Frontier labs land in London: the talent squeeze nobody priced in

In April, two of the world's frontier AI labs signed for permanent London homes within days of each other. OpenAI took 88,500 square feet in King's Cross, room for over 500 people, more than double its current headcount. Anthropic answered with space for 800 people in the Knowledge Quarter, four times its London presence.

The coverage wrote itself. Validation of the UK ecosystem. London anointed Europe's AI capital. A win for British tech.

All true. And all beside the point, if you are building an early-stage DeepTech company within a Tube ride of those desks.

Because here is the reading almost nobody published: the same announcement that raises your ecosystem's credibility puts every ML engineer on your team one coffee away from a compensation package you cannot match, will never match, and should stop trying to match.

Both labs put the reason on the record. Anthropic's Pip White cited the UK's "exceptional pool of AI talent". OpenAI's Phoebe Thacker praised its "incredible depth of talent". They are not coming for the sandwiches. They are coming for your people.

So the question this week is not whether the expansions are good for the UK. They are. The question is what those 800 desks mean for you, and the honest answer changes shape the longer you look at it. Threat first. Then something more interesting.

The wave, in talent terms

Start with scale, because this is bigger than two labs. CNBC's June survey of the wave counted Cursor opening a London headquarters, Google moving teams into a new eleven-storey King's Cross building, and Databricks, Salesforce and Palantir all expanding, with AI companies signing a record 565,000 square feet of London office space in early 2026, according to property consultancy JLL.

Every one of those announcements cited the same rationale. "It's all about talent," as British Land's Mike Wiseman put it in the same piece.

And the effect on the ground is already being seen. Dan Hyde, founder of executive search firm Erevena, told CNBC that this activity from hugely well-funded US companies is putting pressure on local startups and making it harder to hire top talent.

Note the asymmetry in who is celebrating. The property market is celebrating. The government is celebrating. The people who compete with these firms for engineers are not quoted in the victory laps.

The moat you did not know you had

We covered talent wars in August 2025, when the threat was mercenary hiring and engineers decamping to California once they became senior enough to be noticed.

That story had a hidden moat protecting founders: relocation friction. The frontier-lab offer meant a visa, a house sale, a spouse's career, schools. Plenty of engineers looked at the package, looked at their life, and stayed. Founders were being protected by the inconvenience of geography, whether they knew it or not.

That moat is now gone. The counter-offer your best ML engineer receives now will involve no visa, no move, not even a longer commute. Same city, same flat, same school run, three times the money.

This is the genuine escalation, and it is why the subject deserves revisiting less than a year on. The competition has not merely intensified. It has been physically installed at King's Cross, a few stops up the Victoria line, and the passive protections that quietly did retention work for you have been switched off.

The number that ends the argument

If you believe the answer is to pay more, one story should cure you of it.

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, raised a record $2 billion Seed round at a $12 billion valuation. It has since lost five founding members to Meta. One of them, co-founder Andrew Tulloch, joined after the Wall Street Journal reported a package worth up to $1.5 billion over six years, a figure Meta disputed as inaccurate, while confirming the hire.

Sit with that. The best-funded startup in venture history, run by one of the most celebrated operators in AI, could not retain its own founding team against a determined bidder.

If $2 billion in the bank was not enough to win a compensation contest, no Series A ever raised in the UK will be. This is not a battle you are losing. It is a battle that does not exist. The only strategic question left is what you compete on instead, and for whom.

The exposure map

For whom turns out to be the DeepTech founder's advantage, because not every seat on your team is equally at risk.

The wage inferno is concentrated where models are built. Synthesia's Alexandru Voica told CNBC that the extreme packages sit with companies building frontier models, where skyrocketing demand has met a stubbornly fixed supply of specialists, while companies building products on top of those models face far less pressure.

Map that onto a DeepTech team and the picture sharpens. Your photonics engineers are not on a frontier lab's shopping list. Nor are your robotics hardware specialists, your quantum physicists, your power-electronics people. The labs are not hiring for cleanrooms.

Your ML engineers, however, are fully exposed. Every one of them.

So the discipline this suggests is an audit, not a panic: know precisely which seats in your company are one conversation away from King's Cross, and which are structurally invisible to it. Most DeepTech founders will find the exposed list is shorter than they feared, and clustered in a single function rather than scattered across the team. That cluster is where your attention, your equity refreshes and your succession thinking belong.

Sell what they cannot

Then there is the pull in the other direction, which the poaching headlines obscure: people leave frontier labs too, and what pulls them out is the one thing you have infinite supply of.

Simon Kohl worked on DeepMind's Nobel Prize-winning AlphaFold team. He left to found Latent Labs in London because he wanted a "laser-focused entity" rather than a broad organisation pursuing many things at once. A workstream inside a two-thousand-person lab, however gilded, is not a domain you own.

The data says he is not an outlier. Zeki's global analysis of the world's top AI talent found that companies with fewer than 50 employees employed 19% of that pool in 2024, up from 12% in 2019, in the teeth of the fiercest salary escalation the technology industry has ever produced.

The engineer who wants to ship a physical system into the world, or own a scientific domain end to end, cannot buy that at any frontier lab, at any price. You are not the discount option. For a particular kind of person, you are the only option. Recruit deliberately for that person.

The founder factory next door

Now run the clock forward, because the 800 desks change meaning again.

DeepMind has sat in King's Cross for years, outbidding everyone for talent. It has also become Europe's most productive source of founders: 112 alumni have founded or are believed to be founding startups in the last 18 months alone, and the UK and Europe have captured a third of them. Add earlier offshoots like Mistral and Isomorphic, and total funding for European DeepMind alumni companies passes $5 billion.

Every frontier-lab seat filled in London is, on a five-year view, a future alumnus: trained at someone else's expense, credentialed, networked, and statistically likely to get restless.

This is watching-brief territory, not Monday-morning action. But quietly, over time, the founders who stay close to the cluster's orbit, who build relationships with lab researchers before they leave, will find themselves fishing upstream of the richest hiring and co-founding pool the UK has ever produced. The poacher, given time, becomes the founder factory.

The same desks, twice

So look at those 800 desks once more.

In April they were a threat: a compensation gun pointed at your ML team. Look again and they are a filter, touching fewer of your seats than you feared. Again, and they are a market you can arbitrage, because they sell money and you sell ownership, and some of the best people have stopped accepting money as the answer. Once more, on the long clock, and they are a founder factory, quietly producing the people who will build the next decade of UK DeepTech, possibly with you.

The founders who read the announcement once, and only once, saw only the threat.

How many times have you read it?

 
Let's talk.