The artificial-intelligence business moves at the speed of light down fibre-optic cable, yet it has parked itself in a handful of postcodes. Its world capital is a strip of northern California. Its British seat is the district around King’s Cross, where Google DeepMind has just moved into its first purpose-built headquarters, and where the Alan Turing Institute, the Francis Crick Institute, University College London, OpenAI and Meta all sit within a short walk of one another. Three hundred years ago the business of insuring ships against the sea did precisely the same thing, in the coffee houses of the City of London — and the institution those coffee houses produced is, to this day, a room full of people standing a few feet apart. The technology could hardly be more different. The instinct is identical.
That instinct — clustering — is real, measurable and very old. It is also routinely over-obeyed. Because a cluster’s advantages are genuine, firms conclude that everything they do must sit inside one, and pay the postcode premium on the whole payroll. The conclusion does not follow. The very thing that makes a cluster valuable also tells you, quite precisely, which parts of a business belong in it — and which are burning money there.
The air of the trade
The economics was settled in 1890, when Alfred Marshall asked in his Principles of Economics why industries bunch together when land and labour are plainly cheaper spread out. His answer came in three parts. A cluster builds a deep pool of specialised labour, so employers can always find the skills and the skilled can always find work. It sustains a web of specialist suppliers that no single firm could support alone. And — the decisive part — it lets knowledge leak. Where everyone works at the same trade, Marshall wrote, its mysteries cease to be mysteries: they hang, “as it were in the air, and children learn many of them unconsciously”. One person’s idea is picked up by another, improved, passed on. Proximity does not merely cut costs. It manufactures knowledge.
The modern proof arrived a century later. In Regional Advantage (1994), AnnaLee Saxenian set out to explain why Silicon Valley had pulled decisively ahead of Route 128 outside Boston, when both began with the same universities, the same technologies and the same head start. Her answer had almost nothing to do with technology. Boston’s firms were fortresses — vertically integrated, secretive, loyal to a fault. The Valley was porous: engineers changed employers without stigma, talked across firms, and left to found startups with the blessing of the people they were leaving. The Valley won not because it knew more, but because what it knew would not stay still. And note the kind of knowledge in question — the tacit kind, the half-formed hunch, the overheard problem, the thing read off a colleague’s face. The kind that survives a conversation and dies in a document.
A room full of people
For the purest case on record that co-location can be the product, walk to Lime Street. Lloyd’s enters history in 1688 not as an institution but as a coffee house near the Thames, whose keeper, Edward Lloyd, noticed that his customers — sailors, shipowners, merchants — wanted information more than coffee. He hired runners to bring news up from the docks: which ships were in, which overdue, where the storms and the pirates were. The crowd that gathered for the news became a market. Men willing to carry the risk of a voyage for a premium did business at his tables — the “boxes” — and marine insurance in its modern form was born in the room. When Lloyd moved to Lombard Street in 1691, the market simply moved with him. There was nothing to ship and nothing to store. The product was the right people in one place.
Nor was he alone. The financial City precipitated out of coffee houses. In Exchange Alley, stockdealers thrown out of the Royal Exchange for rowdiness regrouped at Jonathan’s, where in 1698 the broker John Castaing began posting a regular price list — the acorn of the London Stock Exchange. The Baltic Exchange, which still sets the terms of world shipping, began at the Virginia and Baltick coffee house on Threadneedle Street. Daniel Defoe reckoned in 1719 that a man could walk the whole circuit of this district in about a minute and a half. Clustering is not a metaphor the City occasionally reaches for. It is the City’s origin story — and the deeper rhythms of the markets it built are the subject of a standing programme elsewhere on this site.
Here is the detail that should give a sceptic pause. Three and a half centuries on, in an age of free video-conferencing, the market that began in that coffee house is still a room. Lloyd’s trades from Richard Rogers’s inside-out building on Lime Street, and at its heart is a vast hall known simply as the Room, where brokers still walk risks from box to box and underwriters still price them face to face, at desks whose design has barely changed since the seventeenth century. Uniformed waiters — lineal descendants of Edward Lloyd’s coffee servers — still move among them. The pandemic ran the experiment everyone assumed would finish the Room off: the market went remote, proved it could, and came back, because the placing of complex, bespoke, high-value risk turned out to need the air. Only last year a specialist insurer took a box — number 388, third gallery — as a deliberate step into the London market. Lloyd’s own defence of the Room — the chance encounters, the market intelligence, the nurturing of talent — reads as though Marshall drafted it. And King’s Cross is the same mechanism in glass and steel: the research, the specialised talent and the capital of British AI gathered in one square mile — the contours of that field were sketched in an earlier piece in this series — while Old Street’s “Silicon Roundabout”, the earlier and cheaper cluster, has watched the centre of gravity move north. Some work still has to happen where the ideas are in the air.
The other half of the lesson
So clustering is real. It does not follow that everything belongs in one — and the reason is the mechanism itself. If a cluster’s value is the circulation of tacit knowledge, the test for any activity is blunt: does this work actually run on tacit knowledge? Some does. Placing a complex one-off risk; originating a deal; forming an investment view in the crossfire of argument; the senior relationships, market-facing and regulatory, that are built on trust and read from a face. Pull those out of the cluster and you have genuinely damaged them. But much of what a large financial institution does is nothing of the kind — the enterprise finance estate anatomised at length elsewhere on this site: policy administration, claims, fund accounting, the finance and actuarial back office, technology delivery, contact centres, the machinery of data and settlement. That work runs on process, scale, accuracy and cost. It does not get better on Lime Street. It only gets dearer.
Dearer twice over, in fact. Rent is the visible premium, and the smaller one. The quieter cost is the labour market: park a large processing function on the most competitive square mile in the country and you are bidding for staff against every firm on it, for people who can cross the road for a rise on a Friday. You pay over the odds and churn faster for the privilege. Move the same function to a city where you are a flagship employer rather than one bidder among hundreds, and you pay less and keep people longer. The premium bought nothing, because the work never depended on the air.
The discipline of the wrong postcode
The familiar version of this — send the back office to Glasgow, Belfast, Bristol or Bangalore — is true but stale, and invites the sneer that these are only the tasks nobody wanted. The sharper evidence runs the other way: firms that became world-beaters by refusing the obvious cluster. Baillie Gifford manages some £200bn, to the highest reputation in active investment, from Edinburgh — a partnership there since 1908, and pointedly not in the City. Fund management sustains a genuine cluster of its own in Edinburgh, so Marshall’s advantages are not forfeited; and the firm is candid that distance from the herd instinct of the London market positively helps the patient, long-horizon judgement its record rests on. Admiral, meanwhile, built the only FTSE 100 company headquartered in Wales — a motor insurer, from a standing start in Cardiff in 1993, now employing some fifteen thousand people there. Volume personal-lines motor is as far from a Lloyd’s box as insurance gets: it is won on pricing algorithms, data and the economics of a contact centre, and sold through the price-comparison sites Admiral itself pioneered. Proximity to Lime Street would have improved none of it. Both firms grasped that the right question is never “where does our industry sit?” but “where does this work belong?”
I have spent a good part of a career inside exactly that question — cost bases and operating models across a succession of banks and, latterly, an insurer/asset manager — and I can report that it is fluffed at least as often as it is answered. The commonest failure is not extravagance. It is never asking: inheriting a footprint, assuming the expensive postcode is load-bearing, and never once testing whether the air is doing any work.
What the regulator wants
For a regulated insurer or asset manager there is a second push in the same direction, and for once the regulator and the finance director are shoving the same way. Since 2022 the Prudential Regulation Authority and the Financial Conduct Authority have required firms to name their “important business services”, set “impact tolerances” for how long each could bear disruption, and — by a deadline that fell in March 2025 — prove they could hold those tolerances through a severe but plausible shock. Concentration risk sits squarely inside that regime: the supervisor wants to know what happens when a single point of failure fails. A firm that has stacked every critical operation into one building on one square mile has manufactured precisely the concentration the rules exist to expose — one fire, flood, power cut or cordoned street from losing the lot. The recent run of cloud outages and cyber-attacks on household names has only sharpened the point. The rules do not order dispersal in so many words; their whole logic runs with its grain. Spreading critical operations across sites and cities is no longer just a saving. It is a resilience virtue the regulator will thank you for.
On the ground
The instruction is neither caricature — not “get out of London”, and not paying the premium reflexively because that is where serious firms are seen. It is a discipline. Map what the firm actually does, function by function, and put one question to each: does this work run on the air — the tacit, face-to-face circulation of judgement and relationship — or on process, scale and cost? Keep the first kind in the cluster without apology, and pay what proximity truly costs, because there it earns it: the specialty underwriter belongs on Lime Street and the frontier researcher in King’s Cross, exactly as the marine underwriter belonged in Edward Lloyd’s coffee house. Send the second kind down to earth — Edinburgh, Cardiff, Glasgow, wherever it can be done well, cheaply and away from the single point of failure. The result is cheaper, stickier in the labour market and more resilient in the regulator’s eyes, all at once. A firm’s premises, read properly, are a map of where its knowledge actually lives. What must be in the air, keep in the air. Everything else can come down to the ground.