20vc X Saastr This Week: Anthropic At $350b, Openai’s Existential Risk, And Why A16z’s $15b Fund Actually Makes Mathematical Sense

A deep dive into the biggest AI rounds, the future of venture capital, and why California’s wealth tax could trigger a founder exodus

We just wrapped another 20VC + SaaStr pod with Harry Stebbings and Rory O’Driscoll from Scale Venture Partners, and this one was packed. Anthropic just raised $10B at a $350B valuation. Andreessen Horowitz pulled in $15B — representing 22% of all venture dollars raised in 2025. And California’s proposed wealth tax has already sent Sergey Brin, Larry Page, and Peter Thiel fleeing the state.

When you run the numbers on these mega-rounds and mega-funds, the picture that emerges is far more nuanced than the usual Twitter hot takes. Some of these bets look surprisingly rational. Others? Not so much.

Let’s dig in.

Top Takeaways

#1. Anthropic at $350B Looks Expensive — Until You Do the Math

The headline number sounds insane. $350 billion for a company that was doing $100M in ARR at the end of 2023? But here’s the thing: Anthropic 10x’d two years in a row. They went from $100M to $1B to allegedly $9-10B at the end of 2025.

If they “only” 3x next year to $30B in ARR, and you average that with their $10B opening ARR, you get approximately $20B in actual GAAP revenue for the year. That puts the valuation at roughly 17x NTM revenue.

That’s cheaper than Palantir. It’s comparable to Cloudflare.

The old rule holds true: You really CAN pay up for anything that goes 10x year-on-year. The investors who got in at $170B three months ago just 2x’d in four months. Calculate that IRR.

#2. Anthropic Has Won Enterprise AI — And They’re Not Stopping There

Everything I see at the API level? Claude has won enterprise. Nothing is perfectly stable in AI, but so far, nothing has dented Anthropic’s position. They’ve birthed Cursor, Lovable, Replit, Harvey, and Lore. Cursor is essentially a derivative of Claude.

But here’s what’s interesting: Anthropic is now expanding across three distinct markets:

Market 1: Enterprise API — Selling to ISVs and enterprises building on top of their models. This is where they started.

Market 2: Claude Code — Instead of being 50% of a coder’s spend (with the rest going to Cursor), they now capture 100% by owning the coding product directly.

Market 3: Claude Workspaces — The new product announced this week. It’s Claude for non-coders: PowerPoint manipulation, data work, all the knowledge work that isn’t coding. This is them going after Microsoft Office at the AI layer.

If Claude Workspaces works, it’s a terrifying proposition for Microsoft. Every knowledge worker uses the Office suite. What’s the AI version of that suite? Anthropic is betting it’s them.

#3. Cursor Should Be Nervous — But Not Panicking

Every CPO I speak to now instantly mentions Claude Code. The portion mentioning Cursor has dropped dramatically in the last three months.

Would I be nervous if I was a $27B pre-money Cursor investor? Yes. They have two massive competitors — Claude Code (their own supplier) and GitHub — both of whom can bundle with adjacencies.

But here’s the nuance: Cursor has graduated to the big leagues. They’re competing with Microsoft and Anthropic. That’s actually a signal of success. The other 10 coding agents? They’re not even getting a seat at the table.

The scorpion and the frog metaphor applies here. Anthropic has no reason NOT to sting Cursor eventually. They could cut off access to models. They could limit access to the best models. They could simply copy the product — how hard is it to build an IDE?

But the scorpion dies too. And right now, Cursor represents a billion dollars a year of essentially free money for Anthropic.

#4. OpenAI Has Genuine Existential Risk

Let me be direct: I think OpenAI has existential risk.

Not “going to zero” risk. They have 800 million users. They’ll find a model. But the relative value of Anthropic to OpenAI has gone from 10:1 to roughly 2:1 in three years. If you were in a race and leading 10:1, you’re now only 50% ahead with the competition closing fast.

Here’s the bear case that’s actually scary:

• The shelf life of an LLM is less than 100 days. The half-life is incredibly short.
• OpenAI needs ~$100B in the next 2-3 years. More than they’ve spent to date.
• Their competitors are in better shape: Anthropic has superior margins. Gemini has massive cash flow. xAI will probably get a trillion dollars in Trump contracts.
• If macro disrupts capital availability at the wrong moment, OpenAI could be frozen while competitors keep moving.

Would you use ChatGPT from a year ago? Claude from a year ago? NFW. There’s not a one in a million chance any developer would use a year-old model today. If OpenAI can’t raise and gets frozen, they become AOL — still existing, grandma still uses it for recipes, but the world has moved to broadband.

#5. Apple Choosing Gemini Over OpenAI Matters More Than You Think

Apple chose Gemini for Siri over the prior relationship with OpenAI. The big deal comments are these:

Google and Apple have a long-standing relationship where money flows from Google to Apple for search placement. So defaulting to your existing partner makes sense.

But here’s what’s interesting: Apple might be PAYING Google for Gemini, while simultaneously receiving billions for search placement. The products are different.

The privacy angle matters more than we discuss. Apple made clear they felt Gemini was the best solution for user privacy. We don’t talk about privacy enough in AI. Some enterprises think Salesforce isn’t secure enough — they want to run CRM on their own private clouds. Google Cloud has a massive business running Salesforce on private clouds for customers who worry about security.

When Apple says privacy is part of it, there’s another level of enterprise deal that a lot of folks simply can’t touch.

#6. Andreessen Horowitz’s $15B Fund Actually Makes Mathematical Sense

Everyone asks: Can they make 3-5x on $15 billion?

Here’s how Rory broke down the math, and it completely changed my thinking:

Step 1: Is the industry in equilibrium?

a16z raised $15B, representing ~20% of the total. That implies the industry raised ~$75B. The total value of exits this year (not amazing) was around $300B. If Anthropic alone goes public next year, you could see $500B+ in exits. So $75B in, potentially 3x+ out. The macro math works.

Step 2: Can they capture their share?

They’re deploying about 10% of venture capital on a sustaining basis. So they need 10% of everything: 10% of Series A’s, 10% of Series B’s, 10% of exits.

That DST research showed a16z did roughly 10% of all Series A’s that became $5B+ outcomes over the last decade. That’s their market share. Can they maintain it? Yes, if you have a top-two brand.

Step 3: What’s the real risk?

The top three private companies represent ~$1.8T of the $3.6T in total private company value. If you miss even ONE of those top exits, the math gets much harder. You don’t have to do the Series A of SpaceX, but you better show up on the cap table before they hit a trillion.

#7. The “Boutique vs. Mega-Fund” Debate Is More Nuanced Than Twitter Thinks

Here’s what’s interesting: a16z has actually structured themselves as multiple boutiques under one umbrella.

They’ve given Martin his sandbox (American Dynamism, ~$1B+). David Yolovich has his sandbox (Fintech, ~$1B). Alex Rampell has his sandbox (Infra, ~$1.5B). Each of those funds is a focused firm just like Scale or Index.

The difference? They have the air cover of the brand. And they have the $5B late-stage fund to “clean up on aisle five” — covering mistakes at the early stage by doubling down massively on winners.

This is the real power of late-stage funds paired with early-stage investing: You can be more promiscuous at Series A because you have enough late-stage capital to cover errors. If you get one good Series A, you’re willing to get three or four wrong — because you’ll do the B, C, D, and E on your winner.

#8. The California Wealth Tax Is a Trojan Horse — And It’s Worse Than You Think

This is NOT about a one-time 5% wealth tax on billionaires.

The coalition behind this bill has been working on it for years. They’ve attempted versions three times already. This is phase one of a multi-year plan:

• Phase 1: Pass the “one-time” billionaire tax. Easy to sell politically.
• Phase 2: Convert it to an annual tax. (You can’t solve an annual healthcare gap with a one-time tax.)
• Phase 3: Lower the threshold to $50M, then $25M of paper wealth based on last-round valuations.

If this plays out, every founder with a Series B at $500M+ valuation would pay a wealth tax on illiquid paper gains. Every year.

My prediction: If this passes and the next bill goes up, founders will begin massively exiting California in 2027 before the second bill. It becomes “leave before the Series B” as the new playbook. Do YC, stay a year, build your team, then leave.

The voters in California will vote this stuff in — at least if the vote happened today. Many people feel like billionaires are getting wealthier while their company is doing layoffs. When everyone’s in St. Barts competing with yachts while you just got fired from Zoom, you’re going to vote to tax those guys.

#9. ElevenLabs Is Incredible — But There’s Fragility Under the Hood

I built a game this week called Founderscape.ai (try it — I put 200 hours into this thing). You join YC, build your team, fundraise, go public. It simulates the whole founder journey.

I added ElevenLabs for voice, so your CTO talks to you throughout the game. The API is beautiful. I implemented it in literally 90 seconds. It’s the best API I’ve worked with.

But here’s the thing: I burned through $30 in credits with just 20-30 players in 48 hours. At my game’s current usage, it would cost $1,320/month for 11 Labs.

The moment Replit or someone else launches a voice product at a tenth the price that’s 80% as good, I would have to switch. The API is so elegant that integration is trivial — which means switching is also trivial.

ElevenLabs did $330M revenue from nothing in one year. But $5B revenue (which you need to justify a $30B exit at 6x) in speech? That’s a lot. Especially if the revenue is concentrated in large customers who have pricing power over you.

The math works if voice adoption is massively distributed — thousands of developers each spending $5K-$20K. It doesn’t work if it’s Epic Games spending $500M and grinding you on cost.

#10. The Revenue-Per-Employee Revolution Is Coming — And It Will Cause Social Unrest

We’re going to normalize around $1-2M revenue per employee in startups. Replit has 200 employees at $300M revenue. 11 Labs probably has even fewer.

When you can do startups with a fifth the headcount we used to need, it’s not about AI “replacing jobs” — we just don’t need that many people. One-to-two million per employee is the new normal.

Here’s what’s terrifying: When you 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?

One in three Nvidia employees is worth $20M or more. 18,000 people at Nvidia are worth $25M+. The Palo Alto real estate market has literally zero houses for sale.

But if you just got laid off from Zoom, it’s been 12 months, and you can’t find a job? You’re going to vote to tax the AI deca-millionaires. Each year people get angrier.

This wealth gap is going to breed contempt. And that’s why I think wealth tax proposals keep coming — each year they’ll get more aggressive, not less.

Quotable Moments

Jason Lemkin

“I’ve given up on this nervosity of competition and disruption in the age of AI. None of these products even worked 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.”

“OpenAI has existential risk. It is a bet that the best of times lasts at least a decade. And 10 years with no downturn would be a long cycle historically.”

“This wealth tax is a Trojan horse. The goal is it becomes 1% or more forever, then lowers to $50M and $25M thresholds. If you have $25M of paper wealth based on your last round, you’ll pay a 1% wealth tax annually. It will become ‘leave before the Series B’ as the new playbook.”

Rory O’Driscoll (Scale Venture Partners)

“In the early stage, you’re taking uncorrelated business risk. In the late stage, you’re taking 100% correlated valuation risk. When it goes wrong, it’s going to go wrong for all of them.”

“They’ve raised 20% of total venture capital, which means on a sustaining basis they need to get 10% of everything — 10% of Series A’s, 10% of Series B’s, 10% of exits. They’ve done roughly 10% of Series A’s that became $5B outcomes. It’s not crazy if they can maintain that market share.”

“The comment on being scared — if you’re going to be uncomfortable being scared, you need to just go home. In SaaS land, you could compound for seven or eight years. Now there’s existential risk every six months. If you can’t live with that, you probably need to find a different job.”

Harry Stebbings

“If you’re OpenAI, are you not slightly nervous? You’re being eaten away by Anthropic on the enterprise side with incredible model performance, and on the consumer side you’ve got Gemini outperforming. Combined with very high SBC and high churn — it feels precarious.”

“I think people are a lot more promiscuous than we give credit. Since the latest Gemini models came out, you’ve had a 22% drop in ChatGPT usage. My son dropped it. He pays for Cursor out of his own pocket, but he doesn’t pay for ChatGPT anymore.”

“There’s one segment that will always exist in venture: when there’s a glitch in the matrix. When they stumble a bit, or no one sees the re-acceleration. Someone is the hottest company at YC, has a couple great months, reboots, and then re-accelerates 12 months down the road. That’s where you can still find opportunities.”