Anthropic, OpenAI, and a wave of AI firms have leased more than a million square feet of London office space since early 2025, a property story with a contradictory edge, the companies most associated with reducing human labor are becoming some of the city's most aggressive new tenants.
It was not supposed to happen this way. The dominant narrative around AI and real estate for the last three years has been that automation would hollow out the office market. Fewer workers, more software, less need for square footage. That argument was attractive, and in parts of the market it has played out. But the AI companies themselves are not following the script. Anthropic just signed for 158,000 square feet at One Triton Square near Euston, a building large enough for roughly 800 people and substantially more than the company currently needs. That is not a cautious expansion. That is a bet on very aggressive growth.
OpenAI followed with 88,500 square feet at Regent Quarter in King's Cross, its first permanent London office. Databricks pre-let 139,000 square feet at the Network Building in Fitzrovia. Project Prometheus, Jeff Bezos's AI lab, is reported to be in talks for additional King's Cross space. In total, according to City AM, AI firms signed for nearly 400,000 square feet in London in under a month during April alone. CBRE is forecasting that AI-led take-up in London could reach 4 million square feet by 2033, a figure that would represent 43 percent of all unlet development space in central London. These are not modest numbers. They are a structural story about where physical capital is going.
The clustering matters as much as the volume. Almost all of these firms are converging on the Knowledge Quarter, the one-mile radius around King's Cross that is already home to Google DeepMind, Meta, Synthesia, Wayve, and a hundred academic and research institutions. That concentration is deliberate. AI companies are hiring scientists, engineers, and researchers who want proximity to universities, peer institutions, and the talent pipeline that London's technical ecosystem provides. They are not leasing premium space because it is cheap. They are leasing it because the location is integral to the recruiting strategy.
CBRE's head of tech, media, and telecoms noted that these occupiers are growing faster than any previous wave of technology companies. Firms typically start in flexible co-working spaces and move into 20,000 to 30,000 square feet within 18 to 24 months. AI companies are running that timeline at double speed, sometimes before a lease is even signed. That pace changes how property agents and landlords think about these tenants. A standard tech occupier might double in size over five years. An AI occupier might do it in two. That is a fundamentally different underwriting assumption for commercial property, and it is already shifting the outlook for London developers who have spent years worrying about oversupply.
Anthropic is a useful case study. The company currently has around 200 people in London, housed at 107 Cheapside. Its new space at One Triton Square was sized for 800, but CoStar noted that the lease significantly exceeds even that headcount. The implication is that Anthropic is planning for growth that goes well beyond its current 800-person target in Europe. That is a bet on the UK as a serious operational hub, not just a sales office, and it comes with geopolitical context. Anthropic's tensions with the US Pentagon over the use of its AI models have added urgency to the company's European positioning. UK officials were reportedly active in drawing Anthropic to London. The result is a company that is now committing capital at a speed that signals long-term permanence, not a diplomatic gesture.
The Contradiction In The AI Office Story
The irony here is not subtle. AI companies are among the most enthusiastic adopters of AI coding tools, automated workflows, and productivity software. Many of them run engineering teams that are a fraction of the size you would have expected a company of their ambition to carry five years ago. And yet they are signing for some of the largest commercial leases in one of the most expensive cities on earth. The explanation is that the savings from AI productivity do not eliminate headcount requirements for a company that is still in rapid scale mode. They compress the size of the team needed to build, but they do not reduce the demand for researchers, safety engineers, policy staff, enterprise sales teams, legal counsel, and the broader operational infrastructure that comes with going from startup to regulated commercial entity.
In other words, AI makes each person more productive, but the strategic ambition of these companies is expanding fast enough to absorb that productivity gain and still require more people. That is the basic arithmetic behind the leasing surge. The market is not saying AI has failed to improve productivity. It is saying productivity gains do not automatically translate into smaller organizations when the total addressable market is growing at the same speed as the tools that serve it.
What This Signals For Real Estate
For commercial property, the signal is significant and runs directly counter to the bearish narrative that has weighed on central London office valuations since 2020. The pipeline of demand is real, it is growing, and it is concentrated in the kind of premium, well-connected, research-adjacent buildings that are genuinely scarce. British Land has already seen the benefit. One Triton Square, a building Meta had previously paid not to occupy, is now nearly fully leased following the Anthropic deal and a handful of other tech and science signings. That reversal is not small. It says something about where the premium office market is going when the anchor tenants become the fastest-growing companies in the global economy.
CBRE's benchmark comparison to San Francisco is worth sitting with. The Bay Area AI cluster absorbed a comparable volume of office space earlier in the current cycle. London is now following a similar trajectory, which means the demand story has at least two to three more years of intensity behind it. Agents who have spent years managing oversupply in central London are now reportedly turning away smaller tenants because the pipeline of AI occupiers is large enough to fill buildings that previously looked unlettable. That is a market reversal with real consequences for city planning, development financing, and the long-term commercial property cycle.
The wider point is that AI is not just a software story or a productivity story. It is an infrastructure story, and physical infrastructure is part of it. Chips, data centers, fibre, and now premium office clusters in research-rich cities are all part of how the industry is building out its operating base. London is benefiting from that cycle in a way that was hard to predict two years ago. The city bet on becoming a serious AI hub rather than a regulatory obstacle, and the leasing activity suggests that bet is paying off faster than even optimistic forecasts expected.
Also read: TIME's new AI A-list shows the race is now global even if the money is still American • Evan Spiegel sees the AI backlash coming before most of Silicon Valley does • The EU just charged Meta with failing to keep children off Instagram and Facebook and this one has teeth
