The Dinner Table That Ate Venture Capital
In the summer of 2005, a programmer who had spent the previous four years writing essays about Lisp and painting sat down with his girlfriend, his graduate school friend, and a roboticist to fund eight startups over dinner. The checks were $6,000 each — sometimes $12,000 for pairs of founders — wired from
Paul Graham's personal bank account. There was no fund, no LP base, no carried interest structure. The legal documents were so rudimentary that Graham later admitted they were "probably not enforceable." The entire operation ran out of a rented house in Cambridge, Massachusetts, where the founders ate together on Tuesday nights, and Graham dispensed advice between bites. The total capital deployed in that first batch: roughly $80,000. The total enterprise value those eight companies would eventually generate: marginal, with the exception of Reddit, which
Sam Altman would later sell to Condé Nast for somewhere between $10 million and $20 million — a modest exit that barely registered in a venture industry then organized around $100 million fund sizes and $5 million Series A checks.
Two decades later, Y Combinator has funded over 5,000 companies across more than 40 batches. Its portfolio includes Airbnb (peak market cap above $100 billion), Stripe (last private valuation $65 billion), DoorDash, Coinbase, Instacart, Dropbox, Cruise, Brex, Faire, and Reddit (which returned to public markets in March 2024 at a $6.4 billion valuation — a different kind of number than the Condé Nast deal). The combined valuation of YC companies has been estimated at over $600 billion. The accelerator — a word Graham dislikes — now accepts roughly 4% of applicants, a selectivity rate lower than Harvard's. Its standard deal: $500,000 for 7% equity, split between a $125,000 investment and a $375,000 uncapped SAFE note. The batch size has swelled from 8 to sometimes 250 or more companies per cohort. The operation that began as an experiment in disaggregating venture capital has become, by nearly any measure, the single most influential institution in startup formation — a factory, a brand, a network, a filter, and a signal all compressed into a three-month program that culminates in a two-day event called Demo Day, where the founders pitch to a room of several hundred investors who collectively manage hundreds of billions of dollars.
The paradox at the center of Y Combinator is this: it was built to be small, personal, and contrarian — a rebellion against the bloated, relationship-driven, slow-moving venture capital establishment of the early 2000s — and it succeeded so completely that it became the establishment. The institution that argued founders should be treated as peers rather than supplicants now runs the most powerful gatekeeping mechanism in Silicon Valley. The program that insisted great companies could be started with almost no money now writes half-million-dollar checks to hundreds of teams per batch, sight mostly unseen after a ten-minute interview. The essayist who believed startups were an art form helped build a machine.
This is what happens when an idea about how to help hackers becomes an institution that must scale.
By the Numbers
Y Combinator at Twenty
5,000+Companies funded since 2005
$600B+Estimated combined portfolio valuation
~4%Acceptance rate (recent batches)
$500KStandard investment per company
7%Standard equity stake
200–250Companies per recent batch
2x/yearBatch cadence (Winter and Summer)
$15B+Estimated AUM across Continuity Fund and other vehicles
The Hacker's Reformation
To understand Y Combinator, you have to understand what venture capital looked like to a technical founder in 2004 — and why Paul Graham found it intolerable.
Graham was not a conventional investor. Born in 1964, he studied philosophy at Cornell, earned a PhD in computer science from Harvard, then did something unusual: he went to the Rhode Island School of Design and the Accademia di Belle Arti in Florence to study painting. The combination — rigorous logician, visual artist, essayist — produced a mind that thought in analogies and first principles simultaneously. In 1995, he and Robert Morris (the same Robert Morris who as a Cornell graduate student accidentally created the first internet worm in 1988) built Viaweb, the first software-as-a-service company, which let people build online stores through a web browser. Yahoo acquired it in 1998 for approximately $49 million in stock. Graham was 33. He was rich, technically accomplished, and thoroughly disenchanted with the process of raising money.
The VC model of that era was, in Graham's telling, adversarial by design. Rounds took months. Term sheets were punitive. Founders were treated as interchangeable labor — the VCs supplied the "adult supervision" in the form of professional CEOs parachuted in to replace the technical founders who'd built the thing. The average seed round barely existed as a category; you either bootstrapped or raised a $3–5 million Series A from a Sand Hill Road firm that expected board seats, protective provisions, and a multi-year relationship that often felt more like indentured servitude than partnership.
Graham's insight — the one that created Y Combinator — was simultaneously obvious and radical: startups had gotten cheap to start, but the funding infrastructure hadn't caught up. Open-source software, cheap hosting (Amazon Web Services launched in 2006, a year after YC), and the web itself meant that two programmers in an apartment could build a functional product for almost nothing. The bottleneck wasn't capital. It was knowledge, confidence, and the specific tactical advice that could only come from someone who had recently been a founder themselves. The venture industry was structured to deploy $5 million at a time. The opportunity was to deploy $15,000 at a time — to hundreds of teams.
The idea was to make seed funding faster and more streamlined — to apply to startups the same kind of mass-production techniques that had been applied to other things.
He recruited three co-founders: Jessica Livingston, his girlfriend (later wife), who had been vice president of marketing at Adams Harkness, a Boston investment bank, and who brought operational discipline and an interviewer's instinct for reading people; Robert Morris, his Viaweb co-founder, who provided technical depth and a hacker's credibility; and Trevor Blackwell, a roboticist and programmer who'd also been a Viaweb colleague. The four of them put in their own money. There was no institutional capital.
The name itself was a signal. "Y Combinator" is a concept from lambda calculus — a fixed-point combinator that allows recursion in languages that don't natively support it. Naming a startup program after an obscure computer science concept was a deliberate act of identity formation: this was by hackers, for hackers. If you didn't know what a Y combinator was, you probably weren't the target audience.
The Cambridge Experiment
The first batch — Summer 2005, officially called "Summer Founders Program" — funded eight startups. The structure was almost comically informal. Founders moved to Cambridge for the summer, worked out of a rented space, gathered weekly for dinners where Graham and the partners gave advice, and presented their companies to a small group of angel investors at the end. Reddit was in that first batch: Steve Huffman and Alexis Ohanian, two University of Virginia seniors, had originally pitched a mobile food-ordering app. Graham told them the idea was terrible but liked them enough to fund a different concept — a "front page of the internet" built on user-submitted links. The pivot happened before the program started.
This detail matters because it reveals the core operating logic that would persist across 5,000 companies: YC bets on founders, not ideas. The application asks about your idea, but the ten-minute interview — the decisive filter — is designed to assess the people. Are they technically capable? Do they think clearly under pressure? Can they adapt? Would they survive the psychological crucible of a startup? The idea is almost a prop. Graham estimated that roughly 70% of the startups in early batches pivoted significantly during the program.
The Summer 2005 batch also included Kiko, a web calendar that Google Calendar would obliterate within a year (its founders later started Justin.tv, which became Twitch, which Amazon bought for $970 million in 2014 — a path dependency that illustrates how YC's real product is not any single company but the network and second chances it creates); Loopt, a location-sharing app that was ahead of its time; and several others that went nowhere.
The second batch ran in Winter 2006, and the pattern was established: two batches per year, a three-month sprint, Demo Day at the end. The deal terms were simple — initially $6,000 per founder plus $6,000 per company — and the equity stakes were modest. The machine was iterating.
Y Combinator's formation, 2005–2008
2005Summer Founders Program launches in Cambridge, MA. Eight startups funded, including Reddit. Total deployed: ~$80K.
2006Winter and Summer batches formalized. Moves to a twice-yearly cadence. Batch sizes grow to 10–15 companies.
2007Dropbox (Drew Houston, Arash Ferdowsi) enters Summer batch. Justin.tv (later Twitch) funded.
2008Heroku funded (acquired by Salesforce for $212M in 2010). YC begins building reputation beyond hacker circles.
Jessica Livingston and the Invisible Architecture
The standard narrative of Y Combinator is a Paul Graham story — the essayist who reimagined venture capital. That narrative is incomplete to the point of distortion.
Jessica Livingston was, by most internal accounts, the operational and emotional core of YC for its first decade. She conducted many of the critical interviews, made many of the final funding decisions, and — crucially — managed the community dynamics that made the program work. Her background was not technical. She had worked in finance and marketing. But she possessed something that proved more valuable than technical fluency: an almost preternatural ability to assess founders' character, resilience, and authenticity in short conversations.
Before YC launched, Livingston was already writing
Founders at Work, a book of interviews with startup founders — including Steve Wozniak, Craig Newmark, and Max Levchin — that captured the granular, unglamorous reality of building companies. The book, published in 2007, became a canonical text for early-stage founders and doubled as a recruiting tool for YC: it demonstrated that the people behind the accelerator understood what founders actually went through.
Livingston's influence operated through what might be called the invisible architecture — the dinner structure, the alumni network dynamics, the norms about how founders should treat each other. Graham wrote the essays. Livingston built the culture. When people talk about YC's "secret sauce," they often mean the social technology she designed: the weekly dinners that created horizontal bonds between founders across companies, the alumni database that allowed any YC founder to email any other YC founder, the implicit social contract that said if you were in the network, you helped others in the network. This was not accidental. It was designed.
The Move West and the Scaling Question
Y Combinator moved from Cambridge to Mountain View, California, in 2009, following the gravitational pull of Silicon Valley's talent, capital, and infrastructure. The batch sizes were growing — 20, then 40, then 60 companies per cohort — and the question that would define the next decade was emerging: could the model scale without destroying the intimacy that made it work?
Graham was explicit about the tension. He believed the quality of YC's output was a function of the partners' ability to give founders individualized attention. More companies meant more partners, which meant diluting the culture, which meant potentially undermining the very thing that produced outlier outcomes. But the math also pulled toward scale. If YC's edge was in identifying talented founders early — and if the distribution of startup outcomes followed a power law — then the optimal strategy was to fund as many potentially great founders as possible. Miss one Airbnb and you've lost more than the combined value of the bottom 80% of a batch.
The compromise was to grow the partner group. YC brought in successful alumni founders — people like Garry Tan (Posterous), Justin Kan (Justin.tv/Twitch), and Qasar Younis — as part-time partners who could mentor founders in their specific domains. The model was almost academic: a small group of tenured faculty (Graham, Livingston, Morris, Blackwell) supplemented by a rotating cast of practitioners.
By 2012, the batch sizes had reached 80-plus companies. The dinners still happened. Demo Day had become the most important two days on the startup calendar — investors flew in from around the world, and the "YC bump" in valuation was empirically measurable. A 2012 study found that YC companies raised follow-on funding at valuations 40–60% higher than comparable non-YC startups. The brand had become a signal — a Michelin star for startups.
The best startups are the ones that seem most surprising. The very best ideas look initially like bad ideas. So if you're too eager to filter them, you'll miss the big ones.
The Airbnb Proof Point
No single company in the YC portfolio did more to validate the model than Airbnb, and the story of how it got funded is itself a parable about what the model was designed to catch.
Brian Chesky, Joe Gebbia, and Nathan Blecharczyk applied to Y Combinator's Winter 2009 batch. They were running a website that let people rent air mattresses in their living rooms to strangers. They had almost no traction. They were nearly broke — at one point they were funding the company by selling custom-designed cereal boxes ("Obama O's" and "Cap'n McCains") during the 2008 presidential election. By any conventional venture capital heuristic, they were unfundable. The idea sounded absurd. The market didn't exist. The founders had no relevant domain experience — Chesky and Gebbia were designers, not hospitality executives.
Graham has said he funded them primarily because of the cereal boxes. Not because selling cereal was a good business, but because the willingness to do something that desperate and creative to keep a company alive revealed a quality he cared about more than market analysis: the founders were, in his word, "cockroaches" — nearly impossible to kill.
Airbnb entered YC in January 2009, in the teeth of the financial crisis. During the program, Graham gave Chesky a piece of advice that became one of YC's most repeated maxims: "It's better to have 100 people who love you than a million people who sort of like you." Chesky and Gebbia spent the program flying to New York, going door-to-door to meet their hosts, personally photographing listings. They grew revenue during those three months. By Demo Day, Sequoia led a $600,000 seed round. The rest — $100 billion peak market cap, a pandemic-era near-death experience and recovery, the transformation of global hospitality — is well documented.
The Airbnb story became YC's founding myth: the proof that the model could identify outlier founders before anyone else, at the cheapest possible price, in the shortest possible evaluation window. YC's 7% stake in Airbnb at a sub-$3 million post-money valuation was, on paper, one of the most profitable bets in the history of investing.
The Sam Altman Era and the Industrialization of Batch
In February 2014, Paul Graham stepped back from day-to-day leadership, and Y Combinator named Sam Altman — a 28-year-old who had been in the very first YC batch as co-founder of Loopt — as president. The appointment was Graham's idea, and it marked a decisive strategic pivot: from the artisanal phase to the industrial phase.
Altman was a different kind of leader. Where Graham was a philosopher-hacker who wrote beautifully and thought in analogies, Altman was a systems builder — ambitious, operationally intense, drawn to scale. He'd grown up in St. Louis, studied computer science at Stanford, dropped out to run Loopt, and after Loopt's modest exit (acquired by Green Dot Corporation in 2012 for $43 million), had become a prolific angel investor and one of the most connected young figures in Silicon Valley. Graham saw in him someone who could take YC's proven model and apply it at an entirely different order of magnitude.
Under Altman, batch sizes grew dramatically — from roughly 85 companies per batch in 2014 to over 200 by 2019. Altman also expanded YC's strategic ambitions beyond the core accelerator program:
YC Continuity Fund (launched 2015): a dedicated growth-stage fund that could invest in YC companies at Series B and beyond, allowing the organization to maintain its ownership stakes as companies scaled. The fund, initially sized at $700 million, signaled that YC was no longer just a seed investor — it was building a full-stack capital platform.
YC Research (launched 2015): a non-profit research lab tackling fundamental problems. Its most consequential initiative was OpenAI, co-founded by Altman with
Elon Musk, Greg Brockman, and others in December 2015, initially housed under YC Research before becoming an independent entity. The chain of events — Altman incubates OpenAI at YC, leaves YC in 2019 to run OpenAI full-time, OpenAI becomes the most consequential AI company in the world — is one of those path dependencies that makes the early history of Y Combinator feel like a single organism generating branches.
YC Fellowship (2015, later discontinued): a lighter-touch program that gave $12,000 grants to very early-stage teams without the full YC experience — an attempt to move even earlier in the startup lifecycle.
YC Growth Program (2017, later discontinued): a separate track for later-stage companies, offering the YC treatment to Series B+ startups. It struggled to find its footing — the intimacy and urgency that worked for seed-stage founders didn't translate to companies with hundreds of employees — and was quietly wound down.
The Altman era was fundamentally about asking: what if Y Combinator is not just an accelerator but a platform? A platform for company formation, capital deployment, talent networking, and institutional knowledge transfer? The answer, by the time Altman departed in March 2019 to lead OpenAI full-time, was that the platform theory worked for some extensions (the Continuity Fund became a major financial asset) and not for others (YC Growth, YC Fellowship). The core three-month batch program remained the engine.
I am confident Y Combinator is the best place in the world to start a company, and we should therefore fund as many companies as we can.
The Standard Deal and the Economics of Mass Selection
The financial mechanics of Y Combinator are unlike any other institution in venture capital, and they explain why the model is simultaneously admired and resented.
As of the most recent terms, YC invests $500,000 in each company for 7% equity. This is split into two instruments: a $125,000 investment on a post-money SAFE (Simple Agreement for Future Equity) for 7% equity, plus a $375,000 uncapped SAFE with a Most Favored Nation (MFN) provision — meaning it converts at the best terms of any subsequent financing. The post-money SAFE at $125,000 for 7% implies a post-money valuation of approximately $1.8 million at the time of the YC investment.
For a batch of 250 companies, the gross capital deployed is $125 million per batch, or $250 million per year across two batches. The Continuity Fund and related vehicles manage billions more for follow-on investments.
The economics work because of power-law math. In a typical batch of 200+ companies, most will fail. Perhaps 30–50% will return the initial investment or slightly more. Perhaps 10–20% will generate meaningful returns. And one to five companies — the Airbnbs, the Stripes, the DoorDashes — will generate returns so extraordinary that they make the entire portfolio economics work many times over. A 7% stake in a company that reaches a $10 billion valuation is worth $700 million. A 7% stake in a $100 billion company is worth $7 billion. Against a $125,000 cost basis.
This math creates a specific and unusual incentive structure: YC is optimized not for average returns but for maximum coverage of the right tail of the distribution. The marginal cost of funding one additional company is $500,000. The marginal opportunity cost of not funding a company that turns out to be Stripe is incalculable. This is why batch sizes keep growing. This is why the acceptance rate, while low, is not as low as it could be. YC would rather fund 250 companies and have 200 fail than fund 50 and miss the one that was number 51 on the ranked list.
The incumbents of venture capital initially dismissed this approach. A partner at a major Sand Hill Road firm reportedly called YC "spray and pray" in 2010 — a pejorative term for investing widely without conviction. By 2020, many of those same firms had launched their own seed programs, scout programs, and accelerator-adjacent initiatives. None replicated the YC model with comparable success.
The Network as Product
The most underappreciated asset Y Combinator has built is not its deal flow, its brand, or its partners' advice. It is the alumni network — and the specific way that network generates compounding value.
As of 2024, the YC alumni network includes the founders and early employees of over 5,000 companies. These people share a set of common experiences (the batch, Demo Day, the ten-minute interview, the weekly dinners, the emotional intensity of the three-month program), a set of common norms (help other YC founders, respond to cold emails from the network, default to transparency), and access to a proprietary platform — Bookface, an internal social network and directory — that allows any YC founder to find and contact any other YC founder.
The network effects are multi-dimensional:
Hiring. YC companies disproportionately hire from other YC companies. A failed YC founder becomes a senior engineer at a successful YC company. The talent stays in the ecosystem.
Customer acquisition. YC companies sell to other YC companies. Stripe (YC S09) processes payments for hundreds of YC startups. Segment (YC S11) provided analytics. Clerky handles legal incorporation. The internal economy is real.
Fundraising intelligence. The network acts as a distributed information system for fundraising. Founders share term sheets, compare investor behavior, and collectively negotiate from a position of information asymmetry that individual founders could never achieve. This has measurably shifted the power balance between founders and VCs in Silicon Valley.
Signal amplification. The YC brand is a signal to investors, customers, and talent that a company has passed a rigorous filter. This signal reduces information costs at every subsequent stage. Investors at Demo Day can evaluate 250 companies in two days — a throughput that would take months in a traditional fundraising process.
The network is also, to some critics, a mechanism of exclusion. The 96% of applicants who are rejected don't just lose the $500,000 investment — they lose access to the network, the signal, the fundraising intelligence, and the hiring pipeline. YC has created a two-tier system in the startup world: companies that are "YC-backed" and companies that are not. The former have measurably better outcomes in fundraising, and the question of whether that's because YC selects better companies or because the YC brand creates a self-fulfilling prophecy is genuinely unresolved.
The Garry Tan Restoration
When Sam Altman left to run OpenAI in March 2019, the presidency passed to Geoff Ralston, a veteran technologist who had been at Yahoo and had joined YC as a partner in 2011. Ralston led the organization through the pandemic — which forced the batch program online and, paradoxically, enabled geographic expansion by removing the requirement that founders physically relocate to San Francisco — and through a period of extraordinary startup formation driven by zero interest rates and COVID-era digital acceleration.
Batch sizes ballooned. The Winter 2022 batch was reportedly over 400 companies — a number that raised serious questions about whether the program could maintain any meaningful level of partner-to-founder engagement. The internal debate that had simmered since the Altman era — quality versus quantity, intimacy versus coverage — reached a new intensity.
In January 2023, Y Combinator named Garry Tan as its new CEO and president. Tan, a designer and engineer who co-founded Posterous (YC S08, acquired by Twitter) and later co-founded Initialized Capital (a venture firm that invested in Coinbase, Instacart, Flexport, and other breakout companies), represented a return to what some internally called "the founder's founder" model. He was technical. He was a YC alumnus. He understood both sides of the table — the founder's experience and the investor's economics.
Tan's early moves signaled a recalibration. Batch sizes came back down — the Summer 2023 and Winter 2024 batches were reportedly closer to 200–250 companies, still large by historical standards but a deliberate retreat from the 400+ peak. He invested heavily in AI-related companies, positioning YC as the default launchpad for the generative AI wave. He increased the standard deal from $500,000 to its current structure. And he became, in his own right, a prominent public voice for the startup ecosystem — active on social media, vocal on policy issues, willing to pick fights.
The strategic challenge Tan inherited was existential in the way that only successful institutions face: Y Combinator had won. It was the dominant force in startup formation. The question was no longer whether the model worked but whether it could sustain its edge as the landscape it had reshaped — with easier access to capital, commoditized startup advice, and AI tools that lowered the barrier to building products to near zero — made the next generation of great companies look different from the last.
The job of YC is to find the most ambitious people in the world and give them the highest-leverage help at the most critical moment.
The Demo Day Machine
Demo Day is the spectacle, but it is also the mechanism — the point where Y Combinator's entire value chain converges.
Twice a year, the graduating batch presents to an audience of several hundred investors — partners at Sequoia, Andreessen Horowitz, Accel, Index, and virtually every other firm that matters, alongside hundreds of angels and family offices. Each company gets approximately two minutes. The presentations are rehearsed obsessively during the final weeks of the batch, with YC partners coaching founders on pitch structure, slide design, and the specific metrics that will trigger investor interest.
The event operates under a set of carefully designed rules. Investors cannot make offers to companies before Demo Day — a restriction that preserves the information symmetry and competitive dynamics that benefit founders. After Demo Day, a frenzy of fundraising ensues, with the best companies closing rounds within days or even hours. The compressed timeline is deliberate: it creates artificial urgency that shifts pricing power to founders.
The Demo Day format has evolved. It moved online during COVID and partially stayed there — "Demo Day" now includes a virtual component that allows international investors to participate. The presentations have become more formulaic (a consequence of coaching at scale), and the sheer number of companies presenting has made it physically impossible for any single investor to evaluate them all carefully. Investors have adapted by specializing in specific sectors, relying on pre-Demo Day networking through YC partners, and using the Bookface platform to triage.
The economic dynamics of Demo Day are fascinating. The event creates a temporary market — a two-day bazaar where the supply of investable companies is released simultaneously to a large pool of capital. The result is price discovery under competitive pressure, which consistently produces higher valuations for YC companies than they would achieve in a bilateral fundraising process. This is not accidental. It is the mechanism.
The AI Pivot and the Present Tense
The generative AI wave that began with ChatGPT's launch in November 2022 represented a kind of perfect storm for Y Combinator. The organization that had spent two decades selecting technically brilliant founders, compressing startup formation timelines, and betting on software's ability to eat adjacent industries found itself at the center of the most significant platform shift since the mobile internet.
YC's batches since 2023 have been dominated by AI companies — by some estimates, 60–70% of recent cohorts are building AI-native products or infrastructure. The accelerator has funded companies working on AI-powered legal tools, AI drug discovery, AI code generation, AI customer service, and dozens of other applications. It has also funded companies building infrastructure for AI deployment — model fine-tuning, evaluation, observability, and safety.
The thesis is characteristically aggressive: if AI reduces the marginal cost of software creation toward zero, then the bottleneck shifts from engineering capacity to taste, speed, and market insight. This is a world where two founders with a laptop and API access can build in weeks what previously required a team of twenty and months of development. The YC model — which was always premised on the idea that startups could be started cheaply and quickly by small teams — becomes even more powerful in this environment.
But the AI wave also introduces risks to the YC model. If foundation model providers (OpenAI, Google, Anthropic, Meta) extend into applications, the addressable market for YC-style startups could shrink. If the cost of building a viable product drops to near zero, the flood of applicants and the difficulty of differentiating between them increases. And if the most valuable AI companies require enormous capital for compute — as frontier model companies do — the $500,000 YC check becomes less meaningful as a proportion of total capital needs.
Tan has addressed this by positioning YC not as an AI company picker but as an AI company factory — the place where the smartest technical founders come to find co-founders, get tactical advice on navigating the rapidly shifting AI landscape, and access the network of investors and operators who can help them scale. The bet is that the YC brand and network remain the scarce resource, even as the technology they're applied to changes.
The Stripe of Startup Infrastructure
There is a way to understand Y Combinator not as a venture capital firm or an accelerator but as a piece of infrastructure — the Stripe of startup formation.
Stripe (itself a YC company, Winter 2009 batch) made it trivially easy to accept payments online by compressing what had been a months-long integration process into seven lines of code. Y Combinator did something analogous for starting a company: it compressed the process of incorporating, raising money, finding co-founders, learning the basics of company building, and accessing a network of investors and mentors into a standardized, three-month, $500,000 package.
Before YC, starting a startup required navigating a fragmented, opaque, relationship-dependent maze. You needed to find a lawyer (and know which ones were good), find investors (and know which ones were honest), learn how to structure equity (and avoid the common mistakes), and build credibility (without having any). YC bundled all of this into a single product. The SAFE note — invented by YC in 2013 as a simpler alternative to convertible notes — became the standard instrument for seed-stage financing across the entire startup ecosystem, used by companies that had never applied to YC.
The infrastructure analogy illuminates both YC's power and its potential vulnerability. Infrastructure providers benefit from standardization and scale — the more companies that use the YC playbook, the more the playbook becomes the default, which makes it harder for alternatives to gain traction. But infrastructure also commoditizes. If the most valuable thing YC provides is a standardized startup-formation package, then any well-capitalized competitor can replicate the package. If the most valuable thing is the network and the brand, those are harder to replicate — but they're also the assets most susceptible to decay if quality declines.
The Essay Engine
One thread connects everything: the essays.
Paul Graham published his first online essay in 2001. By 2005, when YC launched, paulgraham.com was already one of the most widely read destinations for technical founders. Essays like "How to Start a Startup" (2005), "
Do Things That Don't Scale" (2013), "Maker's Schedule, Manager's Schedule" (2009), and "Mean People Fail" (2014) became canonical texts — operating manuals for a generation of founders who had no other accessible source of honest, practitioner-informed advice about building companies.
The essays were not marketing in the conventional sense. They were genuine intellectual contributions — carefully argued, often surprising, always rooted in specific examples from Graham's experience as a founder and investor. They established the conceptual vocabulary that YC founders would absorb during the program: "ramen profitable" (a startup making enough revenue to cover the founders' living expenses), "do things that don't scale" (manual, unscalable work that bootstraps the early growth of a company), "default alive" versus "default dead" (whether a company's current trajectory leads to profitability before it runs out of money).
This vocabulary did something subtle and powerful: it created a shared language. When any YC founder says "we're default alive" or "we need to do things that don't scale," they're invoking a specific framework that every other YC founder understands instantly. Language creates culture. Culture creates norms.
Norms create trust.
Trust creates network effects.
The essay engine also served as the top of YC's funnel. The founders most likely to succeed at YC were, almost by definition, the ones who had already absorbed the essays — who had internalized the Paul Graham worldview before they ever applied. The application process was, in some sense, a test of cultural fit: had you read the essays? Did you think like this? The self-selection mechanism was powerful, and it was powered entirely by freely available content.
Sam Altman continued this tradition during his tenure with a series of lectures and blog posts. Garry Tan has extended it through video and social media. But the Graham essays remain the canonical texts — the ur-documents of the YC worldview.
The $1.8 Million Post-Money Question
In the early days, YC's deal — $20,000 for 6% — implied a post-money valuation of roughly $333,000. Founders accepted it gladly because there was no alternative. The current deal — $125,000 for 7% on a post-money SAFE, plus a $375,000 uncapped SAFE — implies a post-money valuation of approximately $1.8 million. This is a deliberately low number.
A company that has already built a product, assembled a team, and demonstrated traction could reasonably argue that $1.8 million is below fair market value. YC's response, implicit but clear, is that the 7% equity is not purchasing a proportional share of current company value. It is purchasing access to the YC program, the brand, the network, the Demo Day fundraising machine, and the follow-on investment opportunities from the Continuity Fund. The equity is the price of admission, and the price is non-negotiable.
This creates an interesting adverse selection dynamic. The founders most likely to accept a below-market valuation are either (a) first-time founders who cannot raise elsewhere, or (b) founders sophisticated enough to calculate that the YC premium — measured in follow-on valuation uplift, hiring advantage, and network access — more than compensates for the dilution. The latter group is who YC most wants to attract, and the continued strength of the YC brand suggests they're still coming. But the tension is real. As the startup ecosystem has matured, more alternative pathways have emerged — other accelerators, angel syndicates, solo GP funds, AI-enabled company-building tools — and the most experienced founders increasingly have options.
The uncapped MFN SAFE is a particularly clever instrument. Because it converts at the best terms of any subsequent financing, it effectively gives YC a free option on the company's next round. If the next round is priced at a $20 million valuation, the $375,000 SAFE converts at that valuation. If the next round is at $100 million, it converts there. YC's economics improve as its best companies raise at higher prices — a self-reinforcing mechanism that aligns YC's incentives with the founders' ambitions.
What the Machine Produces
The list speaks for itself, and also doesn't.
Airbnb, Stripe, DoorDash, Coinbase, Instacart, Dropbox, Reddit, Twitch, Cruise, Brex, Faire, Gusto, Zapier, Webflow, Retool,
Scale AI, OpenSea, Razorpay, Ginkgo Bioworks, Matterport, PlanGrid, Amplitude. Each of these companies, and dozens of others, passed through a ten-minute interview, spent three months eating dinner on Tuesdays, and pitched to a room of investors on Demo Day. The combined enterprise value is measured in hundreds of billions of dollars.
The failure rate is equally telling. The majority of YC companies fail within five years. The median outcome for a YC company is probably somewhere between zero and a small acqui-hire. The distribution is brutally power-law: the top 1% of YC companies account for the vast majority of the total portfolio value. This is not a bug in the model. It is the model.
What the machine produces, in the aggregate, is not a portfolio of successful companies. It produces a system — a self-reinforcing loop of talent identification, rapid formation, network embedding, and capital allocation that generates outlier outcomes at a rate higher than any other institution in the startup ecosystem. The individual companies are the outputs. The system is the product.
And the system's most consequential output may not be any single company but the culture it has propagated — the set of norms, practices, and beliefs about how startups should be built that have become default assumptions across the technology industry. The SAFE note. The emphasis on speed. The cult of the technical founder. The suspicion of premature scaling. The belief that great companies can start in unexpected places. These ideas existed before YC, but YC systematized, codified, and distributed them at scale.
On a Tuesday evening in San Francisco, several hundred founders sit down to dinner. Some of them are building the next trillion-dollar company. Most of them are not. All of them are inside the machine. The plates are cleared. The conversations continue. The batch is three weeks from Demo Day, and the clock is running.
