London vs Berlin vs San Francisco for AI founders in 2026

Every AI founder I know has had this argument at least once: London, Berlin or SF? They are the three cities where the AI scene is actually deep enough to matter, and the trade-offs between them are real. Cost, talent, capital, lifestyle, regulation — they cut differently in each.

I am London-based and biased, fine. I have spent enough time in the other two to write an honest take. This is not a "which one is best" piece because that depends entirely on what you are building and who you are. It is a "what is each one actually good for" piece.

For the London-side of things specifically, the AI events feed gives you the week-by-week pulse — Anthropic London talks, AI House demo days, the open-source crowd at Conway Hall.

The honest summary

San Francisco wins on senior research-engineer talent, capital density, and pure scene velocity. It loses on cost of living, quality of life for anyone with a family, and visa friction.

London wins on talent breadth (especially applied ML and full-stack), capital that is actually willing to lead seed rounds, regulatory clarity for enterprise sales, and being a sane place to live. It loses on the very top end of the talent pool and on the velocity of the SF founder-investor flywheel.

Berlin wins on cost, lifestyle, the open-source and research community, and a frankly more interesting city. It loses on capital density (German VCs are slower and more conservative) and on the pace of the scene compared to the other two.

If you are doing pure foundation-model research with a billion dollars to raise, SF. If you are building applied AI for enterprise customers, London is the better bet. If you are building open-source infrastructure on a tight budget, Berlin.

The talent question

This is the one most founders care about. The honest version:

San Francisco still has the deepest senior research-engineer pool in the world. OpenAI, Anthropic, Google DeepMind's SF office, Mistral's SF presence, the entire Y Combinator AI cohort each year. If you need to hire someone who has shipped a frontier model, SF is where they are. The downside is the salary war is genuinely insane — total comp for a senior research engineer is now routinely over $1m a year. If you are early-stage, you cannot compete on cash, and the equity story has to do a lot of work.

London has Google DeepMind's headquarters, Anthropic's London office, Mistral's growing London presence, and a much deeper applied-ML and full-stack pool than people give it credit for. The Imperial, UCL, Cambridge and Oxford pipelines feed in real talent every year. The salary expectations are 30-40% lower than SF for equivalent levels. The downside is the very top of the research pool is thinner — the people who could be at OpenAI but chose London are a smaller group.

Berlin has Mistral's German office, the Aleph Alpha team in Heidelberg-but-Berlin-adjacent, and a strong open-source and research community. TU Berlin and the Max Planck institutes feed in good people. Salaries are even lower than London. The downside is the senior applied-ML pool is thinner, and a lot of the strongest German engineers move to Zurich for Google or to SF for the bigger names.

Capital

SF capital is the most aggressive and the deepest. Sequoia, a16z, Founders Fund, the SF arm of Lightspeed, the AI-specific funds like Conviction and Radical. Cheques are bigger, decisions are faster, and there is more competition between funds, which is good for founders. The downside is they expect you to be in SF, or at least visiting every six weeks.

London capital has matured considerably in the last three years. Index, Atomico, Balderton, Accel London, Hoxton, LocalGlobe — all of them lead AI rounds at sane valuations now. The Sifted-tracked deal flow is real. Slower than SF, fewer competing offers, but the seed-stage market is liquid in a way it was not five years ago.

Berlin capital is the slowest of the three. Earlybird, Cherry, Speedinvest from Vienna, Project A. Good funds, conservative pacing, smaller cheques on average. German LP money is risk-averse compared to American or even British capital. The upshot is valuations are friendlier, but you will spend more time fundraising for the same dollars.

Customers

This is the underrated dimension. Where are your enterprise customers actually buying?

The big US enterprise market — Fortune 500 — buys in dollars and mostly buys from SF-based startups, full stop. If your ICP is the US enterprise, being in SF makes sales materially easier. The chance encounters at SF dinners, the warm intros, the fact that your competitors are also there — it compounds.

UK and European enterprise is a different game. Banks, insurers, the FT, the public sector — they buy from London-based vendors with a real London presence and a clear regulatory story. The EU AI Act, GDPR, the FCA's rulings on AI in financial services — these matter, and being in London with a UK entity is a meaningful sales advantage for those customers.

Berlin is somewhere in between, but particularly strong if your customer base is German Mittelstand (the mid-cap industrial companies). They buy from German vendors. Speaking German is genuinely a sales advantage there in a way it is not in London or SF.

Cost of living and the lifestyle question

San Francisco is the most expensive of the three by a margin. A two-bed flat in a normal neighbourhood is $5-6k a month. A senior engineer who can afford to buy a house probably cannot, in SF proper. The quality of public services is, charitably, mixed. The food is genuinely good. The weather is fine.

London is expensive by European standards, cheap compared to SF. £3-4k a month for a two-bed in Zone 2. The food scene has gone from a punchline to genuinely world-class in the last decade. Public transport actually works. The weather is what it is.

Berlin is cheap. A nice flat in Mitte or Prenzlauer Berg is €1.5-2k. The food is fine, the bars are excellent, the city is full of culture, the parks are huge, and you can live well on a salary that would be poverty-level in SF. The downside is the pace is slower in a way that some founders find frustrating. The bureaucracy is real.

The scene velocity

SF's AI scene moves faster than the other two combined. Demo days, party-houses, founder dinners, hackathons, the AGI House crowd — there is something every night and the rooms are dense with people who can move things forward. The downside is it is exhausting and self-referential. A non-trivial number of founders burn out on it.

London's scene has matured a lot. AI House at King's Cross is the centre of gravity now. Anthropic London, DeepMind, the Imperial AI lab, the Hawksmoor founder dinners. Slower than SF but the quality of conversation is, in my biased opinion, occasionally better. Less posturing.

Berlin's scene is smaller and less commercial. The open-source crowd, the research crowd, the Mistral-adjacent community. If you want a city where the AI scene has not been overrun by people pitching, Berlin is it. The downside is you have to seek it out — it is not constantly happening at you.

Visas and immigration

If you are not American, SF is the worst of the three. The H1-B lottery is brutal, the O-1 path is expensive and fragile, and the pace of policy change makes planning a multi-year career hard. A lot of the European AI talent that "should" be in SF is in London or Toronto for visa reasons.

London's Skilled Worker visa is annoying but predictable. Five years to settled status, then citizenship. The Global Talent visa is the better bet if you have the credentials. Worth the paperwork.

Berlin's Blue Card route is fine for senior people. The general work visa is fine for normal hires. The bureaucracy is German bureaucracy, which means it works but slowly.

Who should be where

If you are pre-seed and need the maximum velocity of meeting investors and engineers, SF.

If you are seed-stage and your customer base is European or your team has strong UK ties, London. The Anthropic London office is becoming a real hub. AI House is good. Capital is fine.

If you are open-source-first, research-first, or building developer tools and you do not need US enterprise revenue immediately, Berlin is genuinely a great place to build. Cheap, smart, and the team can have a life.

If you are doing pure foundation-model research with megaround ambitions, SF and only SF, untill the AI labs in London and Paris start funding rounds at the same scale (which is starting, slowly).

What I would actually do

If I were starting an AI company in 2026, I would start in London. The talent is good, the capital is functional, the regulatory clarity for enterprise is a real advantage, the cost is sane, and the visa story is workable. Then I would open an SF office at Series A if the customer base demanded it.

A non-trivial number of London-founded AI companies do exactly this. The London-first, SF-second pattern is real, and it is becoming the default for European founders who do not want to uproot their lives.

For the week-by-week London side, the AI events feed is what I keep open. The other two cities have their own scenes and their own logic. Pick the one that fits the company you are actually building, not the one with the loudest Twitter footprint.