S26: The $2M Token Trap

Miles: So, May twentieth, Sam Altman walks into a YC event, and by the time he's done talking, Grant: Wow. Miles: every startup in the room has an offer on the table. Two million dollars. Except, not cash. Speaker 3: Compute, API tokens, and in exchange, each startup signs an uncapped SAFE giving OpenAI future equity. TechCrunch called it a mic drop moment, and honestly, that's about right. Miles: Yeah, but mic drop is doing a lot of work there. Because the moment you look at what an uncapped SAFE actually means on a cap table, the math gets complicated fast. Speaker 3: That's exactly where we're going today. Miles: With pleasure. So here's the shape of the episode. We're going to walk through the deal mechanics, what uncapped actually means when OpenAI is sitting on your cap table alongside YC and your seed investors. Speaker 3: Then we talked to two founders in the same batch, same offer, completely different calls. One signed, one walked. And the reasons tell you more about this deal than any term sheet analysis will. Miles: Nodding. And there's a YC partner lens on the back half, pattern matching this against the hundreds of deals they've seen. They've seen blow up or pay off at Series A. Speaker 3: Because the real cost of this thing is, nobody finds out until the Series A room-that's the whole problem. Miles: Founder had a sharp read on that timing risk and American Bazaar Online flagged the founder control questions early. We'll pull both threads. Speaker 3: All right, let's start where everyone in that room started-Altman's offer, what it actually said, and why the room went Miles: Yes. Speaker 3: quiet before it went electric. Miles: We all just looked at each other. Nobody said anything for maybe five seconds, and then the room went completely electric. Speaker 3: Five seconds of silence in a room full of founders. That's a long time. Miles: It really is. So let me set this up for anyone who wasn't there. Tuesday night, May 20th, Sam Altman walks into a YC event. Not a demo day, not a big public stage, just the current batch. And according to TechCrunch, he offers every single startup in the room $2 million worth of OpenAI API tokens. tokens in exchange for equity. Speaker 3: All one hundred and sixty nine of Miles: Yeah. Speaker 3: them at once. Miles: At once! YC partner Tyler Bosmeny went on X immediately afterwards and called it a mic drop moment, which honestly is earned. Speaker 3: So what does $2 million in tokens actually mean in practice? Because, cash I understand, tokens are different. Miles: Right. So the founder piece on this put it well. We're talking about API credits, compute runway. According to their analysis, that's roughly 12 to 24 months of serious... various AI infrastructure for most early-stage teams. You don't hire the servers, you don't negotiate a cloud contract, you just build. Speaker 3: No cash out of pocket. Miles: Nothing. And that's what made it feel like a gift before it felt like a deal. Speaker 3: So the room goes quiet, then electric. What was the first instinct before anyone started running numbers? Miles: I think the first instinct was pure relief. If you're six weeks into your batch, you're burning through whatever YC gave you, you're trying to figure out your AI infrastructure costs, and someone just handed you a year's worth of compute? I mean the emotional reaction is obvious. Speaker 3: Yeah, the math looks clean on the surface. American Bazaar Online flagged a vendor lock-in angle pretty quickly. Critics raised it within hours, but in the room, at that moment... Miles: In the room, nobody's thinking vendor lock-in. They're thinking, I can actually build the thing I came here to build. Speaker 3: Which is exactly when you should be most careful. Miles: I've seen this play out: the deals that feel frictionless at signing, those are the ones worth turning over, because what you signed and what you thought you signed are sometimes two different things. Speaker 3: So what did they actually sign? That's the question nobody in that room had fully answered yet. What does an uncapped SAFE really mean when the person across the table is Sam Altman? Miles: So the structure Jared Friedman confirmed to TechCrunch uncapped SAFE that converts at the Series A. Walk me through what that actually means for a founder who signed it. Speaker 3: Right, so a SAFE is a promise of future equity – no interest, no maturity date. It just sits there until you raise a priced round. And uncapped means no valuation ceiling. The higher your Series A valuation, the smaller OpenAI's actual bite. Miles: Which sounds great, right? On paper, founders want uncapped, the investor bears the valuation risk, not you. Speaker 3: And Tom Blomfield, YC partner, basically confirmed that math on X. He framed it as OpenAI exchanging roughly $800 million in compute for about 2% equity across 400 startups. So they're betting on volume. Miles: I mean, that's a wild bet, but here's my question. Who in that room was actually modeling that? Speaker 3: Nobody. Miles: Nobody. You're at a YC event. Altman just did what Tyler Bosmeny called the mic drop moment, and the vibe is free compute for a year. Your brain goes there first. That's exactly what I've seen happen with term sheets that look clean on the surface. Speaker 3: You see the number, you feel the relief, and the structural details just don't land until later. Agreed. The number was real. Miles: $2 million in API credits, useful compute, $2 million in API credits, useful compute, but the equity is the part nobody priced out that night. Speaker 3: Okay, so let's actually run it. Uncapped means OpenAI converts at whatever your Series A values the company at. If you raise at $10 million, they get a bigger slice. If you raise at $100 million... Right, Miles: They get a smaller slice. The founder who builds fast and raises high actually dilutes OpenAI down. Speaker 3: which is the founder-friendly read. And Founder's analysis noted this is the most aggressive equity-for-compute structure any model lab has tried. Most of the coverage called it founder-friendly, and technically, they're not wrong. Miles: I mean, I've pushed back a little on that framing. Speaker 3: Go. Miles: Founder-friendly on paper assumes you're going to raise a big Series A fast. A lot of these companies won't. And the uncapped structure also means there's no floor on OpenAI stake, no cap protecting anyone in either direction. Speaker 3: And notably, Friedman told TechCrunch this version doesn't include a most favored nation clause. So if you later sign a capped SAFE with another investor at better terms, Better terms, OpenAI can't automatically match those terms. Miles: That's actually the one thing that is founder-friendly here. YC's own standard deal includes MFN. Speaker 3: So they gave a little back, strip the MFN. Miles: Which is real, but most first-time founders don't even know what MFN means until their lawyer explains it after the fact. Speaker 3: That's the honest answer. The math is founder-friendly if you raise well. The problem is you're signing the deal before you know how you'll raise. Miles: Yeah, and that gap between the paper terms and the moment in the room, that's where I think the interesting decisions actually got made. Speaker 3: And speaking of decisions made in that room, Miles: Two founders who were sitting there walked out with completely different answers to the same offer. Grant: Two founders, same room, same offer, different answers. Miles: And the gap between those answers is where this gets interesting. So walk me through the one who signed. What was the actual tipping point? Grant: So she said something I keep coming back to. It wasn't really about the commute. She's got enough runway. What she wanted, and she was pretty clear about this, was that OpenAI name on her cap table before she walks into a Series A pitch. Miles: A signal. Grant: A signal. She's raising in a market where every deck looks the same. Having a model lab as an investor says something that two million in cash doesn't. Miles: Okay, but did she run the lock-in math? Grant: That's where it got quiet—I asked her directly, "What happens if you need to switch providers eighteen months from now?" and she paused for a while before she answered. Miles: What did she say? Grant: She said she'd cross that bridge when she got there. Miles: Yeah, that's not a strategy, That's a hope. Grant: Founder published an analysis of the actual switching costs on something like this. Their read: moving from one frontier model to another runs about six weeks of engineering for a small team, plus another couple weeks of quality regression to close. And that's before you touch pricing risk. OpenAI has moved API prices six times in the last two years. Miles: Six times. Grant: Mm-hmm. Miles: So you're not just locked in architecturally, you can't even forecast your own cost. Right—and none of that shows up in the SAFE document; it shows up in your engineering org chart a year later. Grant: Okay, Miles: so flip to the founder who passed. What did he see? He was blunter: Grant: he'd actually thought about the platform risk question before anyone handed him a term sheet. Meaning the Calacanis thing? Miles: He knew that argument; Jason Calacanis posted publicly on X warning founders that "there's a non-zero chance that OpenAI will study exactly Grant: what you're doing. Speaker 4: what you're doing. Grant: Exactly what your startup is doing and replicate it into a free offering. Classic platform playbook. Miles: And look, TechCrunch pushed back on that. They pointed out OpenAI can watch what you're building through API calls whether or not you've signed equity over. The deal doesn't give them new visibility. Grant: True. But this founder's point was different. He said right now, OpenAI has a financial incentive to want me to succeed. If they also start competing with me in my category, which they might. I'm holding equity from a direct competitor. Miles: Your investor becomes your competitor. Grant: He said it pretty cleanly: I don't want a vendor on my cap table, full stop. Miles: Did that cost him anything inside the batch, I mean? Grant: Socially, a little. He said there was a period where people in the cohort treated it like he'd turned down free money, like he'd made a mistake. Miles: Which is interesting, because the founder who signed, her reasoning was about the room. The signal-the social proof. Grant: And his reasoning was also about the room-just a different read of it. He thought the pressure to sign itself was a signal worth paying attention to. Miles: I don't think either of them is wrong. Grant: No. And neither of them could tell me, on the day they made the call, whether they'd made the right one. She doesn't know what her Series A will look like; he doesn't know if his competitors are already burning through those tokens to outship him. Miles: Which is the honest answer? You're making a bet under incomplete information and the outcome isn't scored for another two years. Grant: That's the part neither founder could fully answer-and honestly, I'm not sure anyone could. We put both of those conversations to a YC partner who's watched this exact fork across hundreds of batches. What they said about the pattern changes how you hear both of those founders. Let's get into them. So the two founders gave us their reasoning in the moment. What I want to know is whether that reasoning holds up against what YC partners actually see happen next. Miles: And that's the pattern math question, because any one founder thinks they're deciding on compute. A partner who's watched hundreds of companies go through this sees something different. Grant: Walk me through the cap table math first, because I don't think everyone's visualizing it. Miles: Okay, so you come into YC, you've already taken the YC standard terms. Terms you may have a pre-seed SAFE sitting unconverted now you add the OpenAI uncapped SAFE on top founder ran the scenario where a stacked unconverted instruments reach Series A representing 35 to 45 percent of the cap table before Grant: Before a single Series A dollar comes in, does Miles: a single dollar and the OpenAI instrument is one more line in that math dramatically Grant: the Series A timeline change the calculus at all? Miles: founder flagged this specifically Only 15% of seed-funded startups now reach Series A within two years, down from 51%. The uncapped SAFE converts at a price round that, statistically, most founders won't see on that timeline. Grant: So the instrument may just sit there. Miles: It sits there, it accumulates, and the longer it sits, the worse the conversion math gets if the company is growing. Grant: That's the thing that token maxing framing obscures. You're not just deciding on compute. Cute. You're deciding on a permanent relationship with your infrastructure provider before you know what your product actually needs. Miles: Which brings me to the question I kept coming back to: is token maxing a real operating strategy or is it a rationalization? Like, did these founders sign because the economics made sense or because Sam Altman was in the room? Grant: I mean, those are not mutually exclusive. Miles: Right, right. But when you strip out the room, the moment, the social pressure, the deal still needs to work on paper and the conflict of interest question is real. YC was advising on a deal that came from its own alumni network. Grant: One partner I spoke to put it this way. The signal value the founder mentioned, the OpenAI name on the cop table, That's only worth something if the Series A investor reads it as validation. Some will; some will read it as vendor dependency. Miles: And you won't know until you're in the room. Grant: You won't know until you're in the room. Which means the S-26 founders who signed three weeks ago, the verdict isn't in yet. Tokens are live, products are being built, but the test hasn't arrived. Miles: The real question every founder in his batch is sitting with is what does your cap table look like when that room happens? Grant: And whether OpenAI being in it helps you or complicates the conversation. Miles: That's where this lands for now. Grant: So right now, somewhere in a co-working space, an S-26 founder is three weeks into building on those tokens. The product's moving, but the Series A hasn't happened. Miles: Right, and that's the thing. The real price of this deal hasn't been paid yet. No one knows what that equity actually costs until they're in a Series A room. Grant: And it's not just S-26 anymore. According to American Bazaar Online, the deal reportedly extends to the summer 2025. From our S26 Batch 2, YC even extended the application deadline specifically for founders who wanted in, Miles: So we're talking hundreds of founders who still haven't faced this decision. Grant: which Miles: The question is live. Grant: means the real question every founder has. Miles: To answer not just the ones who signed is what's your compute plan for the next 24 months and is it built on a single point of failure? Grant: And the honest answer for most early stage companies right now, probably yes, because cheap and available compute wins in the short term. Miles: I've watched founders make this call before, not with tokens, but with AWS credits, with Azure deals. The infrastructure relationship always outlasts the original terms. Grant: The tokens run out; the SAFE doesn't. Miles: The founders who signed weeks ago are building real things; the cost is still theoretical, and that gap between what you signed and what you'll eventually owe-that is where this story actually lives right now. So, two founders, same offer, completely different calls. That's what stuck with me. Grant: Yeah, and the honest answer is neither of them was wrong. It really does come down to how fast you build and how high you raise. Miles: That's the thing about the uncapped SAFE structure. The math only resolves at Series A. Right now, every founder who signed is carrying equity they haven't priced yet. Grant: According to Foundra, the offer extends to the Summer 2026 batch. So hundreds more founders are about to sit in the same room, same silence, same five seconds. Miles: Five seconds before the room went electric. Grant: Right. And now they'll have this episode. Miles: Wormley, if you know a YC founder in their first year who'd actually tell their story, send them our way. yearone at hey matto dot com. Grant: And if this episode helped you see the deal differently, leave a review. It matters more than you think. Miles: Thanks for being here. We'll see you next week. Grant: See you then.