Jason Lemkin | 20VC Summary

Participants: Harry Stebbings, Jason Lemkin, Rory O'Driscoll

1. Key Themes

Anthropic Is Quietly Lapping OpenAI and the Race Is Already Decided

The growth trajectory data leaked out is stark: Anthropic did in Q1 2025 alone what it did all of last year, while OpenAI's Q1 revenue was only ~30% of its full-year 2024 figure — and they're nearly the same size now. More critically, Anthropic just became profitable with gross margins expanding from negative 60% to 70% in roughly two years.

"Anthropic has done as much in Q1 as all of last year. And OpenAI has done 30% of what they've done last year in Q1. And they're only slightly bigger than Anthropic. You play that out for a couple more quarters... within a couple of quarters, Anthropic, on the current trajectory, will be visibly and obviously ahead. Profitable, growing more quickly and bigger. That's Pareto dominant on all three vectors." — Jason Lemkin 00:35:33

"Gross margins expanded from 38% to 70%. They're going to have a $559 million operating profit in Q2 projected." — Harry Stebbings 00:19:04

The AI ROI Question Is the Central Economic Question of 2025

The Uber COO couldn't measure ROI on AI spend. Microsoft is reportedly pulling back from Anthropic's Opus over cost. As companies move from experimental budgets ($3M) to real budgets ($300M), the scrutiny will intensify. The key insight: token consumption goes up as models do more agentic reasoning, even as per-token prices fall — meaning total compute costs rise.

"I can't remember where corporate America was so convinced of the ROI of something as it is right now of AI... And I do think the next shoe to drop will be, OK, that was fun. We spent $10 billion. What did we get? And it's just the way the narrative has to play out." — Jason Lemkin 00:18:23

"The trend here is exponentially increasing costs as you move from simple chat interaction all the way to full-on agents. So even though the cost per token stories are yay, amazing, going down, the actual cost to do something goes up a lot." — Rory O'Driscoll 00:14:19

The Layoff Wave Is AI Efficiency, Not COVID Overhiring — And It Will Accelerate

The discussion tears apart the popular narrative that current tech layoffs are just COVID overhiring corrections. With 20%+ annual attrition since 2020, those hires have already turned over. What's actually happening: companies are reallocating payroll from low-performers to high-performers who are now orders of magnitude more productive with AI — and paying them 2-5x more.

"You have no excuse. Like, you couldn't manage... You had five years to manage your low performers out, and now you're blaming it on overhiring?" — Harry Stebbings 00:26:43

"I'm terminating people to make room for people that might cost more per head that are just of different skills... what Zeb really said is I just want to pay my current high performers more." — Harry Stebbings 00:30:55

"If there is efficiency gains from AI across R&D and sales and marketing in particular, everyone's going to be doing 20% layoff just because it's 20% more efficient. My aha is this discussion and the last discussion are, in fact, the same discussion. In one case, we're focusing on the winner, which is Anthropic. And in one case, we're focusing on the loser, which is employees." — Jason Lemkin 00:31:29

2. Contrarian Perspectives

SpaceX's S1 Is Financial Engineering, Not a Tech Story — And the Numbers Don't Support $2 Trillion

The mainstream narrative celebrates SpaceX as the greatest company in history. The contrarian read here: it's Elon rolling failed assets (Twitter/X, X.AI) into a great but relatively modest private business to bail them out, at 100x trailing revenues.

"It could be the GeoCities deal of the AI era. We could look back on this... 100 times trailing sales, we may look back and say these were some good companies. But 100 times, my God, my God. That's worth two trillion? Renting out chips because Jensen's your buddy and you're a good customer." — Harry Stebbings 00:00:24

"It's SolarCity on steroids and we're all here for it because we love AI, but it makes no sense. This conglomeration of friends of Elon makes no sense to anyone but the folks getting bailed out on Twitter." — Harry Stebbings 00:50:54

The AI Productivity Divide Will Get Worse, Not Better — Most Workers Will Fall Further Behind

The optimistic take is that agentic tools will democratize and everyone will become more productive. The contrarian: the frontier keeps moving faster than people can retrain, and the gap between those who master it and those who don't will compound into unemployability — not a reskilling opportunity.

"I have some lingering concerns that instead what happens is the agentic experts get another 10x better and everyone else falls further and further and further behind and are unemployable. And we go to 4 million in revenue per employee and 5 million in revenue per employee." — Harry Stebbings 00:33:34

"If you're on this exponential curve that Dario and Sam keep talking about, then the knowledge to be frontier keeps growing every year. And it's just super hard to keep up. One of my favorite Max Planck quotes is, science advances funeral by funeral." — Rory O'Driscoll 00:37:07

Spending More on AI Tokens When You Claim the ROI is Great Is Actually Irrational — You Should Be Token-Maxing NOW

When the CEO being discussed said "if Anthropic doubled prices we wouldn't change usage," Rory and Jason flipped it: that's actually an admission you're underinvesting, not a testament to value. If you can absorb 2x price, you haven't pushed usage to where marginal cost equals marginal value.

"What he basically said is the ROI on this AI is so good, I could pay twice as much. Well, in that case, let me tell you what you should do right now. You should go in and tell your people, use twice as much... If Anthropic could double your prices and this project is still economically viable, then what's the next project on your to-do list?" — Rory O'Driscoll 00:17:22

Agents Need Entirely Different Infrastructure — Human Software Tools Are Obsolete for the Agentic Era

Most investors are still betting on human-facing SaaS surviving the AI transition. The harder insight: agents don't use Zoom, Dropbox, Salesforce, or Google. They need a fundamentally different toolset. Companies building for agents (not for humans using AI) are in a different category entirely.

"Agents don't hop on Google and do search. They don't. And they don't create files on Dropbox and they don't hop on Zoom. Like agents have a whole, true agents have a different set of workflows and tools they use... Google was the killer use app for humans. Your agents need it, but they don't need Google." — Harry Stebbings 00:05:15

"Maybe we should only be investing in those things and leave those human tools behind guys. Enough investing in human software." — Harry Stebbings 00:06:06

3. Companies Identified

Anthropic Description: Foundation model company, makers of Claude Why mentioned: Dramatically outpacing OpenAI on revenue growth, gross margin expansion from -60% to 70%, now profitable — Pareto dominant on growth, size, and profitability

"Profitable, growing more quickly and bigger. That's Pareto dominant on all three vectors." — Jason Lemkin 00:36:01

NVIDIA Description: Semiconductor company, dominant GPU supplier for AI Why mentioned: $81.6B revenue quarter, ~$56B profit in a single quarter, potentially on track for $200B annual profit; 7% of all American 401k savings

"It's not just a great revenue business growing at 80%. It's a wildly profitable operating margin business... that's 200 billion of profit a year." — Jason Lemkin 00:06:21

Exa (formerly Metaphor) Description: Search infrastructure purpose-built for AI agents Why mentioned: Raised at $2.2B; identified as one of the core primitive tools agents need — analogous to what Google was for humans; backed by Benchmark; Harry is a small investor and early user

"Exa nailed one of the core... It wasn't even obvious to me until I started. There was an obvious issue. Like your agents do need to find current information... it's a look into the future." — Harry Stebbings 00:05:44

OpenRouter Description: Developer platform allowing dynamic switching between 50+ LLM models Why mentioned: Raised $150M at $1.3B led by Capital G; identified as a key infrastructure "pick and shovel" for the agentic era; enables enterprises to optimize token spend across models

"The OpenRouter one is interesting because going back to what Jason says, if people are spending $300 million on a premium product and having to lay off people as a result of that, there's going to be some interest in exploring cheaper costs." — Rory O'Driscoll 00:03:47

Benchmark (VC Firm) Description: Tier-1 venture capital firm Why mentioned: Identified as investor in both Exa and another breakout company (Monaco/Jack Altman deal); praised for willingness to break rules and invest at moment of traction inflection in AI

"Another Benchmark company. This fund is going to be one of the best performing funds in history. I mean, seriously." — Harry Stebbings 00:08:52

Supabase / Neon Description: Open-source database backends (Neon is Supabase's competitor) Why mentioned: Cited as example of agentic-native infrastructure with staggering adoption signal — over 90% of databases on Neon now built by agents, not humans

"At Databricks, Neon, which is their super-based competitor, over 90% of the databases are built by agents, not by humans." — Harry Stebbings 00:12:51

4. People Identified

Jensen Huang (CEO, NVIDIA) Description: Founder and CEO of NVIDIA Why mentioned: Projected $3-4T in AI CapEx by 2030; framed as the central figure determining whether the entire AI infrastructure investment thesis holds

"Jensen Wang said this week, on AI CapEx infrastructure spend that we'd reach $3 to $4 trillion by 2030." — Rory O'Driscoll 00:09:22

Tony Xu (CEO, DoorDash) Description: Founder-CEO of DoorDash Why mentioned: Used as a counter-example to Uber's skeptical COO; aggressive founder-led company that reportedly sees clear ROI on AI vs. engineering cost savings — if he turns skeptical, that changes the debate

"I haven't seen DoorDash say, let's use less token, guys. If I see it from both... from one of the most aggressive founder-led companies out there... then I'll be like, okay, we got ahead of ourselves for real companies." — Harry Stebbings 00:16:30

Zeb (CEO, ClickUp) Description: CEO of ClickUp, project management SaaS Why mentioned: Praised for unusual transparency in explaining layoffs — explicitly stating the purpose was to pay high performers 1M+ in an AI-augmented org, rather than hiding behind vague restructuring language

"I liked what Zeb said, though, from ClickUp, even though he got hazed more than math. One of the things I liked that he said, and it was very transparent. It's like, listen, I'm laying off 22% of my company so I can pay a million dollars to my high performers in the age of AI." — Harry Stebbings 00:30:06

Jack Altman Description: CEO of Lattice (and referenced deal: Monaco); brother of Sam Altman Why mentioned: Cited as example of smart late-seed/early-growth investing in AI — joined Benchmark, invested in Monaco post-traction (after Founders Fund pre-revenue), and marked up nearly 2x quickly; illustrates the "invest the hour it blows up" strategy

"Jack Altman just joined, right? And he just did Monaco, which I know well. I'm a small shareholder. And they marked up the deal very quickly, almost 2X from Founders Fund. But when Founders Fund invested, it had no revenue." — Harry Stebbings 00:09:22

5. Operating Insights

The $2M Revenue Per Employee Standard Is the New Benchmark for All Startups

This was a number once reserved for Apple-tier public companies. The panel argues it is now achievable and expected at the startup level with AI augmentation. Operators who aren't calibrating toward this benchmark are structurally over-hiring.

"I think $2 million per employee is going to become the new normal as you scale. And that will mean that your high performers can make two to three times what they used to make... Instead of the high performers making 40 or 50% more than your mid-pack, it should be 5x." — Harry Stebbings 00:32:28

Pay Your Best People 5x the Mid-Pack, Not 40-50% More

The ClickUp playbook — lay off to concentrate compensation, not just reduce headcount — is an underappreciated operating tactic. The ratio of high performer comp to average comp needs to completely reprice.

"Instead of the high performers making 40 or 50% more than your mid-pack, it should be 5x... Anthropic, because you just don't need as many people. And you're going to do what ClickUp did is you're going to funnel your compensation to your high performers." — Harry Stebbings 00:32:49

Agent Idleness, Not Token Cost, Is the Real Bottleneck for Efficient AI-Native Orgs

For companies already operating at high AI maturity, the constraint isn't spend — it's human throughput to process agent output. The implication: hire fewer but much higher caliber humans specifically to act as intelligent processors of agent output, not to do the work agents can do.

"Our agents are idle. We don't have enough brain sight... if you wake up and build me 50 features every single day, how many features can I even qualify overnight?... You can't implement 21 ideas a week." — Harry Stebbings 00:18:43

"If the AI is now giving more ideas than one human can process, and they're good ideas, hire a second human who will be now more effective. And the return on human capital will go up." — Rory O'Driscoll 00:19:57

6. Overlooked Insights

Anthropic's "Public Storage for Compute" Lock-In May Be the Stickiest Enterprise Contract in Tech History

This was mentioned almost in passing, but it's remarkably significant. The X.AI/Anthropic deal ($1.25B/month, ~$15B ARR, 90-day cancellation clause) was framed as a win for Elon. But Jason Lemkin flipped the insight: Anthropic went in thinking it was temporary, but compute demand is so acute and alternative capacity so hard to build that they may be locked in for years — at escalating rates. This is the self-storage business model applied to frontier AI compute: customers go in expecting to leave, never do, and pricing power accrues entirely to the provider.

"This is like that. And you go in, I'm sure Anthropic went in and thinking, we're only going to pay him 1.25 billion for four months. But they're relying on other people building data centers, and life is tricky, and not everyone will bulldoze it through like Elon did... you cannot take it away... at 1.25 billion, it's a great deal. His ROE on that is pretty damn high in a low ROE business." — Jason Lemkin 00:56:02

The "Invest the Hour It Blows Up" Strategy Is a Structurally New and Superior VC Approach for AI

Briefly mentioned in the context of Exa/Benchmark, but deeply underappreciated: in AI, the window between "no traction" and "obvious product-market fit" has compressed from 18-24 months to weeks. This invalidates traditional seed-stage conviction investing as the risk/return sweet spot. The new optimal entry point is the precise moment of traction inflection — which requires real-time market intelligence, fast decision-making, and willingness to pay up — but dramatically reduces binary risk.

"The amazing thing about these AI companies is that moment used to be a year, year and a half, maybe two years of early product market. Now what you're seeing is you go from early product market fit to, oh my God, it's incredibly obvious just given the take rate literally sometimes in weeks, if not months... You've got to find that moment. And yeah, you've got to pay." — Jason Lemkin 00:10:22

"You want to invest the hour it blows up. The minute it blows up, you want to get the DM and just wire the money in AI. That's the play." — Harry Stebbings 00:09:51