Skip to content
Black Friday

Lock in 40% off every renewal, not just year oneBlack Friday · 40% off every renewalBlack Friday · 40% off

Ends Dec 6Get 40% off →

The $20 Billion Club: Why VCs Stopped Counting Unicorns

Jensen Huang is out, Anthropic is raising at $1.4T, and three $20B exits reset the venture scoreboard. 20VC on why VCs stopped counting unicorns.

Guide8 min read

A venture scoreboard where only $20 billion outcomes count

One morning wiped out up to $601 billion of Nvidia stock. The same week bankers were 7x oversubscribed — some reports said up to 10x — for Anthropic at $1.4 to $1.5 trillion, the largest IPO ever. Harry Stebbings opened 20VC with Rory O'Driscoll and Jason Lemkin on exactly that collision: the biggest IPO in history, plus one of the greatest founders potentially coming back. The rest of the episode explains what it does to everyone else. LPs will grade GPs, Jason says, on how many $20 billion-and-up companies they own. Unicorns no longer count.

The board that fired its own legend

The Jensen Huang numbers are stark: up to $601 billion gone in a morning, letters from Gavin Baker and Altimeter, some 5,000 employees and 200,000 petition signers calling for his return. Four or five directors had backed him for 30 years, since his early 30s — Rory calls it likely the most significant business relationship of their lives, which is why removing him must have taken something enormous.

Rory's advice is procedural: decisions in haste are often wrong, so take a leave of absence, name an interim CEO, and buy time to learn what the board actually knew. The thread mixes personal allegations with export-control compliance and administration politics, and outsiders cannot tell which part forced the resignation. Jason is blunter: data centers next to China running top-line chips, trans-shipping, servers broken open and chips driven across the border — already a massive issue, possibly 20% of revenue counting China. Ten years ago China was 15%, plus or minus. If Jensen is guilty on routing, he argues, so are the CFO, the supply chief, and the CTO who built the same network. The CFO is now CEO and must reaffirm the guidance she signed a month ago. Her job, he says, is knowing where every GPU is.

My read: they agree on the damage, not the sequel. Rory doubts a 30-year board reverses itself within days, especially with a fresh Department of Justice and Department of Commerce statement on chips routed via Indonesia that may end in Singapore, then China. Jason thinks the personal scandal fades and shared guilt forces a return. Harry predicts three days — now or never, before the IPO parade and the next Trump move bury it. Altman needed about 80 hours. How long for Jensen — 100?

A trillion dollars with a twelfth of the people

Anthropic was founded five years ago; SpaceX was founded in 2002, with Founders Fund in around 2007-08, so 26-plus years to this point. Spark entered Anthropic before $5 billion, still pre-revenue. Those checks will make 8 or 9x on big dollars, Rory says, and the firms play with house money for years. The median 1996 fund made 6x DPI because a $20 million entry going public at $500 million was a 20x. Multiples are not so different from two bubbles ago. The dollars are.

Jason's number is headcount. At roughly $100 billion in revenue run rate, Google employed 81,000 and Meta 71,000. Anthropic will be under 6,000 — a twelfth, not 30% smaller. It took about $10 billion to train the last Anthropic model, per the S-1 discussion, and 70 to 75% of compute goes toward the next model. Each frontier lab spends a billion or more on training data alone. Five to ten customers, each spending a billion. My read: that customer list fits on one slide, and it is why late-stage investing is now about paying anything to stay inside the few names that can spend like that.

Unicorns are dead. Bring me twenty billions.

LPs grade on $20 billion outcomes, Jason repeats, at A, B, or C — whatever it takes, since smaller outcomes may never get liquidity. Every mega platform is raising its biggest-ever fund in Q1 on a 12-month redeployment clock. Rory warns that fast raises tend to earn less: the clever Anthropic checks were written in early 2023 and 2024, when fear ruled. Now 10 becomes 20 becomes 40 in six months, and deployment speed itself fuels markups, compressing 8x toward 3 or 4x.

The show's investment-committee game prices three names. Kalshi at $65 billion, doing $5 billion in revenue: sports betting is wildly addictive, but Jason wants a sober count of bettors under full legalization, plus lawsuit and regulatory risk. Mercor raising at $20 billion: labs hate the dirty work of data, and if every company owns its intelligence, data budgets keep growing — Harry sketches $10 billion from model providers plus $10 billion more. Fractile at a $7.5 billion entry for inference chips against Groq and Cerebras, with a $250 million Anthropic order: not the value play, all agree, against a mid-twenties-market-cap public comp shipping about a billion this year toward three billion next. The panel picks Mercor, reluctantly, because capital must be deployed before January.

The Nvidia-weakness kicker is pure venture mechanics. With Jensen gone, Jason says the liquidation preference weakens: Nvidia always bought the kids — Reflection, Poolside, Hugging Face — backstopped Nscale, rescued teams. No new CEO pays $17 billion for Poolside or $15 billion for Clem to win open source. Poolside already admitted it could not raise the $10 to $20 billion gap, which is why it sold. My read: quotable, but built on deal gossip rather than filings.

Customers stopped buying software. They buy outcomes.

Jason's lighthouse metric is net-new logos. HubSpot closed 10,000 net new customers a quarter by clockwork; then 6,000, then 5,000, guided below 5,000. Monday fell from over 1,200 to under 200, down 80%. Price hikes and net retention hid it for years; below a hard deck of new logos, he says, you enter deep-value mode. Founders are back everywhere — HubSpot, Mongo, Workday, UiPath — and it always helps, yet no old SaaS name has reaccelerated above 30%, Palantir's doubling aside.

Small businesses simply refuse the $40,000 HubSpot renewal, stretch Sheets six more months because ChatGPT makes it good enough, and pay $500 to $1,000 a month for front- and back-office agents that close the day they demo. The episode's image: a receptionist quits every Wednesday, so an $800 agent that reschedules and bills around the clock gets signed that afternoon. SaaStr went from 20 people to 3 humans plus 21 agents; collections that cost $3,000 to $4,000 a month now cost one billed hour. Fifty Instinct automations are painful to move, and a $30,000 HubSpot quote dies once Muse pulls the top 200 founders from Salesforce and invites them to dinner. It even ordered closing flowers unasked — no card on file yet. Rory's pricing reply: users gravitate to whoever can bundle all that at a tenth of delivery cost, which with billions in reach is probably Meta, not Instinct.

The stack underneath is shaking

Microsoft stayed down about 36 hours — Outlook, Teams, Copilot — on a load bug that sat unfixed for four months. The FAA ordered a nationwide ground stop with roughly 4,000 planes airborne over a router failure tied to an L3Harris contractor; the agency is finally retiring floppy disks, at least air-gapped. NSA internal comms went down the same stretch. Rory stresses none of this is alleged AI cyber. These are classic IT failures with devastating blast radius.

The startup lesson is Fireworks. A testing corner leaked keys; Notion, Cursor, and Harvey named the vendor outright, which never happens. The team is best in class with maybe 150 SecOps and DevOps people, and attackers still found the edge. The collision defines next year, Jason argues: one-person billion-dollar startups with no security versus a world where only S-tier security teams can be trusted — he would buy new analytics or payments from nobody except Stripe or Adyen. Scale was never easier; risk was never higher.

So the question is not whether Kalshi, Mercor, or Fractile is worth its price. It is who becomes one of ten or twenty $20 billion outcomes when funds must deploy by January, buyers sign $800 agents but not $40,000 suites, and mail plus air traffic halt for a day. Ask what is not getting done, Jason says — five times in ten the agent already does it better than whoever is left in the office. The other five are why security now decides whether the venture math works.

More podcast deep dives from Podwise.

Podwise

Skip the six steps. Just listen.

Podwise transcribes, summarizes and maps any episode for you.