OpenAI Kills Sora & Hits $100M ARR on Ads | Oura Going Public & Whoop Raises at $10BN

Summary

• Anthropic printed $6B of revenue in 28-day February — “more than Databricks has done in their entire lifetime” — while accidentally leaking ~3,000 unpublished assets around Claude Mythos, reportedly a 10-trillion-parameter model “extremely good at detecting cyber issues,” withheld because of how powerful it is and set to cost far more to serve and buy. Anthropic blamed human error; Jason’s math says get used to it — agents working 1,000x faster with a tenth the mistake rate still produce 100x more leaks, daily and weekly in the agentic era.
• The Mythos-driven cyber selloff (CrowdStrike/Palo Alto/Zscaler -6%, Okta -7%, Tanium -9%) is “back-ass backwards”: Jason’s contrarian call is this is the golden age of security — threats exploding as unread, agent-built code ships everywhere — so “everyone should be a rocket ship,” and the fact the market can’t see it means “we’re in a true panic, which is hard to predict a bottom.” Expect CrowdStrike and Palo Alto to swoop agent-security winners for $300-500M before it’s certain.
• OpenAI shooting Sora “in the head” is the bigger tell: single-digit millions of revenue against enormous compute means the accountants have walked in — “we have a scarce resource here… let’s devote this compute to the people who can pay the most for it.” Harry calls it “a huge own goal — our strategy was wrong”; Rory says only two existential bets remain: ads ($100M ARR is “in the noise” — $20B within a couple of years or “they’re not even in the game,” $50-70B to grow into the market cap) and a competitive coding/enterprise model.
• The WSJ story on why Dario left OpenAI reads as a permanent drama tax — nobody would work for Brockman, Sam telling everyone they were in charge, Dario demanding to report only to the board. Rory: that much board and senior turnover “is probably the highest warning signal that you can have as a board member about how your CEO’s doing.” Harry’s fix — buy Sierra, install Bret Taylor as day-to-day CEO, make Sam the fundraiser — Rory would favor as a board member “but I’m not going to say that publicly.”
• SoftBank’s $40 bridge loan to buy OpenAI stock leaves the group ~1.5-2x levered — a 30-40% decline in its two big assets (OpenAI and ARM) “wipes them out.” Rory: “He’ll go as deep as they let him… You haven’t lived till you’ve seen an 85% decline in an index.”
• ARR is fiction at the margins: Anthropic ~$19B run-rate reported gross vs OpenAI ~$25B net; the same tokens get resold and re-recognized down the chain (“how many times do we get to resell these poor little tokens?”); Emergent Labs’ “0 to 100M in 8 months” likely counts $0-first-month trials as $240 of ARR — yet its product beat Replit, Lovable and V0 on Jason’s test. Tranche rounds compound the game: followers pay the full headline billion while the lead’s blended basis is 600 — “that’s the price of being cool.”
• Oura heading public and Whoop raising ~$500M at $10B validates Jason’s health-data call a year early — but “it’s not ServiceNow ARR,” people switch like Peloton. Rory’s retort: “get the f* over it… capitalism is hard” — and with durable software trading below the S&P 500, maybe nobody cares where the R comes from anymore.
• Manus’s founders being unable to leave China kills the playbook — “every Chinese founder… is going: if I do this deal, I am never going home again”; Benchmark’s risk “bounced the right way,” but “you just can’t do the next one.” Same theme stateside: Jurvetson decamps to Incline Village (“that’s real coin” — $260M on a $2B gain), Washington passes a 9.9% millionaire tax, and Rory warns the marginal dollar lost funds “marginal services to marginal people,” not teachers.

Deep dive

1. Mythos leaked itself — breaches go daily in the agentic era

• The setup: Anthropic’s 28-day February did $6B in revenue, more than Databricks’ lifetime — but Harry’s pick for the week’s real story was the accidental leak of ~3,000 unpublished assets around Claude Mythos: “a 10 trillion parameter model, apparently,” being withheld “because of how powerful it is.” The leaked memo’s claims: way more powerful, way more expensive to serve and to buy, and “extremely good at detecting cyber issues.”
• The irony was acute — a model billed as amazing for cybersecurity leaked via a cybersecurity lapse — and Anthropic “blamed human error.” Rory’s line: “We may be at the stage where we throw the humans under the bus, not the AI anymore, which I think at some level is pretty terrifying.” His guess at mechanics: staged CMS assets for a future launch pushed public early, like the UK budget press release.
• Jason’s acceleration thesis: agents are goal-seeking and work 1,000x faster, so even making mistakes 10% as often they produce 100x more mistakes — expect source code, data and PII leaks “daily and weekly in the agentic era.” His Adobe anecdote as the trade-off: source code was the crown jewel and banned from the cloud, and their on-prem tooling meant a month per release versus 60 a day now — “there’s trade-offs there, and I’ll take them.”
• From the leaked Kairos material, Jason flagged two things: an always-on background assistant and agents that “sleep, wake, and self-resume without any prompt” — the token-devouring autonomous future “open claw” hinted at, with agents debating your portfolio 24/7 while “we give up all of our personal freedoms and autonomy as part of it.”

2. Sora shot in the head — “our strategy was wrong,” and only two bets remain

• Harry framed a chasm: Dario’s “progression of force” versus “a faltering, confused and dazed OpenAI wandering around the product desert trying to find some water.” Rory’s pushback — worth keeping: “narrative is overdone on both sides”; killing Sora is embarrassing four or five months after hyping it, “but if it’s a mistake, give them credit for at least saying it’s a mistake.”
• Harry went further: Sora made single-digit millions and was reported as consuming a million a week, “which actually sounds way too low… It must have consumed billions.” Killing it is probably right, “but man… this is a huge own goal. Our strategy was wrong” — the all-in consumer bet on owning image and video, which Anthropic never even attempted, is in retreat.
• Rory reads it as healthy discipline: “the economists, the accountants have wandered into the room… let’s devote this compute to the people who can pay the most for it.” Video generation is extraordinarily compute-intensive with minuscule revenue; codegen is orders of magnitude cheaper with real dollars attached. “Net net it’s a positive. Better late than never.”
• On ads, the math is existential, not optional: consumer conversion runs ~5% off ~500M uniques — a $10-15B business — and most consumers won’t pay $20 a month, so ads must work. The $100M ARR everyone’s ragging on is “in the noise”: Facebook does ~$200B on a $1.7T cap, Google $260B+ on $2T — “if these guys aren’t doing 20 billion within a couple years, they’re not even in the game,” and growing into the cap takes $50-70B. That plus “oh my god, we should have done coding all along” are “the only two things they’re doing and the only two things they should be doing.”

3. The drama tax — Harry would rather be at Anthropic

• The Wall Street Journal’s account of why Dario left OpenAI: Greg Brockman couldn’t recruit — Dario and his sister wouldn’t work for or talk to him; Sam told Dario he was the boss while telling Ilya and Greg they could fire him at any time; Dario would stay only reporting directly to the board. Harry: “I’m exhausted… a company organized like [Anthropic] — same founders, same goals — is just going to out-execute someone with that level of drama.”
• Harry’s provocation: should OpenAI buy Sierra, install Bret Taylor as day-to-day CEO, and make Sam the fundraiser? Rory: “I think you’re right, Harry, and I would favor that as a board member, but I’m not going to say that publicly… I am too unimportant for Sam to even give a f* about.”
• Rory’s board-craft takeaway: that much board-level and senior turnover over an extended period “is probably the highest warning signal that you can have as a board member about how your CEO’s doing” — cut the drama, build a team, “call a shot and play it for more than six months at a time.” Jason on CEOs load-balancing talent versus teams rowing together: “the backside of Pluto and the front side of Mercury.”

4. Masa’s $40 bridge — wiped out at minus 30-40%

• On SoftBank borrowing $40 to buy OpenAI stock, Rory: “He’ll go as deep as they let him. If they give him another 20, he’ll borrow that, too.” SoftBank Group runs roughly 1.5-2x levered on equity, so a 30-40% decline wipes them out — the equivalent, he says, of a $1 venture fund borrowing $1.8 billion and investing the lot.
• The two big assets are the OpenAI position and ARM — “amazing companies… easily imaginable of both of them declining 30%.” Masa’s qualification is survival: he lived through 2002, when “the Nasdaq went down 85%. You haven’t lived till you’ve seen an 85% decline in an index” — plus the ~$12B WeWork lesson.
• Jason’s half-defense: real estate funds max out leverage by design, and “if venture had access to more debt, we’d all load up.” Rory’s rebuttal: real estate can lever because cash flows are predictable — “the SoftBank portfolio is more like Jason’s fund than it is a real estate fund.”

5. The cyber selloff is “back-ass backwards” — this is the golden age

• The damage: CrowdStrike, Palo Alto and Zscaler down 6%, Okta and Netscape (likely Netskope) down 7%, Tanium down 9%. Harry’s dissection: Mythos helps at the code-development stage — application security and secure code review may have to incorporate it or go redundant — but it isn’t real-time perimeter defense and it isn’t Okta’s single sign-on. Selling all of them off is “just baby with the bathwater,” a knee-jerk rather than anything thought through.
• Jason’s tradeable contrarian call: “this is one where it’s just back-ass backwards… in the agentic world, this is the golden age of security.” Threats are going up orders of magnitude — agent-built apps nobody reads, corners cut everywhere, “you can’t stop the rogue engineers that vibe coded something that accessed your data.” “Everyone should be a rocket ship… the fact that the market doesn’t see it shows we’re in a true panic, which is hard to predict a bottom” — see Robinhood down ~10% merely on Elon routing a tender through E-Trade.
• Rory agrees with the caveat that it mightn’t be the old guard that wins: nobody yet knows how to defend against agents given full root access and told to work overnight, but the threat is adoption velocity times solution power — and the smart incumbents “know damn fine” to “spend your 300 million bucks, your 500 million bucks, and just swoop up the winner” long before it’s certain.
• The meta-observation on Anthropic: they are “masters at selling fear” — releasing Mythos first to CSOs with the message “this is the new terrifying weapon we’ve invented. Please give us a million dollars and we’ll let you defend yourself with it also. Great marketing.”

6. ARR is negotiable — resold tokens and the $0-a-month trick

• In fairness first: OpenAI and Anthropic define ARR sensibly — trailing four-week average × 13, actual realized revenue flowing through the system. On that basis Anthropic runs ~$19B and OpenAI ~$25B. But the methods diverge on partner revenue: OpenAI reports net of Microsoft’s cut, while Anthropic books gross through AWS and hands ~$20 of every $100 back as cost of sale.
• Then the double counting, per Michael Cannon-Brookes’ point: Cursor and friends resell the same tokens and recognize them again — Jason: “how many times do we get to resell these poor little tokens?” His demo-day reductio: at zero gross margin, “everyone in batch 0001 has a million ARR its first week because we just resold our tokens to each other.” Rory’s punchline: “until we all have to get profitable, all this can continue.”
• Jason’s own evidence from inside: a portfolio company over $100M sends him three revenue numbers a month, ARR the smallest, “asterisks and daggers” everywhere — “I cannot for the life of me tell what an ARR is in 2026.”
• On Emergent Labs’ “0 to 100M in 8 months” homepage banner: the likely mechanism is recognizing a $0-first-month, $20-thereafter trial as $240 of ARR instantly — “a sus practice… I don’t love that kind of gray art,” though “we’d probably have to shoot half our portfolio companies” that do PLG AI the same way. Yet Jason actually tested the product: it beat Replit, Lovable and V0 on his saster.ai homepage test — top 10% of vibe-coding apps, “a legit business,” an order of magnitude better than Make, in a field where most wrappers lean on “Claude code does 90% of the work for you.”

7. Tranche rounds — paying 50% more to be cool

• Harry went off on one: “X raises from Sequoia at 5 billion — trust me, Sequoia got in at one,” with the tranches clubbed together and only the headline announced. Rory’s mechanics (amusingly, “neither Claude nor GPT” knew the modern meaning): a first close at 250 pre and a second at a billion pre, blended basis ~600 — silly but rational optics, except “you better be damn sure your next round you’re at 1.5 billion, otherwise you’ll have the optics of a down round.”
• The invidious version is multi-participant: the lead takes the whole 250 tranche plus half the billion while less-marquee followers pay the full billion at the same moment. “I don’t believe there’s right or wrong in money, there’s just money” — but the follower has to look in the mirror: “that’s the price of being cool… I’m paying 50% more because I just can’t access that deal.”
• Jason’s zoom-out: valuations are “utterly gamified” — YC’s cheaper-before, higher-at, +20-30%-after demo-day pricing is institutionalized — and “I just don’t think raising at five or eight billion when you’re at 80 million or 100 million of suspect ARR is the most exciting accomplishment in the world… I’m going to send a few thumb emojis on the email, but that’s about it. They’re all fake anyway… These are not public companies.”

8. Oura and Whoop — real businesses, not ServiceNow ARR: “get over it”

• The news: Whoop raised ~$500M at $10B and Oura is heading toward an IPO — vindicating Jason’s prediction that 2027 would be the year for human health data and longevity, “and it looks like it might even be 2026.” Rory: this is not an AI-envy story — standalone products with a clear consumer value proposition that are “not going to be cloud-coded on Friday.”
• Jason’s hedge: these are recurring-revenue products with fairly expensive subscriptions, “exciting until, like Peloton, they aren’t” — he runs 5 miles a day, 360 days a year, and would switch to a better device tomorrow. “It’s not ServiceNow ARR.” And his Thiel-shaped meta-anxiety: “competition’s for losers… if these are unmonopolizable markets, are they good ones for venture?” — though “a better landing place than investing in bars.”
• Rory’s retort, verbatim: “Let me be direct, get the f* over it. Not every business on the planet has 5-year design debt” — Coca-Cola fights Pepsi daily, bars re-earn every drinker — “it turns out capitalism is hard. If you want to make 10 billion in value, you got to deliver value to your consumers.” Peloton, he argues, was a COVID demand pull-forward like Zoom, not a competition story; GoPro died of saturation (Jason mildly disagrees — a step function left them behind). And hardware is genuinely hard: “we’ll be done with wearables long before we’re done with sneakers” — Allbirds just sold to AMAX (as heard) for $39M.
• The flip that matters for boards: durable software now trades below the S&P 500, so Harry asks whether anyone still cares what type of R it is — “we’d be in board meetings where you would torture companies so they would have more ARR… that seems archaic today.” Rory: “you should conform your company around your customers and your model, not your VCs.”

9. Manus founders trapped — Singapore-washing is over, and capital keeps moving

• The facts per Jason: Manus was China-founded, redomiciled to Singapore, Benchmark invested, and Meta acquired it — his understanding is the deal closed and the money moved — but two key founders were in or summoned to China and can no longer leave, the Chinese government taking “a dim view” of talent leaching overseas. Jack Ma is the precedent, “and there are significantly worse consequences than that.”
• Harry’s verdict: “this whole Singapore-washing thing is over” — venture talked itself into the cut corner, it “bounced the right way for Benchmark and friends,” but “you just can’t do the next one.” Rory: every Chinese founder contemplating the move now knows “if I do this deal, I am never going home again.”
• For Meta, $2B is small and the product runs smoothly short-term, though “when you lose your founders, you lose your heart and soul of your company.” Harry hopes Meta brokers “an amicable end… you don’t want to leave people you just acquired in limbo,” and the sober coda: “the worst thing that can happen is not just that you lose your money. There are worse outcomes than that.”
• The same mobility logic hit California: Jurvetson, at his “DPI moment” after riding SpaceX and Tesla, bought Incline Village’s most expensive home — saving 13% on realized gains plus 5% on all gains if the wealth tax passes. Rory wouldn’t uproot for income tax, but for a one-time $2B SpaceX gain: “Honey, for the next 2 years, why don’t we live in Incline Village 165 days… we will save $260 million. That’s real coin.”

10. Killing golden geese — and VCs as bit players

• Washington just passed a 9.9% millionaire income tax (Howard Schultz already left); California’s billionaire-tax math “assumed massive amounts from Larry Ellison, who’s been gone a half decade.” Jason’s worry isn’t politics but the tipping point: “they’re all leaving — everyone that doesn’t work at OpenAI and Anthropic,” and his old line that founders leave after the Series B “may happen by show 100.”
• Rory’s argument for the other side of the table — since “being mean to a billionaire is actually a feature”: taxing without attention to collectability shrinks revenue, and “somewhere in Sacramento, someone will zero out a line item… it won’t be payments to the teachers, it won’t be payments to the firemen, it’ll be marginal services to marginal people that your crass stupidity and desire to make a political point has ended up costing them money.”
• Separately filed under candor: Epic Games quietly laid off 25% — over 1,000 people — with what Rory called a no-BS announcement (“we sell less stuff, we have less people”), no AI story attached; Rory sees entertainment as the preview of every labor market — “the Fortnite circle coming for everybody,” even Fortnite itself — with the WSJ documenting Hollywood employment in permanent decline.
• And the week’s most crystallizing tweet: asked on Jack Altman’s Uncapped about Ron Conway’s claim he’d helped Cloudflare navigate significant issues, Matthew Prince answered, in effect, “well, maybe — I don’t remember any of it.” Not mean, says Jason — it “crystallized the whole value-add idea in a single tweet.” Rory’s proof: business biographies barely mention VCs — money in, money when backs are to the wall, hire/fire the CEO, “anything after that is at best an assist… We’re not the stars in the drama. We’re bit players who get well paid for our part.”