The bull case for London property has always rested on three pillars: supply is constrained, global wealth treats it as a store of value, and jobs require physical presence in the city. You had to be there for the career, there wasn’t enough housing, and rich people worldwide wanted a foothold in a stable jurisdiction. These forces pushed prices relentlessly upward for three decades.
The AI transition is about to stress-test this thesis.
Cities are expensive because people need to be in them. This sounds circular, but the causation runs through employment. High-paying knowledge work clustered in cities because knowledge workers needed to collaborate, and collaboration required physical proximity. Agglomeration effects meant that talent attracted talent, which attracted capital, which attracted more talent. London, New York, San Francisco, Singapore: these became extraordinarily expensive places to live because being there was the price of admission to the best careers.
A 25-year-old graduate moves to London and pays £1,800 per month to flatshare in Zone 2 because the job is in Canary Wharf and remote isn’t an option. This calculus has driven urban migration for decades. The rent is painful, but the career upside justifies it.
What happens when the job doesn’t exist, or doesn’t require presence?
If LLMs automate significant portions of knowledge work, the demand for knowledge workers falls. If the remaining knowledge work can be done remotely, the requirement to be in London falls. Either way, the gravitational pull that justified paying £22,000 per year just for a bedroom weakens considerably.
The most exposed asset class is commercial office space. The logic is straightforward: companies needed offices because employees needed desks. If companies need fewer employees, or those employees don’t need to be on-site, office demand collapses.
This is not speculative. COVID-19 already demonstrated that most knowledge work can be done remotely. Office occupancy in London remains well below pre-pandemic levels; Knight Frank data from late 2024 showed Central London office vacancy rates around 9%, up from historic averages of 5-6%. This is before AI has materially reduced headcount in most industries. The occupancy rate reflects changed preferences; the headcount reductions haven’t even started.
If a typical financial services firm currently employs 10,000 knowledge workers in Canary Wharf and can reduce that to 3,000 with AI augmentation, and those 3,000 only need to be on-site two days a week, office space demand falls by perhaps 85%. The firm doesn’t need a 500,000 square foot headquarters; it needs 75,000 square feet of flexible collaboration space. This arithmetic is coming for every industry that employs people to manipulate information.
Commercial landlords are exposed to a structural demand shock that isn’t cyclical. It’s not that firms will need more space when the economy recovers; it’s that the model of filling towers with knowledge workers is ending.
Cities are ecosystems. The office workers buy lunch from Pret, get their shirts dry-cleaned, drink after work at the pub on the corner. An entire service economy exists to serve the knowledge workers who serve the corporations. When the knowledge workers leave, the service economy follows.
London’s economy is approximately 90% services. Financial and professional services alone account for roughly 25% of London’s GVA. These are precisely the sectors most exposed to AI automation. The downstream employment in retail, hospitality, and personal services depends on the continued presence and spending of the primary knowledge worker population.
The question is not whether this affects London’s economy. The question is how fast, and whether anything replaces it.
Commercial property has a clean thesis: fewer workers means less demand for desks. Residential is messier because people live in cities for reasons beyond employment.
London has genuine amenity value. World-class cultural institutions, healthcare, education, restaurants, diversity, global connectivity. Rich people worldwide still want a London flat as a store of value and a foothold in a stable, rule-of-law jurisdiction. These forces don’t disappear because AI can write code.
Immigration continues to drive population growth. London’s population has grown by roughly 1 million since 2000, primarily through international migration. If the UK remains an attractive destination, people will keep arriving, and they need to live somewhere.
Housing supply remains genuinely constrained. The green belt, planning restrictions, and NIMBY politics mean London cannot easily build its way out of scarcity.
These factors support residential prices. The question is whether they’re strong enough to offset the weakening of the jobs magnet.
Consider a simplified model. London has approximately 4 million jobs. Perhaps 40% are knowledge work that could be significantly impacted by AI automation, either eliminated or made remote. That’s 1.6 million jobs. If half of those jobs disappear or leave London over ten years, that’s 800,000 fewer workers needing to be in the city.
Those 800,000 workers have partners, children, flatmates. The population effect might be 1.5 to 2 million people who no longer need London housing. London’s current population is around 9 million. We’re talking about a potential 15-20% reduction in housing demand from this effect alone.
Against that, set continued immigration, the amenity value for the wealthy, and the supply constraints. My guess is these forces roughly offset for a while, producing stagnation in real terms rather than collapse. Nominal prices might stay flat or grow slowly while inflation erodes the real value (an effect we’re already seeing). The dramatic appreciation of 1995-2020 is probably over; whether it reverses depends on the speed of the transition.
The deeper question is whether cities as a concept retain their value. For decades, the answer to “why live somewhere expensive and cramped?” was “because that’s where the work is.” If that answer no longer holds, we need a new one.
Maybe the new answer is “because cities are where the other people are.” Humans are social animals; we value proximity to other humans for its own sake, not just for economic collaboration. Restaurants, theatres, dating pools, serendipitous encounters: these require density. If people have more leisure time because AI handles the work, perhaps they value urban amenities more, not less.
Maybe the new answer is “because I already own here and I’m not selling.” Residential property markets are sticky. People don’t move in response to economic signals the way capital does. Homeowners absorbing a paper loss on their property will simply stay put, reducing transaction volumes without crashing prices.
Or maybe there isn’t a good new answer, and we’re at the beginning of a slow unwinding of the urban premium. Cities became expensive over a specific period for specific reasons. Those reasons are changing. It would be strange if prices didn’t eventually reflect that.
I don’t know which scenario plays out. I suspect the muddled middle: some decline in real terms, masked by nominal stickiness, with the adjustment happening over 15-20 years rather than in a dramatic crash. The people most exposed are those who bought at peak prices with high leverage, expecting continued appreciation. The people least exposed are those who own outright and value the amenities regardless of the investment return.
What I’m confident about is that the old model, where you could buy London property and assume it would appreciate 5-7% annually forever because jobs would always require presence, is broken. The jobs are changing. The presence requirement is changing. If you’re making a 25-year bet on London property, you need a thesis that doesn’t depend on knowledge workers commuting to offices. I’m not sure what that thesis is.