# Paul Graham just put a number on the post-YC geography decision — and the press missed it

URL: https://www.thedeepfeed.ai/posts/2026-05-13-paul-graham-yc-stockholm-half-as-likely-unicorns/
Category: Business
Published: 2026-05-13
Author: the-deep-feed
Tags: paul-graham, y-combinator, silicon-valley, stockholm, startup-hubs, founders
Kind: deep

> YC now has internal data on what happens to startups that go home after the batch. They're only half as likely to become unicorns. Graham disclosed it on stage in Stockholm; coverage led with the wrong sentence.

## TL;DR

- **The buried lede:** Graham disclosed that YC internal data shows startups that go home after the batch are **only about half as likely to become unicorns** as those that stay. This is the first time a number has been put on the post-YC geography effect.
- **The press misread:** Business Insider framed the talk as Graham *"throwing cold water on the Stockholm-is-the-next-Silicon Valley idea."* He did the opposite — he explicitly endorsed Stockholm: *"You could make it the Silicon Valley of Europe. That job is still up for grabs."*
- **The 16-year arc:** Every structural argument in the talk — talent density, serendipitous meetings, faster investor decisions, the prophet-in-his-own-country effect, big-fish-small-pond — appears verbatim in Graham's 2006–2011 hub essays. The only new arguments are the unicorn stat and the *"call our API"* pitch.
- **The 60-year-culture claim is new:** Graham now dates Silicon Valley's pay-it-forward custom to ~1965 and says it has fully de-coupled from economic incentives. *"In Silicon Valley, people help you for no reason."* That's the cultural-fossil framing, not the original-incentive framing he used in 2006.
- **The contest he's calling:** *"Where's the Silicon Valley of Europe? The job is still available."* In May 2026, the candidates with strongest current signal are **Stockholm** (Lovable $6.6B, Legora $5.55B, Sana $1.1B exit), London, Paris, and Berlin. Stockholm's unique tell is that Graham himself flew to it.

## The data point everyone missed

![Two hand-drawn editorial bar columns on cream paper: a tall grey "STAYS IN SF" column on the left, a half-height red "GOES HOME" column on the right, connected by a dashed midpoint rule — the half-as-likely-to-unicorn gap visualized as a 2x penalty](/post-images/2026-05-13-paul-graham-yc-stockholm-half-as-likely-unicorns/half-as-likely-divergence.jpg)

[**Paul Graham**](https://x.com/paulg) flew to Stockholm on April 29, 2026 to give a 22-minute talk at YC's first European founder event. The talk [went up on X and YouTube on May 13](https://x.com/i/status/2054570434025460199), where it has picked up **3,500+ likes and 390 reposts** on the YC account and another **52,000 YouTube views** in a week. YC framed the upload around the geography question, not the data point inside it:

> Paul Graham (@paulg) whether founders should move to Silicon Valley, and what it takes to build a startup hub anywhere else. Live from our YC | Stockholm event on April 29, 2026.
>
> — [@ycombinator](https://x.com/ycombinator/status/2054570434025460199), May 13, 2026

The press read it as a Silicon Valley pep talk. The headline that lapped the others was [Business Insider's](https://www.businessinsider.com/paul-graham-on-stockholm-becoming-next-silicon-valley-hub-2026-5): *"Paul Graham just threw cold water on the Stockholm-is-the-next-Silicon Valley idea."* That headline is wrong, and the line everyone should have led with is buried 17 minutes into the talk:

> "YC now has a lot of data about this. And the startups that go home after YC don't do as well as the ones that stay. Startups that go back home after YC are only about half as likely to become unicorns."
>
> — Paul Graham, [YC Stockholm, 18:48–19:01](https://www.youtube.com/watch?v=QHJkUw31YX8)

That is a number nobody outside YC has had before. **Y Combinator** has graduated [more than 5,000 companies](https://www.ycombinator.com/about) across 40+ batches since 2005. A meaningful fraction of those founders are not American, and a meaningful fraction return home after the program. Until this sentence, the cost of going home was a folk belief. Now it is a roughly **2x penalty on the probability of crossing $1B**, disclosed on stage by the firm that has every dataset on it.

For European, Indian, and South American founders deciding whether to stay in San Francisco after Demo Day, that one number is the actionable line of the talk. The rest is structural argument, and most of the structural argument is a republication of essays Graham wrote between 2006 and 2011.

## The press got the headline backwards

Business Insider's claim, that Graham *"threw cold water"* on Stockholm, survives only if you read the talk as a binary: SV or not-SV. That isn't the talk Graham gave. The actual two-question structure he opens with is:

> "I'm going to talk about two questions today. The first one is about you. Should you go to Silicon Valley? The second one is about Sweden. What should Sweden do to thrive as a startup hub?"
>
> — [00:00–00:14](https://www.youtube.com/watch?v=QHJkUw31YX8)

His answer to question one is *go to Silicon Valley, but you can come back.* His answer to question two is *Stockholm can be the Silicon Valley of Europe; that slot is still open.* Those are complementary, not opposed. The bridge between them, in his own words:

> "If you do well enough at transplanting Silicon Valley here, you could do more than just make Stockholm thrive as a startup hub. Here is the exciting idea. You could make it the Silicon Valley of Europe. That job is still up for grabs."
>
> — [20:13–20:30](https://www.youtube.com/watch?v=QHJkUw31YX8)

A speaker who travels to Sweden to deliver that line at the Waterfront Congress Centre is not throwing cold water. He is **publicly endorsing a contestant**. The relevant questions are who else is in the bracket and what the implied bet means for European founders.

## Sixteen years of the same argument

![A hand-drawn axonometric stack of eight thin horizontal slabs on cream paper, each labeled with a year (2006, 2007, 2008, 2009, 2011, 2016, 2026) and the topmost slab "STOCKHOLM 2026" rendered in editorial red — the talk's structural arguments traced through Graham's hub-essay canon](/post-images/2026-05-13-paul-graham-yc-stockholm-half-as-likely-unicorns/sixteen-year-thesis-stack.jpg)

The structural arguments in the talk are old. The freshness, and what makes this talk worth watching rather than re-reading, is the *evidence base* underneath them. Every claim Graham used to argue from anecdote in 2006 he now argues from sixteen years of YC dataflow.

| Claim in the 2026 Stockholm talk | First appears in |
| --- | --- |
| Big centers concentrate talent — *"Paris for painting in 1870, Göttingen for math in 1900, Hollywood for movies in 1950"* | [How to Be Silicon Valley (May 2006)](https://www.paulgraham.com/siliconvalley.html) |
| Serendipitous meetings dominate planned ones | [Cities and Ambition (May 2008)](https://paulgraham.com/cities.html) |
| Investors decide faster in SV because of competition for deals | [Why to Move to a Startup Hub (Oct 2007)](https://paulgraham.com/startuphubs.html) |
| *"No prophet in his own country"* — leaving home raises your local status | [Why to Move to a Startup Hub (Oct 2007)](https://paulgraham.com/startuphubs.html) |
| Big-fish-in-a-small-pond → calibration against known big fish | [Cities and Ambition (May 2008)](https://paulgraham.com/cities.html) |
| Founders should *"go for a bit and then come back"* to seed home hubs | [Can You Buy a Silicon Valley? (Feb 2009)](https://www.paulgraham.com/maybe.html) |
| Pay-it-forward culture in SV is unusual and durable | [Why Startup Hubs Work (Oct 2011)](https://paulgraham.com/hubs.html) |
| All you need to start a hub is "a place founders want to live and a critical mass" | [How to Be Silicon Valley (May 2006)](https://www.paulgraham.com/siliconvalley.html) |
| Mountain View was a backwater; geography is downstream of people | [How to Make Pittsburgh a Startup Hub (Apr 2016)](https://paulgraham.com/pgh.html) |

This is not a criticism. It is a feature. When a thesis survives 16 years of YC dataflow without the author having to walk a single sentence back, that is signal. The opposite pattern, where a public figure has revised the same prediction multiple times in a year, looks like [the Altman roadmap we covered last week](/posts/2026-05-01-hassabis-2030-agi-disease-compression/). Graham's hub thesis has held the line through ZIRP, COVID dispersion, the remote-work hype cycle, and the AI capital surge, and now appears, in his telling, validated by YC's internal cohort statistics.

What the table also shows is that the *only* structural claims in the talk that weren't already in print before 2016 are:

1. The half-as-likely-to-unicorn stat (new in 2026)
2. The *"call our API"* line — that any government hoping to engineer a Silicon Valley adjacency can simply ship founders to YC (new in 2026)
3. The 60-year-old cultural fossil framing (sharpened from 2011)

Three new claims, two of them quantitative-ish, in a 22-minute talk. That's a good ratio for a public-canon speaker.

## Decomposing the 2x unicorn gap

YC does not publish per-cohort unicorn rates. Public estimates from outside the firm put YC's all-time unicorn rate at roughly **5–6% of graduated companies**, with [a list of more than 300 YC unicorns documented as of 2025](https://www.ycombinator.com/topcompanies). If half-as-likely is roughly right, the implication for a foreign founder is this:

| Cohort behavior after the batch | Estimated probability of becoming a unicorn |
| --- | --- |
| Stays in the Bay Area | Baseline (call it **X%**) |
| Returns home post-batch | Roughly **X/2 %** |

There are honest caveats, and Graham himself names two of them on stage:

> "First of all, there's a lot of selection bias in this. It's not just measuring the effect that Silicon Valley has on startups. It's measuring also the confidence and the determination of the founders. If you're more confident and determined, you're more likely to move to another country to do your startup."
>
> "Also, it's not just measuring the effect of Silicon Valley on the company's performance, just their valuation. And everybody knows companies in the Bay Area can raise money at higher valuations."
>
> — [19:07–19:32](https://www.youtube.com/watch?v=QHJkUw31YX8)

Translated: the 2x gap is a mix of three things — (a) genuine Bay Area causal lift on company outcomes, (b) selection on founder traits that correlate with both moving *and* succeeding, and (c) Bay Area valuation inflation. Graham doesn't decompose them; he can't, without a randomized trial nobody is going to run. But the *unicorn* metric is a financial-event metric, which means (c) is doing real work in the gap. A founder who comes home and builds a $500M company in Stockholm in 2032 is not in the unicorn count; the talk explicitly retools this as a feature:

> "Even if you only do half as well, that's still pretty good. If you'd have become a billionaire in the Valley, so that you only have 500 million — well, there's no difference between those two things. In fact, in Swedish, it sounds even better. You have **five billion kronor**."
>
> — [19:36–19:54](https://www.youtube.com/watch?v=QHJkUw31YX8)

So the actionable read is: **going home costs about half your unicorn-event probability and approximately none of your shareholder value at scale.** That is closer to a tax than to a wall, and it is consistent with the talk's overall thesis — go, then come back.

## Three first-disclosures in the 2026 talk

Set aside the canon. Three things in the talk are first-disclosures or recently sharpened.

**1. The 60-year cultural fossil.** Graham now dates the SV helpfulness norm to roughly the founding of Fairchild Semiconductor and Shockley:

> "The custom is 60 years old now. Now it just seems to people in Silicon Valley, that's how everybody acts."
>
> — [14:37–14:43](https://www.youtube.com/watch?v=QHJkUw31YX8)

In 2011's [Why Startup Hubs Work](https://paulgraham.com/hubs.html), Graham described the norm as something close to an active economic incentive — be nice to nobodies because some of them will become Mark Zuckerberg. In 2026, he is saying the explicit incentive is no longer load-bearing. The custom has detached from its origin and become a cultural fossil that reproduces itself by imitation. That is a different and stronger claim, because *imitable customs scale to other geographies in a way that one-off economic incentives do not.* It is also why he thinks transplanting SV culture to Stockholm is possible.

**2. The "call our API" pitch.** Graham makes an unusual policy claim — that YC has *already* become the cheapest and most efficient mechanism a government could buy to expose its founders to Silicon Valley:

> "If the Swedish government designed a program to help Swedish founders experience Silicon Valley, they couldn't do better than this. And it doesn't even cost them anything because it's funded by Silicon Valley investors... And they don't even have to license it. They can just call our API."
>
> — [18:16–18:36](https://www.youtube.com/watch?v=QHJkUw31YX8)

This is the first time we have seen Graham frame YC as a *public-good substrate for non-US innovation policy.* It is also a recruiting line. Sweden's government has spent the better part of a decade [trying to retain its AI talent](https://www.ft.com/content/cb371b43-d5af-4666-a9f5-28ec3e00ea63) as US firms acquire them; Graham is suggesting that the inverse, strategic export with mandated return, is the better policy. The economic argument is tight: YC is funded by SV LPs, takes 7% standard, and produces measurable cultural-import flow. A founder fellowship program with comparable selectivity and outcomes would cost a national budget on the order of tens of millions a year. Sending applicants through YC costs zero state dollars.

**3. The Dropbox/Sequoia fax story.** Graham has alluded to Dropbox's funding trajectory in print before, but the [Stockholm telling](https://www.youtube.com/watch?v=QHJkUw31YX8) adds detail we haven't seen:

> "For the past year, a big Boston VC firm had been keeping a benevolent eye on them, offering them lots of encouragement and advice. Not money, just encouragement and advice... And when Dropbox went out to Silicon Valley, they came to the attention of Sequoia. When the Boston VC firm found out that Sequoia was interested, their opinion of Dropbox changed so fast that they probably gave themselves a whiplash. They sent Drew, the founder, a term sheet. They faxed it to him... with a blank valuation."
>
> — [07:23–08:29](https://www.youtube.com/watch?v=QHJkUw31YX8)

Color, not data — but the *blank-valuation faxed term sheet* is a sharper artifact than the existing public record had. It also re-makes the local-investor-bias point: the Boston firm's price discovery on Dropbox was set by Sequoia's interest, not by Dropbox's underlying metrics. That is the cleanest single illustration in the talk of why investor latency outside SV is not a small operational cost.

## Stockholm's case for the European slot

![Three hand-drawn editorial bar columns on cream paper, sized proportionally to AI-company valuations: Legora $5.55B (left, grey), Lovable $6.6B (middle, editorial red — the tallest), Sana $1.1B (right, grey, shortest), with a "STOCKHOLM 2024-2026" rule above — three AI unicorns from a metro of 2.4M](/post-images/2026-05-13-paul-graham-yc-stockholm-half-as-likely-unicorns/stockholm-unicorn-cluster.jpg)

Graham closes the talk by formally inviting Stockholm to compete:

> "Where's the Silicon Valley of Europe? It's not just like, you know, everyone doesn't immediately say like, there's a certain answer. I mean, if there were an answer, it would seem a ridiculous question to ask. Nobody asks like, where's the Silicon Valley of America? Duh. So that means the job is still available."
>
> — [20:30–20:51](https://www.youtube.com/watch?v=QHJkUw31YX8)

The case for Stockholm in 2026 is unusually strong on company evidence. The headline data, all from the last 18 months:

| Stockholm-born AI company | Most recent valuation / exit | Date | Source |
| --- | --- | --- | --- |
| **Lovable** | $6.6B Series B (vibe-coding) | Dec 2025 | [Unified AI Hub](https://www.unifiedaihub.com/ai-news/lovable-secures-330m-series-b-at-6-6b-valuation-era-of-ai-powered-builders) |
| **Legora** | $5.55B Series D + €42M extension (legal AI) | Mar–Apr 2026 | [TechCrunch](https://techcrunch.com/2026/03/10/legora-reaches-5-55-billion-valuation-as-ai-legaltech-boom-endures/), [EU-Startups](https://www.eu-startups.com/2026/04/swedish-legal-ai-startup-legora-raises-e42-million-extension-to-series-d-bringing-total-to-over-e500-million/) |
| **Sana Labs** | $1.1B acquisition by Workday | 2026 | [EQT ThinQ](https://eqtgroup.com/thinq/technology/sana-ai-fueled-journey-to-exit) |
| **Klarna** | Public; multi-tens-of-billions market cap | 2025 IPO | Prior |
| **Spotify** | Public; ~$100B market cap | Continuing | Prior |

Three AI unicorns in 18 months from a metro of 2.4M people is the densest founder yield in Europe over that window. It is also the part of the talk where Graham is most evidence-aware: he repeatedly references **Max Junestrand** of Legora and **Gustaf Alströmer**, YC's Stockholm-born partner who has been [pushing the European program internally](https://linkedin.com/in/gustafalstromer) for years.

The case *against* Stockholm, and the case for London or Paris, runs through density, capital, language, and immigration policy. Stockholm sits at 2.4M metro vs. London at 9.5M and Paris at 11M. UK and French capital pools are larger, and English-language defaults are easier for inbound talent. Stockholm's counter, articulated in the [FT's 2026 Europe hubs report](https://www.ft.com/content/cb371b43-d5af-4666-a9f5-28ec3e00ea63), is that smaller-and-denser is currently outperforming larger-and-diffuse — the city is producing per-capita more billion-dollar AI companies than any other European capital, and the founder network reads more like Y Combinator's early batches than like London's enterprise SaaS scene.

What Graham does *not* address is the dilemma that the Bens Bites coverage [framed as the structural question for the cohort](https://news.bensbites.com/posts/48729-swedish-ai-startups-lovable-legora-and-sana-which-is-set-to-be-acquired-by-workday-highlight-a-dilemma-in-european-tech-sell-out-or-bet-on-getting-big): *sell out or bet on getting big.* Sana exiting to Workday at $1.1B in the same year Lovable is fundraising at $6.6B is a fork in the road for the European playbook. Graham's *"come back home"* policy is implicitly a bet-on-getting-big argument — coming home to absorb a $1.1B acquisition is not the move he is recommending.

## The decision matrix for a non-US founder

![A hand-drawn axonometric cascade of five rectangular path-tiles on cream paper, falling diagonally from upper-left to lower-right, each labeled (1. STAY HOME, 2. YC THEN STAY, 3. YC THEN RETURN, 4. SOLO TO SF, 5. SKIP YC STAY EU) — the middle tile "3. YC THEN RETURN" rendered in editorial red as the recommended path](/post-images/2026-05-13-paul-graham-yc-stockholm-half-as-likely-unicorns/five-founder-paths.jpg)

Here is what the talk actually implies for a founder in Stockholm (or Bangalore, Lagos, São Paulo, or Berlin) who has read Graham's older essays and is deciding what to do in 2026.

| Path | What the data says |
| --- | --- |
| **Stay home, never go** | Worst expected outcome on unicorn probability; selection-biased downward on confidence. PG calls this the "village" path. |
| **Apply to YC, stay in SF after** | Maximum unicorn probability per Graham's stat; maximum dilution risk via the new [OpenAI $2M-for-uncapped-SAFE](/posts/2026-05-22-openai-2m-tokens-yc-batch-equity/) cap-table pressure. |
| **Apply to YC, return home after the batch** | ~50% unicorn probability vs. staying, per the new disclosure; mostly intact $-value at scale; maximum cultural-import contribution to your home hub. |
| **Skip YC, move to SF independently** | Gains the geography but loses the curated peer set, the "super-valley within the valley," and the YC investor decision-speed effect Graham specifically cites as YC's edge. |
| **Skip YC, stay in Stockholm/London/Berlin** | The mode of every successful European AI company shipping in 2026 (Lovable, Legora, Sana). PG would still tell you to *visit* — even briefly. |

The talk's clean recommendation, if you take Graham at face value, is path three: **YC, then come back.** That recommendation also happens to be commercially good for Y Combinator — they get to graduate founders into European capitals where their LP money can compound for another generation. But the *cohort behavior the recommendation describes already exists in Stockholm in 2026*: most of the YC alumni who founded Stockholm AI companies in 2024–2025 followed exactly this path.

## What this changes

Three things change in the public record after this talk.

🔴 **The post-YC geography decision now has a number.** Founders deciding between staying in the Bay Area and going home can do that math openly. Previously they were guessing. *"About half as likely to become unicorns"* is the first publicly disclosed YC-internal coefficient on that decision. Other accelerators — Antler, Techstars, Entrepreneur First, Combinator-imitators globally — should be expected to publish or refuse-to-publish their own version of this number. The ones who refuse will be the ones whose data is worse.

🟡 **The European hub competition has a publicly endorsed candidate.** Graham did not bless Stockholm casually; he flew to the Stockholm Waterfront Congress Centre and said *"that job is still up for grabs"* in front of the city's founder class. London, Paris, and Berlin do not yet have an equivalent on-stage endorsement from a YC co-founder. That asymmetry will move capital allocation decisions inside YC's LP base and outside it. Stockholm's pitch on talent density just got the strongest available external referee.

🟢 **The 60-year cultural-fossil framing makes the SV culture import argument scalable.** If pay-it-forward is no longer an incentive-driven custom but a cultural fossil, it can be transplanted by founders moving home, not just by economic conditions reproducing themselves. That is a stronger thesis than Graham held in 2006. It also makes the *"go and come back"* policy a culture-import program, not just a network-and-money-import program. The implication: every YC batch is now also a cultural exchange program, and the unit of export is the alumnus.

## The talk in context

Stepping back from the line-by-line: this is Graham doing what he has been doing since 2006, with a sharper evidence base, in front of a city that has earned its slot on the bracket. The press got the headline backwards because they read it as a binary, and Graham's argument has never been a binary. It is a two-step: go to the center, then return. The new data point gives that two-step a price tag. The Stockholm endorsement gives it a destination.

For a non-US founder reading this in 2026, the line worth taking home is not *"go to Silicon Valley"*. Graham has been saying that for sixteen years. It is the half-as-likely number, the *come back if you can* corollary, and the implicit asymmetry the corollary creates: if every YC alum who returns home is paying a tax in expected unicorn probability and almost none in expected wealth at scale, then a hub like Stockholm is essentially being subsidized, at zero state cost, by the choice of every returning founder.

If a critical mass of those founders return at the same time, the math at the end of the talk is no longer a thought experiment:

> "All you need is a place founders want to live and a critical mass of them. Stockholm, I can tell from being here, is the kind of place founders want to live. And who knows how close you are to having a critical mass? That's the thing about critical masses. You don't know until you hit it, and then pow."
>
> — [21:29–21:49](https://www.youtube.com/watch?v=QHJkUw31YX8)

That is the line worth quoting, not the one Business Insider used.

## Sources

- [Y Combinator — Paul Graham, Founder of Y Combinator, Live from Stockholm (13 May 2026)](https://www.youtube.com/watch?v=QHJkUw31YX8)
- [X — Y Combinator status (13 May 2026)](https://x.com/i/status/2054570434025460199)
- [Business Insider — Paul Graham just threw cold water on the Stockholm-is-the-next-Silicon Valley idea (15 May 2026)](https://www.businessinsider.com/paul-graham-on-stockholm-becoming-next-silicon-valley-hub-2026-5)
- [Paul Graham — How to Be Silicon Valley (May 2006)](https://www.paulgraham.com/siliconvalley.html)
- [Paul Graham — Why Startups Condense in America (May 2006)](https://paulgraham.com/america.html)
- [Paul Graham — Why to Move to a Startup Hub (Oct 2007)](https://paulgraham.com/startuphubs.html)
- [Paul Graham — Cities and Ambition (May 2008)](https://paulgraham.com/cities.html)
- [Paul Graham — Can You Buy a Silicon Valley? Maybe. (Feb 2009)](https://www.paulgraham.com/maybe.html)
- [Paul Graham — Why Startup Hubs Work (Oct 2011)](https://paulgraham.com/hubs.html)
- [Paul Graham — How to Make Pittsburgh a Startup Hub (Apr 2016)](https://paulgraham.com/pgh.html)
- [Founder News — Is Stockholm The World's Hottest Startup City Right Now?](https://foundernews.eu/is-stockholm-the-worlds-hottest-startup-city-right-now/)
- [TechCrunch — Legora reaches $5.55 billion valuation (10 Mar 2026)](https://techcrunch.com/2026/03/10/legora-reaches-5-55-billion-valuation-as-ai-legaltech-boom-endures/)
- [EU-Startups — Legora raises €42M extension to Series D (Apr 2026)](https://www.eu-startups.com/2026/04/swedish-legal-ai-startup-legora-raises-e42-million-extension-to-series-d-bringing-total-to-over-e500-million/)
- [Unified AI Hub — Lovable raises $330M at $6.6B valuation (Dec 2025)](https://www.unifiedaihub.com/ai-news/lovable-secures-330m-series-b-at-6-6b-valuation-era-of-ai-powered-builders)
- [EQT ThinQ — Sana's AI-fueled journey to a $1.1bn exit](https://eqtgroup.com/thinq/technology/sana-ai-fueled-journey-to-exit)
- [Financial Times — Sweden's AI start-ups capture growing US interest (Europe's Leading Start-up Hubs 2026)](https://www.ft.com/content/cb371b43-d5af-4666-a9f5-28ec3e00ea63)
- [Y Combinator — Requests for Startups](https://www.ycombinator.com/rfs)
- [Y Combinator — Top Companies](https://www.ycombinator.com/topcompanies)

---

Canonical: https://www.thedeepfeed.ai/posts/2026-05-13-paul-graham-yc-stockholm-half-as-likely-unicorns/
Site: https://www.thedeepfeed.ai
Full corpus: https://www.thedeepfeed.ai/llms-full.txt