OpenAI Buys TBPN & Their Management Team Reboot | Mercor Hack & Why Now is the Time for Cyber

Summary

• Anthropic hit $30B revenue and passed OpenAI — up from $9B at the start of the year, 3.3x growth in four months, while still compute-constrained and Claude sometimes cannot finish chats. The WSJ-leaked kicker: Anthropic’s model training costs are a quarter of OpenAI’s, meaning it caught OpenAI in half the time at a fraction of the cost — Jason’s verdict: “that’s a double code red,” and “the investors in OpenAI got a much worse deal in the last round than the Anthropic ones did.”
• The pair trade writes itself: if both were public, hedge funds would short OpenAI and long Anthropic — “would you short OpenAI at 870 and go long Anthropic at 372? I’m not a risky guy, but even I would contemplate doing that.” Underneath it, Jason argues OpenAI’s ~$840B round was “barely real”: SoftBank tranched and borrowed, Amazon conditioned on IPO-or-AGI, Nvidia “almost all not money” — though Andreessen’s ~$11B was real and came in up front, and other hard dollars changed hands. And the advice to OpenAI employees: “sell it all at 820 the minute a tender comes.”
• OpenAI’s management reboot is rational but risky: COO Brad to special projects, CMO and CRO out, Fiji on leave, and ex-Slack CEO Denise Dresser handed all of go-to-market months after arriving — the “perfect LinkedIn” hire attached to tumult has, in Rory’s experience, “about a 30% chance of success.” The TBPN acquisition is the tell: a “vanity project and absurd” with zero editorial control, with outreach beginning in January and the deal closing last week — “there’s no way that deal’s going to happen today. It’s dead because of management change.”
• SpaceX confidentially filed at a $2T target — the largest IPO ever, up to $75B raised, xAI and X/Twitter involved, on $15-16B of 2025 revenue (125x). Sum-of-parts comes out far lower (last standalone SpaceX mark: $400B; xAI is #4 in models at best, burning $12B/year), but “he will will his 1.75 into existence for one day, at least one day” — 30% retail and underwriters with little leverage. The big three — SpaceX, OpenAI, Anthropic — will exceed every other IPO of the last 20 years combined.
• Cyber is the underpriced theme: the likely Mercor hack is “potentially death” because a hyperscaler executive described zero tolerance for material vendor breaches and data-labeling is “heavily fungible” — the pre-2023 one-free-pass era is over. AI turns hacking into “hacker farms of AI agents cranking 24/7,” so security stocks selling off on Anthropic’s announcement was “absurd” — “anyone who’s cutting back the security budget in 2026 is missing the point.”
• Low ACVs are the trap hiding in AI portfolios: OpenRouter at $1.3B on $50M ARR (up from $10M in October) is a 10/10 product that may be managing ~$2B of inference for that $50M — “the notional bips math doesn’t always translate to the real-world bips math,” and it “could be one of the greatest $200 million ARR companies.” Same logic blesses Supabase at $10B: agents now create more databases than humans, and the winner must parlay into multi-product before “the great crunch… and it will come in a couple of years.”
• The dodgy edge is the preview: a two-person company (“Medvy”) scaled to $1.8B revenue selling GLP-1s with AI deepfakes and edge-of-fraud affiliate tactics at $300-400 per delivered lead — “I both think this is fraud… and the future.” Like SEO farms that built DigitalOcean and Zapier, agentic hyper-personalized marketing goes mainstream within a year or two; the investable play is the software that helps “the normies tool up.”
• Venture housekeeping: Doug Leone returns to Sequoia — Jason reads it as calming LPs mid-raise, Harry as competitive necessity against Founders Fund and Andreessen (“he closes the deal”), consensus word: gravitas. And YC expelled Delve after AI-faked audits and, especially, for forking another YC company’s possibly batchmate Sim Studio’s code and claiming it as their own — “you broke the code of the West, you’re out.”

Deep dive

Anthropic Signals Code Red

• Jason fell out of his chair: $30B in revenue, up from $9B at the start of the year — 3.3x in four months. Salesforce took 25 years to get there; “Anthropic got there in five, but maybe they really got there in three depending on how you count.” And they’re still capacity-constrained — Claude visibly can’t finish chats — so “the estimates that we were just looking at 2 months ago just look incredibly wrong at this stage.”
• The Wall Street Journal leak Jason keeps returning to: Anthropic’s training costs are a quarter of OpenAI’s — maybe because they don’t have to do video, images, and a lot of consumer stuff. Catching OpenAI in half the time at a quarter the training cost, with management turmoil at the competitor: “that’s a double code red… it really feels like the investors in OpenAI got a much worse deal in the last round than the Anthropic ones did.”
• Jason’s Uber/Lyft comparison: Uber out-accelerated Lyft while burning every dollar it had; Anthropic is out-accelerating while being more efficient — same-or-better gross margins, lower training costs. Rory calls it “a bad fact pattern” if you’re the other guy.
• The OpenClaw move fits the same picture: revenue is exploding and they’re still compute-constrained, so you allocate capacity based on money — keep token-devouring OpenClaw agents out of fixed-price base plans, price tokens closer to value (as OpenAI is doing by deemphasizing video). But gently: “the thing you have in your favor with any digital good is the complete certainty that prices per token go down over time” — you still want people addicted.

OpenAI Round Draws Scrutiny

• Jason’s dissection of OpenAI’s raise: Andreessen’s ~$11B came in up front and real; “the SoftBank money comes in tranches. They have to borrow money to pay it. The Amazon money is tranched in part on IPO or AGI. And the Nvidia money is almost all not money” — offsets and compute. “That’s a sign of, like, classically at least, barely getting the round done.” Rory’s pushback, worth keeping: “a bit harsh” — they tacked on another $10B of cold hard cash, enough hard dollars changed hands to represent a bona fide price, and Anthropic does its own round-tripping on compute and distribution.
• Both agree on the relative call: Anthropic at $370B “feels a little more comfortable” than OpenAI at $820-840B. If both were public, “there would be a bunch of those New York hedge funds shorting OpenAI, longing Anthropic… Would you short OpenAI at 870 and go long Anthropic at 372? I’m not a risky guy, but even I would contemplate doing that” — same rough revenue, better trajectory, a management team, half the price, with overall AI risk diversified away.
• For OpenAI employees sitting on tens of millions at the $820 mark: “Sell it all at 820 the minute a tender comes.” The tempered version: “when the liquidity window opens, take it seriously, cuz it mightn’t open again for a while” — all your brethren already have.
• Harry’s inversion game — Anthropic at 850 or OpenAI at 380? One guest takes OpenAI at 380 on one proviso: sit down with the board and ask “what are you going to do about this?… just stop screwing around and focus” — the consumer asset is still dominant, Sam deserves credit for aggressive compute purchases now that constraint is real in ‘26 and early ‘27, and the mission is simple: monetize consumer and ship the Codex competitor. Rory won’t touch it at any framing: “I’m just not into the tumult… not into a non-deeply-technical founder leadership… I want someone Dario or smarter technically running these companies.”

Management Reboot Brings Risk

• The scale of change: Brad the COO moved to special projects, the CMO stepping down for health reasons, the CRO out, Fiji (head of apps) on a short leave. The defense: “you don’t just sit there and make no changes on the team when your competitor over the last 6 months has radically changed the competitive posture… it just makes sense to reboot the team” — though “calling code red 3 months ago didn’t magically change the trajectory here.”
• Rory’s sharpest warning is about the one additive hire: Denise Dresser, CEO of Slack via Salesforce, in the door only a couple of months and handed “basically everything go-to-market.” Bringing in “the person with the perfect LinkedIn,” giving them a massive portfolio, and attaching them to something in tumult “has about a 30% chance of success, just roughly” — “there’s not a lot of time for the get-to-know-you tour.”
• The Oscar Wilde framing, delivered with genuine sympathy for the illness cases: to lose one parent “might be an accident,” “to lose both parents smacks of carelessness… you are getting to the stage of carelessness here.”

TBPN Acquisition Draws Fire

• The acquisition is called start to finish: “I thought the acquisition was just insane.” You don’t launch focus decks and no-more-side-projects decks and then, within a week, do the most side-project thing imaginable. OpenAI is “the most known company on the planet perhaps other than Apple” — Sam meets any world leader he wants — so “if you were to pick the one company who doesn’t need media attention and does need to focus, it would be OpenAI.”
• Jason’s rational bull case: some acquisitions run on autopilot and cost little senior time, and for a profitable B2B company under brutal margin pressure, buying a media asset at scale “is a way to turn a balance sheet into marketing” — the logic behind the Barstool deal that “almost worked but failed.” His own caveat: “every profitable public company should do a deal like this, of which OpenAI is neither.”
• Harry’s attempted justification is the interesting middle: OpenAI has “consistently messed up crisis PR” (lemonade stand, the Anthropic adverts), so maybe TBPN is “vibe maintenance” — except they have no editorial control whatsoever, “so they have zero benefits.” The history lesson is that “owning a media asset in very big takes way more time than you think for way less money than you expect. See Jeff Bezos for details.” If you can’t control the story, “hire a better storyteller.”
• Harry’s meta-lesson is for founders to hear: outreach was January, when the world was different; it closed last week. “There’s no way that deal’s going to happen today. It’s dead because of management change.” So default yes to a good deal — the odds the VP who sponsored it is still there in 12 months with unchanged priorities “approaches single digits.” A guest’s counter from selling a company to GE: everyone was ready to sign and Jack Welch looked at the numbers and said no — the big boss should still kill late-stage deals. The Adobe frame: every executive got one big chip and one small chip a year; TBPN was a small-chip deal that took five minutes.

SpaceX Targets $2 Trillion

• Harry’s framing of the moment: SpaceX plus OpenAI plus Anthropic at IPO will exceed every other IPO of the last 20 years combined — which he found “almost depressing,” with Sam Lessin’s “OpenAI doesn’t even matter” rattling in his head: “maybe nothing we’re doing matters.” Rory’s answer is power-law psychology: “you can have the third best outcome in venture history and be only 1/10 as large as the largest outcome… if you’re going to let it in your head that it’s not two trillion dollars, then you’re doomed and you’re just going to need therapy.”
• The terms: confidential filing, up to $75B raised, largest IPO ever past Saudi Aramco, xAI and X/Twitter involved; 2025 revenue $15-16B with $8B of EBIT — 125x revenue at $2T.
• Rory’s valuation forensics: any sum-of-the-parts comes out far lower, and the gap “is all Elon premium” — which is visibly fading, with Tesla down significantly year-to-date and JPMorgan out with an actual sell and a 60% decline prediction. The walk-up matters: a meaningful transaction at $400B less than 12 months ago, a smaller $800B secondary that may never have happened, then the merger that valued SpaceX at $1T to mark xAI — negative $12B in cash flow, “number four in the model LLM space at best” — at $250B. “It is worth remembering that the last time the useful asset was valued on a standalone basis, it was worth $400 billion.”
• On the mechanics of IPO day: Elon said publicly on X that $2T was too high — but whatever his number is, “the underwriters are not going to be able to argue with him for more than 5 minutes,” 30% retail will whip up demand, and “he will will his 1.75 into existence for one day, at least one day.” Rory’s close: “only in the long term are markets weighing machines; in the short term they’re voting machines” — and index funds are buyers anyway, since it hits QQQ 15 days after the IPO.

Doug Leone Returns

• Doug returns in an investing capacity — leadership stays with Pat and Alfred. Jason’s read from a distance: “it feels like something to calm the LPs” amid enormous fundraising — “LPs say that they’re looking at the new generation… but they are comfortable when the old leadership is still actively involved.” You don’t bring someone back just for an hour of wisdom on Slack.
• Harry’s pushback from time spent with LPs: “the insatiable appetite from LPs for Sequoia has never been more prominent” — this is about competition, with Founders Fund riding Anduril and SpaceX tailwinds and Andreessen more attractive than ever. “Doug is the ultimate winner of deals” — whether it’s Christian Hacker at Trade Republic or the Wiz team, “he closes the deal.” Jason’s tease: “the kids at YC have heard of Doug Leone or even know how to spell his last name? I doubt it.”
• Rory’s balanced take: sensible, not earth-shaking — continuity for LPs, the firm, and entrepreneurs, plus “he’s a damn good investor” by the hiring test (“do you think the next check Doug Leone will write will be better than a check one of us will write?”). And an honest edge: needing this move means the first transition “wasn’t that successful… it just felt a little janky late last year.” The word everyone settled on: gravitas.

Delve Leaves YC

• Jason separates the two sins: making up audits with AI — “we’re going to find more portfolio companies did that” — and the especially serious issue of taking another YC company’s, possibly a batchmate’s, open-source code, not attributing it, and claiming it as their own software to a customer. “We’ve all thrown a few things in the cloud and pretended we did the work… but you can’t hand-wave that one away.”
• Rory’s base-rate defense of YC: 200 companies a quarter is 800-900 founders a year; he indexes that to America’s roughly 1% rate between felonies and misdemeanors and says “statistically… there’s eight of them that within the course of their life commit some kind of crime. It’s going to happen. So, first of all, no drama there.” The only enforcement that works isn’t a priori policing — it’s Old West justice: “you broke the code of the West, you’re out.”
• The open question left hanging over Insight’s $32M check: “should you really buy compliance software from 21-year-olds? That’s an interesting comment.”

OpenRouter Faces Scale Questions

• The business is an interface between app builders and 50-60 LLMs with dynamic routing, charging ~5-5.5% of model spend — “kind of in that Stripe/Twilio business model,” and since Twilio books 20-40%+ gross margins, “maybe there is actually room for margin expansion.” Jason is a 10/10 super fan: market leader, cheap, easy — “it ain’t worth switching for a tiny amount more basis points.”
• Priced at $1.3B on $50M ARR, up from $10M in October — but the worry is that they may be managing ~$2B of inference just to book that $50M. The bull math is tighter than it looks: OpenAI-plus-Anthropic 2029 projections imply maybe $300-400B of enterprise API; if 10-20% goes open source, that’s $40-80B, of which 5% is $2-4B — “and that’s 100% of the market.” Bonus irony: the leaderboard is all Qwen and Kimi — “the Chinese Communist Party is effectively subsidizing the American small independent software vendor. God bless them.”
• Jason’s RevenueCat lesson (he was first investor): ~50% share of paid mobile apps at a half-percent-to-1% net take, up 40% last month on AI, and only now a path to $1B — “the notional bips math doesn’t always translate to the real-world bips math… OpenRouter could be one of the greatest $200 million ARR companies.” He’s nervous about low-ACV AI winners generally: “our investments are being flattered by high ACVs right now” — the Lagoras and the Harveys. Harry’s Visa counter: 15-20 bips on every dollar every human spends “turns out to be remarkable.”
• The operating advice is not to overthink commoditization — during the build-out, anyone solving a rate-limiting step attracts capital, but “the great crunch does come, and it will come in a couple of years,” so parlay into adjacencies (inference, hosting) before it does. Harry goes further: run the AI Rippling playbook and go truly multi-product now — “you literally probably have to build five distinct products to get to that billion.” A Stewart Butterfield-esque reluctance to expand? “I would be less excited to hold stock.”

Supabase Rides Agent Growth

• Jason calls it “an AI tailwind to the maximum”: a 2020, pre-AI fork of Postgres that turned out to be exactly what every agentic product needed — a database that self-deploys without humans. Replit went with Neon (which Databricks bought), but everyone else standardized on Supabase, which white-labeled to Lovable, Emergent and the rest; Replit now attaches a database to every app whether you use it or not, and Supabase monetizes them all. “More databases are being created by agents than humans… why wouldn’t you want to invest in the leader?”
• Rory’s playbook: this is MongoDB’s SaaS-era arc on fast-forward — be the database of the vibe-coder and agent era for 2026-28, so when the Lovables consider in-housing you can say “every developer on the planet uses us. Every agent framework supports us. Why would you do this?” And the movie’s ending is already written: in 10 years someone will say “those legacy Supabase products, they’re almost as bad as MongoDB… but that’s just the movie, and this is Supabase’s time to crank.”
• The durability debate is signal: Harry says “we’ve given up on worrying too much about intellectual durability in these investments.” Rory’s reframe — it’s not that we don’t care, “you just don’t have the luxury”: per Brian from Andreessen, “the barrier to entry is speed.” Five leave the gate, one won’t stumble; second-order moats come downstream. “If you stumble, you lose.”
• Which triggers Jason’s rant on VCs as enablers: too many are running “a pre-AI enabler playbook, where when folks fall behind a tick or two, you see kumbaya activity instead of code red activity” — his example, a nine-figure-revenue company facing AI disruption whose new board member emails “great job, guys.” “Enablers can enable a death spiral that you feel good about” as you approach the event horizon. Rory’s synthesis: not founder-friendly but founder-honest and fact-based — know the direct competitors hands-on and the last three win-losses, or “you’re just cosplaying a board member.”

AI Raises Cyber and Marketing Stakes

• On the likely Mercor hack, with Meta pausing as a customer: Jason recalls a hyperscaler executive telling him, even over a minor vendor incident, “there’s not much higher on our list… we have no tolerance.” Since Rory says data labeling is “heavily fungible” — the largest customers use all the vendors — Jason’s verdict: “I think it’s potentially death… through 2023 you always got a pass once as a vendor. I just don’t know that you get a second one here.” Rory is softer: it was a ransom crew (likely Lapsus$ — “decent honest criminals,” jokes Harry, who asked if he could invest), and the likely outcome is losing revenue and time, spending heavily on defenses, and earning the way back slowly. “Hopefully not fatal.”
• The bigger call: Sam Altman said this week that massive AI-driven cyberattacks are coming, and Jason thinks most B2B companies are sitting ducks — “a hotchpotch of open source and other products that are barely monitored” (the likely LiteLLM weakness that hit likely Mercor is everywhere). Remember Gainsight offline for a month, Drift “permanently destroyed” — and his old CTO’s line: “the only reason we’ve been hacked is no one cares about us.” With AI, someone cares about everyone: “It used to be hacker farms of people in the Philippines, Russia. Now it’s going to be hacker farms of AI agents cranking 24/7. It’s going to be miserable.”
• The tradeable conclusion, per Rory: security stocks falling on the Anthropic announcement “was absurd” — “anyone who’s cutting back the security budget in 2026 is missing the point.” Attacks gain the Red Army property (“quantity has a quality all its own”), everything that matters is now online, and only two errors kill these companies: the app down for a long time, or the app grievously hacked. That’s where the money has to go.
• Then the mirror image: “Medvy,” a two-person company at $1.8B revenue selling GLP-1s via AI deepfakes, misrepresented doctors, and edge-of-fraud affiliate tactics at $300-400 per delivered lead. Jason holds both thoughts: “I both think this is fraud… and the future. We’re watching the future.” The precedent is SEO — a dark art that built DigitalOcean and Zapier before every corporate normie learned it — and within a year or two “only the creakiest companies will be doing marketing and advertising the way they do it today.” Rory’s stolen insight of the day: the correct play last time was “to be the software provider helping the normies tool up… Now I’ll go look for those companies.”