"Founders should live together" is the kind of claim the technology industry makes constantly and tests rarely. So here is the test, in two parts: what the research says about proximity and ideas, and what actually came out of the houses.
Part one: the research
The strongest available evidence is not from startups at all. It is from science.
In November 2023, Nature published "Remote collaboration fuses fewer breakthrough ideas" by Yiling Lin, Carl Benedikt Frey and Lingfei Wu — an analysis of millions of research papers and patents examining what changes when collaborators are in different cities. The finding is specific and uncomfortable for distributed work:
Among distributed team members, collaboration centres on late-stage, technical tasks involving more codified knowledge, yet they are less likely to join forces in conceptual tasks — such as conceiving new ideas and designing research — when knowledge is tacit.
Remote teams execute. They do not conceive. The authors note that despite enormous improvements in digital tools, remote teams are less likely to integrate their members' knowledge into new, disruptive ideas, and warn that the post-pandemic shift to remote work could deliver smaller, more incremental advances while making breakthroughs harder.
Read that against what an early-stage startup is. A startup is almost entirely the conceptual phase: deciding what to build, for whom, and why the obvious approach is wrong. It is the exact category of work the paper finds degrades at a distance — because the knowledge involved is tacit, half-formed, and transmitted in the kind of conversation that does not survive being scheduled.
This is the honest version of the argument. Not "offices are magic" — the paper is not about offices. It is that the specific activity of fusing two people's half-ideas into one new idea appears to need physical proximity, and that is the activity a founding team does all day.
Part two: the institutions voted with their rules
Watch what the programs did, not what they said.
Y Combinator has required founders to relocate to San Francisco for the full duration of the batch since Winter 2025, with no remote track at all. It now runs four batches a year of roughly 150–200 companies — 199 in Winter 2026, 193 in Spring 2026 — on an unchanged standard deal of $125,000 for 7% on a post-money SAFE plus $375,000 on an uncapped MFN SAFE.
Consider what that mandate costs YC. It shrinks the applicant pool to people who can uproot their lives for three months, in one of the most expensive housing markets on earth, during a period when rents set records. An organisation with YC's data on what predicts company outcomes chose to pay that price anyway. That is a revealed preference, and it is a strong one.
HF0 goes further and compresses it into one building: around ten teams, twelve weeks, one mansion, with food, housing and laundry handled. The Residency pushes the density higher still — up to twenty-five founders per house across five San Francisco mansions, one of them a twenty-two-bedroom Pacific Heights house with twenty-two founders in it.
None of these programs are short on capital or options. They all converged on the same format.
Part three: the receipts
The format's defenders usually stop at vibes. Here are outcomes, with sources attached.
Lovable went through HF0's residency. In February 2025 it raised a $200 million Series A led by Accel at a $1.8 billion valuation. In December 2025, a $330 million Series B led by CapitalG and Menlo Ventures at $6.6 billion. In August 2026, a $400 million Series C led by Menlo Ventures and the Scaleup Europe Fund at a reported $13.3 billion valuation. From residency to roughly $13 billion inside about eighteen months.
The rest of the HF0 portfolio includes OpenRouter, Krea, Browser Use, Coframe, Featherless AI, Forage and Zeta Labs — a concentration of funded AI infrastructure companies out of a program that admits roughly ten teams at a time. HF0's own fund has scaled with it, from reported vehicles of $8 million and $16 million to a reported $100 million.
Mercor is the more extreme data point, and the one that bears on the education argument. Brendan Foody, Adarsh Hiremath and Surya Midha were high-school friends from the speech and debate team at Bellarmine College Preparatory. They dropped out of college together in their sophomore year in 2023 — Hiremath from Harvard, Foody and Midha from Georgetown — and were later awarded Thiel Fellowships. Mercor started by matching freelance engineers with US companies, then moved into supplying expert contractors to AI labs for training data. It reached a reported $10 billion valuation, and as of July 2026 was reported in talks to raise $500 million at a $20 billion valuation. All three founders were still under 23.
The Thiel Fellowship, which pays young people $200,000 to leave or skip college for two years, has produced alumni including Figma's Dylan Field and Scale AI co-founder Lucy Guo, with one analysis putting total value creation above $750 billion and a unicorn hit rate near 14%.
What the receipts do and do not prove
Be careful here, because this is where most writing on the subject becomes dishonest.
These programs select ruthlessly. HF0 takes about ten teams from thousands of applicants and skews heavily toward repeat founders with prior exits or unicorn experience. YC accepts a low single-digit percentage. When you select the most capable founders in the world and then house them well, you cannot cleanly separate the effect of the housing from the effect of the selection. Lovable might well have become Lovable in a WeWork.
So the claim worth defending is narrower, and it survives the objection:
- Proximity measurably improves the conceptual, idea-fusing work that early-stage companies consist of — that part is research, not anecdote.
- Every serious program that could choose otherwise has chosen co-location, including at significant cost to its applicant pool.
- Removing household logistics returns attention to the work, which is why the programs all do it rather than just renting desks.
- Peer density converts extreme effort from a personality trait into an ambient norm.
None of that requires believing a house is a substitute for a good idea. It requires believing that the same team ships more in a house than scattered across four sublets in four neighbourhoods, which is both plausible and, on the Nature evidence, likely.
The operational version
If you accept the argument, it becomes a logistics problem — and this is where most teams lose the benefit they came for.
The failure mode is predictable. A team of six arrives for a three-month batch, cannot find one house that takes them together for three months, and ends up in four separate places across the city. They have paid San Francisco prices and kept exactly the geographic dispersion the research says degrades their best work. The commute is now internal to the company.
What actually preserves the benefit is unremarkable: everyone in one building, or in two houses close together; a kitchen and common table big enough that the default evening is shared; internet that holds; and nobody spending Saturday arranging cleaning or furniture.
That is the whole of what hyperstay does — furnished homes across San Francisco, 30-night minimum, private rooms or an entire house with six or more bedrooms, rent and utilities and internet and cleaning in one monthly price, and twenty-one-plus rooms available across the portfolio when a cohort arrives together. We have hosted founders from Y Combinator, Hexa, Stanford's Mayfield Fellows Program, Latitud and General Catalyst-backed companies.
The research says proximity is worth something. The programs agree. The only remaining question is whether your team actually ends up in the same building — which is a booking, not a philosophy.
Bringing a team or a cohort to San Francisco? team@hyperstay.com.