital, and the Bletchley-to-Seoul-to-Paris-to-San Francisco AI summit cadence was meant to keep that flywheel turning. Losing the second Anthropic office to Tokyo or Zurich is the kind of signal CHROs at British banks and consultancies will notice. We expect at least two follow-on consequences: a tightening of the UK's AI talent visa scheme to compete on access rather than volume, and a more aggressive equity-grant posture from London-based Series-C AI startups trying to retain candidates who would otherwise have waited for Anthropic to call.
The forecast: Tokyo wins on a 60/40 read. METI is further along in formalising the kind of co-developed evaluation framework Anthropic's policy team has been pushing for, and Japan's combination of regulatory clarity, a stable yen-cost base, and an unsaturated frontier-lab presence is unusually attractive. Zurich is the conservative pick — a defensible legal jurisdiction with a globally portable workforce — and would likely become a policy-and-enterprise office rather than a research one. Both finalists tell the same underlying story: in 2026, frontier labs are choosing geographies for regulatory leverage first and for engineering supply second.
The lesson for Fortune 500 talent leaders watching Anthropic move is straightforward. The next round of AI-related site-selection decisions across enterprise will not be driven by where the engineers live. It will be driven by where the company can negotiate jurisdiction. Saudi Arabia understood this with PIF and HUMAIN. The UAE understood it with G42. Singapore has been positioning for it since IMDA's 2023 framework. The labs are now confirming what sovereigns built for. The question for every CHRO planning an international AI footprint in 2026 is no longer "where is the talent?" — it is "where can our policy team get a meeting?"
For more on how frontier labs are reshaping the global hiring map, see the ENTRA Intelligence ranking of the Top 30 AI Founders to Watch in 2026.