Anthropic’s $10B Raise | a16z’s $15B Fund: Is the Middle Dead in VC? | How OpenAI Could Go to Zero?

Summary

• One panelist’s math on Anthropic’s $10B raise at $350B: revenue went $100M (end-‘23) → $1B (end-‘24) → ~$9-10B (end-‘25) — 10x two years running. Assume it “only” 3x’s to $30B, average the opening and closing ARR to ~$20B GAAP revenue, and you’re paying ~17x NTM — a lower multiple than Palantir, “kind of comparable with Cloudflare.” “It turns out you really, really can pay up for anything that goes 10x year on year.” One panelist expects this is the last private round before IPO, and raising only $10B probably signals healthy unit economics.
• Anthropic has three enterprise categories: the premium enterprise API, coding via Claude Code (enterprise coding revenue “plus or minus comparable” to Cursor and GitHub), and — as of the day before recording — a workspace product for non-coder knowledge work. One panelist’s zoom-out: the AI version of the Office suite is “a huge ass idea” and “a scary one if you’re Microsoft.”
• The scorpion-and-frog warning for Cursor: Anthropic already cut off xAI’s access this week, and could limit, degrade, or simply clone — “how hard is it to build an IDE that’s just the same as Cursor?” One panelist: “I would be nervous if I was a 27 billion pre Cursor investor, but they’ve created something amazing.” The panelist discussing the scorpion would still invest, but sees the sting as possible.
• On OpenAI, one panelist’s scoreboard: OpenAI’s lead over Anthropic has gone from 10-to-1 to only 2-to-1 in three years — “You’re still in the lead. Don’t blow it.” Scott’s bear case is sharper: OpenAI needs $100B in the next 2-3 years, more than it has spent to date; if a macro shock freezes it in time it becomes Detroit or AOL — “Would you use ChatGPT from a year ago? NFW.” “OpenAI has existential risk. It is a bet that the best of times lasts at least a decade.”
• On likely Andreessen Horowitz’s $15B raise (22% of all 2025 venture dollars), one panelist reverse-engineered the model and convinced himself it works: on a two-year cadence it’s ~10% of venture capital, so they need ~10% of every great Series A and B — which DST data says is already their share of Series A’s that became $5B outcomes. The real engine: “You can be promiscuous at the A if you have enough late-stage stuff to cover it up” — the growth fund is “clean up on aisle five.”
• The trade-off nobody’s pricing: “In the early stage you’re taking uncorrelated business risk and in the late stage you’re taking 100% correlated valuation risk.” The example — likely Databricks at ~$100B is ~25x revenue on 40%+ growth; if growth slows to 20%, those companies have historically traded ~6x. “If the growth is there for one more year, it looks cheap” — and if it isn’t, the dislocation hits every late-stage book at once.
• One panelist’s new worry, born from burning $30 of 11 Labs credits in 48 hours on his vibe-coded game: this is the year substitution risk gets real in AI — “probably the back half of the year… Marc Benioff will actually be right, we will rotate out for cost.” He’d back CEO “Motti” no matter what but wouldn’t touch $11B without more work; another panelist’s counter is the Nuance lesson — customers who swear they’ll swap vendors often never do.
• Both see California’s proposed wealth tax as a Trojan horse: Jason says the coalition behind it has already tried three times to set thresholds at $50M then $25M of paper wealth, annually, priced off the last venture round — so the rational founder move becomes “leave before the Series B,” and “you could have a Detroit in the Silicon Valley.” One panelist bets voters reject it, but “even trying to do it has had an economic cost” — Brin has already joined Page in leaving.

Deep dive

1. Anthropic at $350B: you really can pay up for 10x

• One panelist walked through the arithmetic he’d run inside his own partnership: Anthropic went “from 100 million in ‘23… to a billion at the end of ‘24 runway to allegedly about between 9 and 10 billion at the end of ‘25” — 10x two years in a row. Assume next year they “only quote only 3x” to $30B run-rate; averaging opening and closing ARR gives ~$20B GAAP revenue, so the round is ~17x NTM revenue — “a much lower revenue multiple than Palantir. It’s kind of comparable with Cloudflare for God’s sake.”
• His conclusion: “If the growth is there for one more year, it looks cheap. It’s the old rule — it turns out you really, really can pay up for anything that goes 10x year on year.” And the investors who paid 170 three months ago? “They’re a 2x in 4 months. Calculate that IRR.”
• One panelist’s read on the raise size: “only raising 10 billion… is actually a sign that the unit economics are probably healthy” — minimal dilution, and Anthropic “own not just enterprise but they own code creation, application creation… everything we have spent our lives working on.” His hedge-free gut call: “I know it’s the trite VC thing to say, but it’s hard not to believe we’re in the first inning.”
• One panelist expects this is the last private round before an IPO: “they’ve stated they want to and it feels like they can… then logically it should get done.”

2. Three markets, and the third one should scare Microsoft

• One panelist split Anthropic’s enterprise position into three: the premium API business (with the standing risk that ISVs route commodity work to cheaper open-source models), then Claude Code — “coding is probably the single largest use case for what we make, let’s build a coding product” — where instead of being 50% of a coder’s revenue at 50% gross margin “you’re getting 100% of the revenue because you’re selling the product.” His estimate: enterprise coding revenue “plus or minus comparable” to Cursor and GitHub, maybe a little lower but growing nicely.
• The third leg, announced the day before: a Claude product for non-coder knowledge work (“a Claude workspace, I think it’s called — I could be wrong on that”). The inversion matters: “Instead of bringing the AI to the Excel spreadsheet, which is what Copilot tried to do at Microsoft, you bring all these tools into the workspace.” Early reviews he’d read: “yes, it’s amazing, but it’s a bit janky.”
• The zoom-out — “a scary one if you’re Microsoft”: every knowledge worker buys Office, so “what is the AI version of the Office suite?… The idea that for every knowledge worker there can be some bundle like this is a huge ass idea. I don’t think this is it yet,” but the direction of travel is to be “the place where you do knowledge work,” not the chat interface.

3. The scorpion will sting Cursor — invest anyway

• Harry’s provocation from his 20Product interviews: every CPO he asks now names Claude Code, and “the portion of people that said Cursor has gone down dramatically in the last 3 months.” One panelist: “Depends on the price I got in at… I would be nervous if I was a 27 billion pre Cursor investor, but they’ve created something amazing.” His league metaphor: Cursor has graduated to playing Microsoft and its own supplier — “are you a little scared? Yeah, but you’re damn glad to be playing there, cuz the other 10 coding agents — any of them going to get to play?”
• Jason’s fable — worth keeping whole: Anthropic cut off xAI’s access this week, and “there’s no reason that Anthropic just might not sting Cursor just before it gets to the other side of the river… It would be naive to assume otherwise.” The sting has many forms: limit access to top models, degrade it, or simply copy — “how hard is it to build an IDE that’s just the same as Cursor?… They can build Replit and Lovable too. These are not the greatest challenges of mankind. So all of them are at risk… but I would still invest.”
• One panelist’s meta-posture on fear: “these products didn’t even work a year ago — how nervous can you be holding a large position in a product that didn’t work a year ago? You can only be so nervous or quit the game.” Another panelist agreed: “In SaaS-land [as heard: ‘Fastland’] you could compound for seven or eight years. Now there’s existential risk every 6 months. If you can’t live with that, you probably need to find a different job.”

4. Apple choosing Gemini: distribution, plus the privacy layer VCs ignore

• One panelist’s framing of the Siri deal: Google and Apple have a long-standing arrangement where ~$10B a year flows to Apple for search placement, “the best distribution on the planet” — but here, with no advertising model, “Apple might be paying Google for Gemini.” If OpenAI ever ships an ad model, “the dynamics of the money move can flip.” At the margin, losing a billion phones stings: “it’s not like OpenAI blinks at bad economics — those guys have an economic indifference curve that would make your head hurt.”
• Another panelist added the underrated variable: “we underestimate how mission-critical privacy is in the deep enterprise.” Google Cloud runs “a massive business running Salesforce on their own private clouds for customers who somehow worry Salesforce isn’t secure enough” — so when Apple says Gemini was the best answer for user privacy, that’s a real bar most startups can’t clear: “We’re not more secure than Salesforce. It’s a high standard.” And bluntly: “Certainly today Google feels like a far more stable partner than OpenAI.”

5. OpenAI: still in the lead, but the lead went from 10:1 to 2:1

• Harry pressed the bear case — eaten by Anthropic in enterprise, outperformed by Gemini in consumer, “very high SBC and high churn… it feels precarious.” One panelist’s deadpan opener: “Well, luckily you’re a nonprofit, so whoever wins is great for the global economy” — which another panelist corrected: OpenAI isn’t a nonprofit anymore, its largest shareholder is, so “if OpenAI’s value goes down, the largest loser is a charity. The second largest is Microsoft. It will survive.”
• One panelist’s scorekeeping — the cleanest frame of the episode: the relative value of Anthropic to OpenAI “has gone from 10-to-1 to now only 2-to-1… You were in the lead 10-to-1 over the other guy. You’re still in the lead, but he’s now 50% behind you and coming on fast. Are you nervous? Yeah. You’re still in the lead. Don’t blow it.” He called the Sebastian Mallaby goes-to-zero take “absurd” — 800 million users, subscriptions, a real business.
• Scott’s rebuttal is a concatenation, not a single failure: OpenAI “needs a hundred billion in the next 2 to 3 years. That is more than it has spent to date.” If a macro shock freezes it: “Would you use ChatGPT from a year ago? NFW… There’s not a one in a million chance any developer would use a year-old model today.” The company “would be like Detroit… or AOL and dialup. Grandma’s fine with ChatGPT from a year ago because it helps her with recipes, but the rest of the world’s moved on to broadband.”
• One panelist conceded the corner case on his 2x2 — macro breaks and scaling laws keep working — but called it “concatenated probabilities that are fairly low,” invoking the Bill Gates rule: always hold two years of opex cash, “because you’ve got the world’s best fundraiser.” Scott’s wider warning stands regardless: “we have ascribed the odds of a downturn to less than zero… it feels like late 2020, 2021… OpenAI has existential risk. It is a bet that the best of times lasts at least a decade.” On consumer stickiness, one panelist said ChatGPT usage is down 22% since the latest Gemini models — and Rory said his son dropped his $20 ChatGPT subscription while paying for Cursor out of his own pocket.

6. Likely Andreessen Horowitz’s $15B: get 10% of everything, and the math closes

• One panelist’s take on the raise (22% of all venture dollars raised in 2025): “so what… you might as well hoover up 51% of the capital and then just shut down your competitors.” What’s remarkable is the 2x2: “Andreessen not only raised the most capital but has the strongest founder brand. That’s hard to do both” — plus published returns that are top-tier, with no fund below 3x.
• One panelist came in expecting to argue it can’t work and reversed himself on his own math: $15B at 20% share implies the industry raised ~$75B; exits this year were ~$300B in an unremarkable year, and “if Anthropic alone goes public next year it’s 500 billion of exits” — so the industry is “roughly in equilibrium” and can 3x total invested capital. On a two-year fund cadence, the firm is ~10% of sustaining capital, so “they’ve got to get 10% of everything — 10% of the great Series A’s, the great Series B’s” — and the DST partner’s decade-long study showed they already did roughly 10% of Series A’s that became $5B outcomes.
• The dependency he flagged: total private value is ~$3.6T, but “if you chop off just the top three deals you’re down well north of a trillion bucks… You simply can’t make this kind of math work without getting those top exits. You don’t have to do the A of SpaceX, but you better show up on the cap table before they hit a trillion.”
• The scaling limits, per one panelist: more deals means more pickers, and “it gets hard to be smart in a room with more than five or seven people” — the DST data shows the firm’s Series A market share is higher than Benchmark’s but “the hit rate is much lower.” Jason’s rejoinder: conflicts are “a super solvable problem” — law firms solved it — so why not 50% market share, leaving “Sequoia and those General Catalyst guys the scraps.”

7. The late-stage fund is “clean up on aisle five” — and the middle fights back

• Jason dismissed the fund-you-at-every-stage pitch — “I don’t think founders go skipping down the street when they hear that” — but one panelist’s third argument landed: the growth fund lets you make errors at the A. “I’m willing to get three or four of them wrong because in that good Series A I’ll do the B, C, D and E and the other stuff gets lost in the noise… We made some whoopsies, but we’ll just clean it up. You can be promiscuous at the A if you have enough late-stage stuff to cover it up.” It’s the engineer’s approach to venture — the firm “engineered an overall system” where Benchmark tries to pick.
• Harry’s version of the alpha: “the ballooning of your growth assets means you have ever-increasing price elasticity on your early assets — they can just come in and bid 300 when we’re bidding 150, doesn’t freaking matter, because David George is going to put in a 300 million check at three or four billion.” Then the knife, via Alex Rampell’s “the middle is dead” thesis: “I mean this in the nicest and most loving way, Rory… are you not the middle?”
• Rory’s defense — worth keeping in full: crudely on AUM, yes, “you are the middle.” But boutique doesn’t mean small, it means focused — and the firm itself has “split the fund up into four funds roughly our size… very Alfred Sloan,” giving Martin, Yolovich, and Rampell their own sandboxes because “you’d see deterioration of investment quality” otherwise. “Each of those funds is a freaking boutique fund, just like us” — what they have that a $900M fund doesn’t is the air cover of the brand and the $5B late-stage cleanup. His survival condition: “you have to know something the general funds don’t… you have to get there earlier. If you wait till it’s consensus, you’re probably going to lose.”
• One panelist’s history lesson on scale entrants: since Benchmark in ‘95, only two have truly broken in — the likely Andreessen Horowitz firm (founders-as-engineers who “systematized” the business) and Founders Fund (founders who were incredible investors). Benchmark and Index prove you can still run seed/A brilliantly on $500M to $1B.

8. The meta question: can you still find a $10B gem outside the system?

• Jason’s frame for why he stays in the game: “Can you still find a $10 billion gem outside of the boundaries of this system? If you cannot, then this is all a game of fees… it’s all performative. It’s all 25K checks into hot YC companies and it’s all a lifestyle joke on Twitter.” With YC, Project Europe, HF0 and South Park Commons locking up discovery, “there are a lot of VCs that think the only thing left is inception investing” — or pre-inception: “we’ll go to middle school.”
• One panelist’s counter-evidence that picking still matters: roughly 20-something percent of unicorns went through YC — 80% didn’t — and “venture does a stunning job of missing the turn.” Salesforce struggled to raise a dime; “the first venture round at OpenAI was 23 billion pre with Thrive, and the first venture round at Anthropic was 4 billion pre with Spark and Menlo.” Anthropic itself: “the very first time they tried to raise money, 22 out of 23 VCs said no.” Market efficiency is real but partial — his firm’s “hard to beat” names appear in 40-50% of A rounds and ~80% by the C.
• Jason’s remaining niche: the “glitch in the matrix” — the second seed, when a hot YC company stumbles and “reaccelerates six months, 12 months down the road… I just invested in one that, because of Anthropic and friends, reaccelerated two years after YC.” Jason’s own specimen: “Owner was a glitch in the matrix when Redpoint didn’t see it and I did the seed — and then they came in and put in every single round since.”
• Harry’s challenge to the whole premise: Europe’s three breakouts — Lovable, 11 Labs, and an entity heard as “Lorra” — have grown “entirely linear… no faltering.” One panelist’s concession-with-warning: if growth is linear and obvious to everyone, picking degrades into “a beauty contest — and you’re probably going to rank lower than some people who have $15 billion and the guy who invented the browser… Be very nervous when you think everything’s going to work.”

9. Uncorrelated business risk vs. 100% correlated valuation risk

• The core line, the one to keep: “In the early stage you’re taking uncorrelated business risk and in the late stage you’re taking 100% correlated valuation risk — and when it goes wrong, it’s going to go wrong for all of them.” The mechanism: “when it’s obvious, people pay up because the only risk left to take is valuation risk” — the best firm wins the round but pays market price; “Lovable is not saying I’ll take six billion when I could get 8 billion from someone else.”
• His worked example, picking on “one of the best companies out there”: likely Databricks at ~$4.5B revenue, 40%+ growth, cash-flow positive, valued around $100B — ~25x revenues. “Across the last two decades, 20% growth companies trade around six times… If the growth stays, the valuations stay. If the growth slows down even slightly, you have a dislocation to the downside” — magnified because “everyone’s leaned in so far” for three straight years.

10. Substitution risk arrives in the back half — the 11 Labs stress test

• Jason’s live experiment: he vibe-coded Founderscape.ai (“200 hours… it simulates everything — fundraising, batchmates, going public”), added an 11 Labs-voiced CTO — “it’s so effing good, a 99 out of 100 product” — and burned $30 in credits in 48 hours with 20-30 players. Replet quoted him $1,320/month at current usage. His conclusion: “if I could do something a tenth the price that was close to as good, I would have to switch” — and that from a customer who doesn’t even look at his Replet bill.
• Asked if he’d invest at $11B: he’d back CEO “Motti” “no matter what — even if the ship went down,” and 11 Labs did $330M revenue from nothing with “without question the best API I’ve worked with — I implemented it in 90 seconds.” But at $11B, “I’m not smart enough to take it… there’s an underlying fragility to it” — while conceding a 3-5x exists “if the whole world uses voice the way all the VCs talk about it.”
• One panelist’s on-the-fly underwriting: a 3x needs ~$30B, “at scale you’re going to trade at six or seven times because that’s the way human life is, dude — get over it” — so ~$5B of speech revenue. The deal works only if the market stays distributed: “tens of thousands of people spending no more than 20 or 30 grand” rather than an Epic Games paying half a billion and designing you out. And his Nuance scar tissue cuts the other way: a rival speech company’s $2-3M customers all told his reference calls they’d swap it out — “you fast forward five or six years, they’re all still on the platform. They never got around to substituting.”
• Jason’s generalization, with his Clerk/WorkOS example ($30/month SSO he’ll “immediately delete” once Replet ships a native equivalent): “I think this will be the year — probably the back half — where we have to take substitution risk seriously in AI… Marc Benioff will actually be right. We will rotate out for cost.”

11. The “entrepreneurs tax”: a Trojan horse, and leave before the Series B

• One panelist’s two technical points: wealth taxes always underperform projections because wealth is mobile — “Norway, France… invariably unwind them” — and this one assesses ownership by voting control, so a founder with 10x super-voting shares (a structure he “didn’t agree with 10 years ago and I’ve changed my mind totally”) gets taxed as if 5% ownership were 50%. “Are you going to sit in California worth $2 billion and give a billion for the privilege of living here? You’re going to leave.” Right policy frame: “not ideological — how can I cost-efficiently milk this cow?”
• Jason’s darker read: “This is a Trojan horse… You cannot solve an annual healthcare gap with a one-time tax.” The coalition behind it has tried three times to pass versions with thresholds lowered to $50M, then $25M of paper wealth — annual, at 1%+, priced off your last round. Hence: “Leave before the Series B… if this goes as far as the folks backing it want, you could have a Detroit in the Silicon Valley” — the meme becomes “come to Dogpatch, do YC, stay a year, build your team, and then leave.” Winners: Miami and Austin, the almost-winners of 2020-21. He’d personally go if the annual lower-threshold version lands: “one year, who cares — but 10 years compounds to 15 to 20% of your net worth.”
• Rory’s optimistic dissent: “the default California voter goes in to say no… my guess is in the end this loses.” But either way it’s already “an own goal” — Brin has joined Page in leaving, Harry said Chamath had reported—he thought—$700B of $2T was gone, and “even trying to do it has had an economic cost.”

12. The wealth gap AI is building — and why the anger compounds

• Jason expects B2B to “normalize around a million to $2 million per employee” — Replit runs ~200 people on $300M revenue — which means “we just don’t need that many people,” and that breeds malaise even inside tech’s own ecosystem. Meanwhile at Nvidia, “one in three employees is now worth 20 million or more — 18,000 folks worth 25 million or more” — and in Palo Alto “there are literally zero houses for sale.”
• The collision he’s watching: “If you just got laid off from a previously high-flying public SaaS company growing 4%, what are you going to do? Who’s going to hire you?… I’m already seeing folks on LinkedIn — ‘after 27 years at Microsoft, I’ve decided January 15th is my last day.’ What happens when the next job is impossible?” His verdict on the St. Barts yacht season: “It’s gross. It’s tacky.” It’s hard for everyone not to want to tax the f out of everybody. Harry’s close, from the UK: the London-versus-everyone disparity is “terrifying” — and one panelist: “It’s going to grow… I do think there is a level of social unrest that will grow over the coming years.”