OpenAI Restructuring: Who Wins and Who Loses & Mercor Raises $350M at a $10BN Valuation

Summary

• OpenAI is free. The deals with Microsoft and the attorneys general of Delaware and California are cut: a PBC under a ~$135B charitable foundation, Microsoft at 27% (a 10x on its $13B), employees and the nonprofit at ~20% each — and Sam Altman still owns zero shares while Elon Musk argues for a trillion-dollar package. Rory’s verdict: “we’re out of the stupid corporate structure getting in the way of everything stage of business,” and Elon can still litigate, but “possession is nine-tenths of the law.”
• The IPO trade is now live. Jason can’t think of “a retail IPO that would be more popular than OpenAI” — nobody will read the prospectus past the $250B of cloud commits — and conversion may unlock ~$200B more equity “to go the distance,” making the circular Nvidia/AMD/Oracle promises more financeable. Jason puts a trillion-dollar mark roughly two years out on current trajectory: $500B on ~$12B GAAP revenue is 40x, and at 40x “if your growth rate slows, you fall sheer.”
• a16z’s $10B raise looks smaller than the headline — $6B growth (post-fees just 22 $200M checks), $1.5B AI apps, $1.5B infra, $1B defense — but the model wins two ways: optioning (pay a “stupid price” at seed because you’re buying the B) and the wall of news. Rory: “quantity has a quality all its own… Andreessen Horowitz is the red army of the venture industry.” The only possible losers are LPs, and only if late-stage returns settle at 12-13%.
• Everything training-data is one trade. Likely Mercor’s $350M at $10B — $500M revenue in 17 months, believed fastest ever — is a great market with two flaws: an unspecified large share of gross goes to the PhDs, and two buyers are 50%+ of revenue. Underwriting a 3x means $30B, i.e. $6B of training-data revenue at 5x — “makes a man pause.” Rory’s frame: “this whole thing top to bottom is one big-ass bet on AI capex hypergrowth” — Jason supplies the Chuck Prince keep-dancing analogy.
• Ultra-late-stage is a different asset class. On Ramp’s reported $30B (from $23B), Jason: “I don’t even consider it an up round. It’s not enough” — dilution can leave the price per share flat. Rory: these are “public stocks hiding in private” — Stripe went 91→110 — and nobody doing billion-revenue companies at $30B is honestly underwriting a 3x or a 30% IRR.
• The Carta data kills spray-and-pray without picking. Of 547 2018 Series Bs, two-thirds returned under 2x, yet the blended distribution still pencils to ~3x net — if you fill the 5x/10x buckets. Indexing the whole vintage returns ~0.2x even with Figma’s 100x inside; only Y Combinator has a mass-production seed business. Three strategies exist — spraying, picking, optioning — and a16z’s 72 seed bets (vs 27 for number two) show whose superpower optioning is.
• Sell or ride: Synthesia’s $3B Adobe test. Rory: if ARR really went 100→150 in under six months, “I can make this conversation really quick — there’s no way they should sell.” Jason would still say sell — “I want to be the guy that gives that advice” — because for a founder $3B vs $10B is “no difference,” and the real question is “do you really want to run a public company?” Roomba is the cautionary tail: Lina Khan blocked the $1.7B Amazon deal and the company is now heading for bankruptcy.
• Amazon’s fortnight from hell, graded. History’s largest white-collar layoff (10%), cloud share down to 38% from ~50% in 2018, Raymond James seeing AI cloud share at 7%. Jason gives Bezos an F for punching out at the 2021 top without selling; Rory’s hyperscaler scorecard: Google built its own model (and owns 14% of Anthropic), Microsoft rented one, “and you did nothing — so you lose.”

Deep dive

1. OpenAI cut both deals — the straitjacket is off

• Rory’s summary of the day’s news: OpenAI settled with Microsoft and the attorneys general of Delaware and California, implemented the restructuring, and can now raise capital normally — a PBC (Patagonia-style, “a different kind of American corporation”) sitting under one of the best-capitalized charities on earth. Elon can still litigate, “but possession is nine-tenths of the law” — once the AGs allow it and conversion happens, it’s very hard to unwind.
• The structure landed roughly where leverage predicted: Microsoft 27%, employees ~20%, nonprofit ~20%. Rory’s litigator’s framing: experienced lawyers ask “what’s the case worth?” and after all the human drama, “this settled out for what the cases were worth.”
• The genuinely unprecedented piece: Sam Altman still owns no shares — in the same news cycle as Elon arguing he deserves a trillion-dollar package “so the robots don’t kill us.” Jason: “Have we ever seen anybody own nothing? It’s crazy.” Rory thinks it cuts both ways: “you can’t assail the man for capitalism when his other billion-dollar entities are the ones that let him finish off the McLaren collection.”
• Winners, per Rory: Microsoft — a 10x on $13B plus IP rights, an Azure contract and residual rev share (“corporate development deserves a gold star from shareholders in a way that frankly Microsoft R&D does not” — the stock popped); a $135B charitable foundation (“I would have laughed in 2016, but they’ve succeeded at a big-picture level”); and Bret Taylor, best board chairman twice in five years — Twitter jammed down Elon’s throat at $44B, now this. “No losers” — nobody got more than the risk they took deserved.

2. The most popular retail IPO of all time, and the trillion-dollar clock

• Jason’s structural point: conversion may unlock ~4x more equity access — if OpenAI could IPO near $2 trillion it might raise another $200B, “and they may need that to go the distance.” Harry extends it: every stock that popped on an OpenAI purchase promise can now point to a real corporate structure — he’s still “skeptical Oracle will collect the last dollar of that cloud contract,” but funding it “is just banking and math” now.
• On retail appetite, Jason is categorical: “I just can’t think of a retail IPO that would be more popular than OpenAI” — buyers won’t read the prospectus caveats or the $250B of third-party cloud commits, they’ll just say “let me get some of that OpenAI.” Jason: bankers “might do it for half nothing just to be on the biggest IPO of all time.”
• Trillion-dollar company in 2026? Jason: on current trajectory probably two years, absent a euphoria moment — $500B against ~$12B of “boring GAAP revenues” is 40x, “and we saw this in ‘21: if you’re trading at 40 times revenues and your growth rate slows, it’s nasty and you fall sheer.” Meanwhile SoftBank is closing a ~$300B-valuation primary while doing employee secondaries at $500B — the same security at two prices, “you’re 40% IRR in an hour.”
• The browser gets a shrug: “it’s tough to keep track of how many stories tech VCs and X create… it could be like MCP — interesting, but not really the game changer VCs think it’ll be.” Jason doubts the average consumer needs a new browser rather than typing ChatGPT into the old one.

3. a16z’s $10B: the red army raise that’s smaller than it looks

• The breakdown — $6B growth, $1.5B AI apps, $1.5B AI infra, $1B defense — prompted Jason’s surprise: “Honestly, I thought they were small.” Sequoia’s new $200M seed fund isn’t much bigger than Harry’s ~$125M seed sleeve, and $1.5B for AI apps “doesn’t get you very far” when seed rounds price at $50M off a YC demo day. Jason runs the growth math: $6B minus fees is $4.8B — 22 $200M checks. “It’s not that much.”
• Rory concedes scale confers two real edges. First, optioning: a $10B colossus can decline to price the seed at all — it’s buying an option on the B, and “any option always trades higher than the intrinsic value of the asset,” so they can pay a quote-unquote stupid price. Second, the wall of news — which Jason no longer buys: “you can get coverage without writing $10 million checks.”
• The exchange worth keeping — Rory to Harry: if you win deals off the pod, “what you’re basically saying is you can win a deal not ‘cause I’ve grafted for 30 years and returned billions to my investors, but because I’m on a pod. If that sentence isn’t true, then 100% Marc Andreessen was right all along” — investing isn’t a media business, but a16z “tilted the table with their media strategy,” and Rory admires people who articulate a strategy and pull it off. His Red Army line: “quantity has a quality all its own… they got the 10 billion and they’re going to march it forward.”
• Two endgames: venture becomes investment banking — “Goldman Sachs, JP Morgan, and the rest of us are boutique players” — or big-fund returns disappoint and allocation tilts back to mid/small-cap venture. The only potential losers are LPs, if late-stage settles at 12-13%; and note the stapling — Harry: top brands often require $2 into other funds per $1 of early-stage access, across all vehicles. Rory: “bundling and rent extraction is fully understood.”

4. Likely Mercor Is an AI-Capex Bet

• The round: $350M at $10B led by Felicis — which led at $2B only eight months ago — on the believed-fastest run to $500M revenue (~17 months), and it’s GAAP revenue, not GMV. The business is RLHF human data for foundation labs: “five years ago we were labeling cats”; now “the models know how to label cats” and the product is PhDs posing frontier math, physics and bio questions — “you pound the model into submission.”
• Harry’s market map: three stacking pillars — talent acquisition → data acquisition (“you provide it to us, not we extract it”) → the implementation layer — each adding pricing power. The offsetting structural fact: “two buyers are 50%-plus of every one of these labeling providers’ revenue.”
• Rory’s checklist: “one enormous positive, great big honking market growing like a weed — and two negatives.” Margins: $500M gross but an unspecified large share goes to the doctors and mathematicians. Concentration: eventually the buyer says do it for 40. “I prefer to be OpenAI and have 800 million customers than to be Mercor and have two.” Jason adds the lived texture: an eight-figure renewal with a model company “was stressful AF” — nobody is shoveling money at vendors without thinking about margins.
• The underwriting fork, which Rory called Jason’s most insightful frame: pencil the end state — 3x from $10B means $30B, which at 5x revenues means $6B of training-data revenue, “a sobering thing… makes a man pause” — or ride momentum, where the $2B round already looks dirt cheap, “just like the Anthropic round at $67 billion earlier this year.” Net: “this whole thing top to bottom is one big-ass bet on AI capex hypergrowth” — and Jason reaches for Chuck Prince’s 2007 line about staying on the floor and dancing. “It turns out you should have sat down.”

5. Ramp at $30B and the asset class with no name

• Jason on the reported $23B→$30B round: “I don’t even consider it an up round. It’s not enough.” His math: a diluted 2% seed position goes from $400M to $480M — “it’s not doubling my position like a classic round” — and with 10%+ annual dilution “you might go from 22 to 30 and have the same price per share.” Impressive if the last round was at 10; “here I’m like, meh, that’s cute.”
• Rory’s partial defense: Ramp is capital-hungry by construction — “every dollar of revenue takes $5 of capital because you’re effectively recreating Amex,” so a $1B-revenue company probably carries a $5B balance sheet and lenders want a $1B equity cushion.
• His bigger reframe: there’s early venture, late venture, and now a third thing that needs its own name — “public stocks hiding in private.” Stripe’s most recent move was 91→110; you don’t expect mid-caps to gap up 3x. “If you’re doing companies at $30 billion pre, you’re probably not underwriting to an overall 3x on your fund — and you’re definitely not underwriting to a 30% IRR.” His two biggest changes of the last 3-5 years: AI, and this ultra-late-stage transformation of the IPO marketplace.

6. The Carta data: two-thirds of Series Bs don’t matter — pick, spray, or option

• The dataset: all 547 Series B investments from 2018 — ~35% below 1x, 18% above 5x, ~10% above 10x, and exactly one 100x (Figma). Rory was “very pleased”: it matches his fund model’s buckets almost exactly, and if you fill the good buckets the blend is 3.7x gross, 3x net to LPs. “The return was available to you.” Jason’s blog headline was the dark half: two-thirds of all deals returned less than 2x — “it makes you remember how much of this business is disappointment.”
• Rory’s taxonomy: three strategies — spraying, picking, and optioning — and skew just slightly worse than the average 66% dead-weight and “your math doesn’t work.” Harry’s supporting data: a16z did 72 seed option-bets versus 27 for the number-two firm; optioning is what lets you afford to spray.
• Against Harry’s “outcomes are so big now, just spray into the winners”: Rory runs the index test. If 2018 had ~1,600 seeds and you bought every one, a single 300x Figma still nets the vintage only ~0.2x. “The only people with a structural business where they can deemphasize picking is Y Combinator — the only mass-production seed business.” Even David Tisch, who “explicitly, respectfully sprays and prays,” picks 50 out of 1,600. “Poor me, I’m stuck having to pick.”
• Rory’s worst-outcome war story, from that morning’s mood: board member, sale fails to close, “you’re on the hook for shutdown costs and have to wire 300 or 400 grand just to pay severance” — which is why Jason now does everything on safes: “I’m only sort of committing, guys… if it doesn’t [work], I’ll write cheery responses to your monthly updates.”

7. Synthesia’s $3B question: exploding ARR versus “are you really an IPO guy?”

• The setup: Synthesia at a reported $150M ARR turned down $3B from Adobe and is raising well north of that. Rory checked the tape — they said $100M ARR when they took the Adobe money in April: “if they really have gone from 100 to 150 in less than 6 months, I can make this conversation really quick. There’s no way they should sell — that thing’s exploding.” Though he’d first tell the CEO: “now would be a really good time to look into your own heart, and if there’s something about the business that’s really worrying you, now would be a good time to share.”
• Jason would still advise selling — “I’m not saying it’s the right decision, but I want to be the guy that gives that advice” — unless the founder is 100% sure of a $10-20B public company. The disconnect: 3-to-10 is the difference between a 1x and 3x fund-returner for Accel, but for a founder at 10%, “what the hell’s the difference? I don’t think Jeff Lawson would be living a better life with half the money or twice the money.”
• His under-discussed question, asked twice of founders in 18 months: “do you really want to run a public company, for real?” At $3B nobody shows up offering five — “you have to assume it’s IPO or bust,” or “become the Collison brothers and compound private forever.” Rory calls it a genuinely excellent question, and a shame: the fact that being public is perceived as a pain “gets in the way of entrepreneurial ambitions… I hope it changes.”
• On DPI pressure, Harry and Jason split. Harry: non-tier-one GPs need to return cash. Jason asked his own LPs, including a stressed university endowment: “they’re like, no — we want you to play another card.” His rule: “there’s DPI pressure if you have a mediocre hand… what’s the point of returning 21% of your fund? That ain’t the full job, friends.”

8. Antitrust Adds Public Weight

• Harry’s field report from a Benioff Dreamforce dinner with ~10 public B2B CEOs: not stress exactly — “the weight… you could smell it out of the pores.” Mike Cannon-Brookes, whose co-founder calls him “the unreasonable man,” carries it visibly; the Ramp and Mercor founders don’t yet show “that anvil-level Wile E. Coyote weight.”
• The Roomba coda to the Synthesia debate: iRobot’s $1.7B sale to Amazon was blocked by Lina Khan’s FTC over — per Benedict Evans, tongue firmly in cheek — “an incipient monopoly in the house vacuum-cleaner marketplace.” The company raised $200M of debt to bridge, it’s spent, and bankruptcy looms. Rory: “an absurd decision at the time that only got more absurd… a horrible and unfair outcome for which the government is entirely responsible.” Jason’s lesson: when an acquirer pays a multiple only they can justify, “you got to take it.”
• Duration compounds the problem: Synthesia could face an 18-month wait; Wiz still hasn’t closed — “everyone’s like, you got a 2.5x in 6 months. No you didn’t. You got a 2.5x in 2 years.” At the margin it pushes founders to push on, or acquirers to the aqua-hire routine — “but when you’re buying the vacuum cleaner company, you want the freaking vacuums.”
• Harry’s conclusion is operational: early-stage managers should work secondaries proactively, because “a 4x fund over 17 years is the same as a 2.5x fund over 10.” Rory’s geek formulation of the objective function: maximize multiple subject to a minimum IRR constraint — say 25%: “a 30% IRR in one year isn’t as good as 25% for four years, but if that 25 starts dipping to 19, 18, 17, you’ve gone to a different place.”

9. Amazon Has a Bad Fortnight

• The tape: history’s largest white-collar layoff at 10%, cloud revenue share sliding from roughly half in 2018 to 38% today, Raymond James projecting Amazon’s AI cloud share at just 7%, plus a billions-in-damage outage. Harry’s needle: contrast Sergey Brin punching back in at Google — “maybe stepping down just before AI hit was suboptimal.”
• Harry’s defense-then-demolition of the timing: Jassy took over July 2021, “the peak of the last era, when products were frozen in time… it was the great time to go to Miami because nothing was changing.” But the grade is an F, not an A+: “if he sold his company, you get an A+” — the A+ goes to Salesloft’s Kyle for the December 2021 exit. And on the layoffs Rory is blunt: “Bezos would lay off half his company in a fortnight. It was the right thing. I don’t think you even care.”
• Rory’s diagnosis: retail over-invested for Covid and is now substituting robotics — painful but “just more of the same.” The real problem is cloud: core AWS didn’t fold, but all the new compute — 10x and 20x larger AI demand — went elsewhere. The hyperscaler scorecard: “Google was able to be relevant because they had their own model. Microsoft went and rented a model from OpenAI — and now the contract’s nearly up. And you did nothing, so you lose.” (Google, he notes, owns 14% of Anthropic.)
• His caveats: don’t overcorrect off one bad news day — retail dominance is intact because Amazon “can deliver … faster than anyone else on the planet” — and the fix is AI relevance without “doing what Oracle I fear is doing, taking on a whole bunch of subpar economic transactions.” On Bezos personally, the cynical read: with a couple hundred billion, “you’re not maximizing money, you’re maximizing psychic pain and joy.”

10. Lightning Round Tests Convictions

• Brex at $13B ($700M revenue, 50% growth) versus Ramp at $30B ($1B revenue): Rory takes the cheaper one — “I can’t help myself. For real.” Even Ramp at $100M vs Brex at $30-40M, “I’m still going to do Brex probably.” His homework assignment is the better answer: find the equilibrium growth rate that makes you indifferent — “it’s actually the core problem you face over and over in venture: how much extra do you pay for 70% growth over 60, over 50.”
• On a16z as best-performing mega platform, Rory outs himself as a convert: having invested alongside partners across the firm, “they are as good as the partners at smaller firms — I am a big bull on Andreessen now where I was not before.” Rory: they solved venture “from the founder back with operational excellence,” the returns “I’ve seen the numbers — they’ve been excellent,” and the churn is survivable because “everybody can have their thing… Marc and Ben sit at the top and life is good.” Harry’s kicker from Europe: “we don’t sit and be frightened of Index or Accel. We fear Andreessen coming into Europe. They’re the ones who beat you.”
• Anduril at $50B: Jason’s out, on self-knowledge grounds — “not interested in building weapons and stuff like that, just not my vibe” — even while conceding “there’s no better brag at a party than Anduril… one of the psychic benefits of investing is bragging.” Harry’s close: “the amount of self-knowledge embodied in that last two minutes is just stunning. The therapy is clearly working, dude.”