Jason Lemkin | 20VC Summary

1. Key Themes

Anthropic's IPO is being engineered to start low and walk up

Rory O'Driscoll argues the pricing is a deliberate setup: a modest range keeps the private-round price as the reference point, leaves room to raise the range, and avoids a failed "start high, come back" IPO. Jason Lemkin is the skeptic, predicting a flat trade on market fatigue.

Rory: "It is far better in an IPO to start low and go up than to start high and come back. Cuz once you start coming back, everyone loses confidence and you've blown it." 00:00:13

He also explains why headlines show different valuations for the same price: "when you do a valuation on the private side, you include all the options. So the same price per share looks like a bigger pre-money... that's why the newspapers say it's roughly a trillion today, but about a trillion-three, a trillion-four if you counted all the granted but not unvested options." 00:24:25

Jason: "I think it's going to trade flat." 00:26:04

Being public beats being private in the compute-capital arms race

Rory sees Anthropic's IPO as a structural advantage over OpenAI because access to public capital markets ends the private fundraising treadmill.

Rory: "They're going to get their IPO done in the next month. They're going to have the money, they're going to have the float, and they're going to be through with the private begging bowl bullshit... I'm willing to bet 6, 9 months from now, that CFO will have a convert on file." 00:27:54

Jason adds that public disclosure itself is valuable: "Now we're going to have Anthropic every quarter having to disclose at least far more than we've ever seen before. It's a gift to all of us." 00:28:45

The AI trade's real risk is demand, not supply

Rory dismisses the $2.5T NASDAQ-100 drop as pod-shop deleveraging rather than evidence of cracking fundamentals, and restates his thesis that only demand can break the trade.

Rory: "The AI trade doesn't crack on the supply side. It doesn't crack because the data centers can't be built... It doesn't track for any of those other reasons. If the demand is there, and that demand will be evidenced in ARR for OpenAI and Anthropic, if the demand is there, everything else will be roughly fine." 00:00:13

On the mechanism: "the good folk at Citadel and places like that, they basically take your checkbook away at 5% and they fire you at 7.5% drawdown. So it doesn't take a lot to have a bunch of these kind of pods just blow up." 00:22:26

Harry relays a countervailing view from a trillion-dollar allocator: "I'm not worried about this 80 billion. I'm worried about the next 100 billion, cuz that layer is very thin." 00:26:48

Frontier-lab leadership is ping-pong, and token price elasticity is the next unknown

OpenAI reportedly re-accelerated to 70% annualized revenue growth in Q3 after a miserable earlier quarter, which both hosts read as proof that developer loyalty is thin.

Rory: "We're going to discover what the demand elasticity of token pricing is... the basis of competition will be pretty clear: compute, capital, and token pricing." 00:30:00

Jason: "When Dev was on the show from Mongo, what he said is there's no loyalty with developers... the entire time we've done this show, it's been ping-pong, right? Ping-pong between these guys." 00:30:53

Agents, not seats, now drive public-market winners and losers

Cloudflare and Twilio are rising on agentic consumption while seat-based SaaS (Atlassian, Figma, Monday, HubSpot) is de-rated.

Jason: "If it's not an agentic beneficiary, I don't want to hear it anymore." 00:51:22

"More and more is going to be agent-driven... I don't see any way that the seat-based guys are coming back." 00:56:20

Rory on Atlassian: "the new trend is agents not engineers, and that means the seat-license thing is brought into question." 00:52:54

Headcount efficiency was latent; AI made it visible

Cloudflare (20% layoffs, now up 39%), Freshworks (crossing $1B with 30% fewer employees), and Palantir ($2M revenue per employee, ~98% growth) illustrate the shift.

Jason: "AI may be the excuse, and we didn't need the people, and AI opened our eyes to the reality of it when it was always staring us in the face in 2021, working 20 hours from home and doing our four side hustles." 00:50:25

Rory: "if you have $2- and $5 million contracts like Palantir, and if revenue doubles, you probably can't hire head count fast enough to catch up. You literally can't stop your margins going up." 00:50:42

Legacy software can't match AI-native shipping velocity

Jason says non-AI-native companies are failing to change how they operate, which makes M&A like Dialpad/Ada a soft landing rather than a real exit.

Jason: "As we're rolling into 2027, guys... these non-AI companies can't get their acts together. They can't run their companies differently than 2021, '22, '23. They can't move faster." 00:45:58

"I don't view it as an exit. If there's no liquidity, I don't think it's technically an exit." 00:45:58

API-key theft is the emerging crime wave of AI infrastructure

Jason details swarming attacks on inference providers to steal Anthropic and OpenAI keys.

Jason: "Everyone's trying to steal API keys for inference... my spending, I didn't even know, was set at $100,000 by default. So if someone stole mine... they could have charged $100,000 in minutes." 00:40:27

"It's the most lucrative criminal trade you can do right now... it's better than selling Rolexes you steal off someone's wrist in London." 00:40:27

He cites Wiz: "nine of the leading AI developers disclosed a swarming incident in October alone," and Unit 42: "bills are $1 million before detection." 00:43:22

Incumbent security vendors are consolidating agent security through M&A

Rory frames CrowdStrike and Palo Alto acquisitions as textbook incumbent behavior.

Rory: "if the new security issue is agent security, then just go do acquisitions, deal with the problem... The trick is to buy the hot one, get intercepted by CrowdStrike or Palo Alto fairly quickly." 00:43:55

Persistent agents are compute-hungry, and the unit economics are unresolved

Jason revised his assumption about agent product costs.

Jason: "I thought... the cost to run [these agents] was like 4 to 5 bucks. That's what it costs to run Lovable or Replit... I was wrong. This is like up to 20 bucks or 30 bucks a month to run these things." 00:31:49

Rory's counter: "Short-term compute prices go up because of supply and demand, long-term compute prices go down because of Moore's Law... The question is are you 2 years away or 5?" 00:33:23

2. Contrarian Perspectives

The evergreen fund was intellectually right but a comp-plan and complexity mistake

Most would read Sequoia winding down its evergreen structure as a retreat from long-term thinking. Rory defends the logic while explaining why it failed: the partners had already been paid, so the LPs were taking the medicine.

Rory: "the real money is made after the IPO... It is arguably almost always true that if you hold a diversified portfolio of really interesting tech assets over the long term, they have compounded well above any kind of S&P return." 00:05:18

"You've already been fully paid for what you're doing. So now you are basically giving the LPs their vitamins. You're basically saying, 'Damn it, you will hold these stocks cuz it's good for you. I've already made my money.'" 00:07:59

Jason, by contrast, calls it "the best idea in venture ever." 00:04:24

Jensen Huang's China chip push was conviction, not compromise

The popular framing is that a Chinese intelligence operation compromised him. Rory argues the "entrapment" had nothing to extract because he was already lobbying for exactly this.

Rory: "Without any lobbying from any Chinese agent, from any entrapment agent, he wants to sell GPUs to anyone who'll buy them... You didn't get anything by entrapping him because, for entirely logical reasons, he's been pushing that agenda pretty consistently." 00:17:32

Jason adds that China is the real loser: "It's going to be harder to not only sell GPUs to China, it's going to be harder to just host them across the border. It's going to all get shut down." 00:00:00

Anthropic's IPO scares fewer people than OpenAI's private raise should

While Jason and Harry's allocator source worry about capital fatigue, Rory argues the market structure favors Anthropic.

Rory: "It's the first chance to buy a pure-play Gen AI next-gen company other than just picks and shovels. So I think there'll be wild interest... this is a company that will have $200 billion of cash on the balance sheet, more cash on the balance sheet than any other company in America." 00:27:14

Buying legacy SaaS with AI-native companies is never worth it

Jason raises the obvious arbitrage (AI-natives buying slow incumbents' contracts), then both reject it.

Rory: "It's never worth doing... you buy someone else's old contracts, and then you inherit all their problems, and the customer's never grateful for you for solving them, they're just mad at you." 00:49:29

Compute-layer startups may be "tourist" trades, but the wave justifies them

Jason picks fal despite concerns that compute becomes a commodity in 3 to 5 years.

Jason: "In venture, sometimes we just have to be tourists, we have to be part of a wave... I know Rory's right, compute will become a commodity in 3 to 5 years. You just can't get compute today." 00:35:10

3. Companies Identified

Anthropic

Frontier AI lab maker of Claude, in the process of IPO-ing. Mentioned as the central story of the episode; the hosts cite $80B run-rate, ~$200B of cash post-IPO, a $70-78 per share recommended range versus $54 in the last private round, and ~$41M average employee payout.

Jason: "Anthropic also said they're at an $80 billion run rate right now, right? So you don't get much better than that." 00:23:30
Rory: "They put a recommended price of 70 to 78 bucks a share, which is relatively low. As a reminder, the last private round was 54 bucks a share." 00:24:25
Jason: "The average Anthropic employee, $41 million in the IPO... 216 billion going to current employees and 160 billion in the pool." 01:01:34

OpenAI

Frontier AI lab behind ChatGPT and Codex. Mentioned for a re-acceleration to 70% annualized revenue growth in Q3 and enterprise becoming the majority of revenue; also for relative disadvantage in being private.

Harry: "Annualized revenue grew 70% in Q3. Enterprise doubled and became majority of revenue... Codex is crushing and really handing it to Claude Code in a lot of ways." 00:29:26
Rory: "If they really did re-accelerate to 70%, that's stunning." 00:30:00

Nvidia

Dominant AI chip company under board investigation relating to its CEO. Cited for a ~$5T market cap, longest-tenured tech CEO, and an unusually long-tenured board.

Rory: "It's hard to argue he's not a success when you've had a $5 trillion market cap." 00:14:31
Jason: "This is too important to AI. It's too important to the US." 00:00:00

Sequoia Capital

Top-tier venture firm restructuring (leaving London, closing the scout program, merging early and growth teams, ending evergreen). Mentioned for its brand and ability to reinvent itself.

Rory: "They are always good at reinventing themselves." 00:11:01
Rory: "they've been at something absurd like 25% of the NASDAQ." 00:05:18

Founders Fund

Venture firm with a tiny partnership and large funds; admired for deployment speed and concentration.

Harry: "Founders Fund deployed the growth fund in 11 months... they're in Stripe, they're in Anduril, they're in SpaceX with that fund, and it's like no LP is complaining." 00:13:13
Rory: "they concentrate down, you know, from I think 20 investments in the first fund to seven or eight. It's exactly how you should do it." 00:13:48

ElevenLabs

London-founded AI voice company, a Sequoia portfolio company. Cited as proof Europe's best companies can be covered from the US.

Jason: "I would almost think ElevenLabs, and if it wasn't for Mati's accent, I would think they were Bay Area companies. They're everywhere." 00:09:48

Revolut

UK fintech at a $115B secondary valuation, up 53% from the last round. The preferred IC pick of Harry and Rory.

Harry: "They are obliterating European incumbents. We will see market share for Revolut unlike we've ever seen for any historical incumbent... there's a 4X from here, with the downside that we have key man risk on Nik." 00:39:48
Rory: "all those customers who are 19 to 25 today are just going to get older, they're going to want mortgages, they're well positioned to be the biggest bank in Europe." 00:39:14

Zipline

Drone delivery company raising $1.5B at $21.5B, three times its January price, with a Walmart contract and US rollouts in Dallas and elsewhere.

Rory: "they've done 3 million deliveries in the US... you're probably only getting 5, 10 bucks a delivery... at 21 billion, you're really leaning into that Walmart contract." 00:36:52
Jason: "I like it better than investing in Waymo. The unit economics are better." 00:34:32

fal

Generative media inference provider valued at $16B on more than $800M of ARR, selected by Jason.

Rory: "You probably get a couple of good years' growth here or there... at 16 billion, you might be closer to the end than the beginning." 00:38:14
Jason: "I think we can flip this at 40 billion, worst case, so we have more than our pref." 00:35:10

Cloudflare

Public infrastructure company that cut 20% of staff and is now growing 39%, nearing $4B revenue, driven by agent traffic.

Jason: "It's agents, and it's agentic spend at Twilio, too." 00:51:22
Harry: "20% layoffs earlier in the year, and they're crushing, up 39%." 00:49:56

Twilio

Communications API company seeing agent-driven growth acceleration.

Jason: "Twilio, you know, moldy-oldie Twilio" is being bid up by markets for agentic exposure. 00:56:20

Freshworks

SaaS company that crossed $1B revenue with 30% fewer employees.

Harry: "Freshworks crossing a billion with 30% fewer employees. God, efficiency at its finest, huh?" 00:49:56

Palantir

Enterprise software company cited for $2M of revenue per employee growing at 98%.

Jason: "Palantir at 2 million, 2 million per employee... growing at 98%." 00:50:17

Figma

Design platform at roughly $10.2B market cap, ~40% revenue growth and strong free cash flow, but with a seat-based model under pressure.

Rory: "You are dealing with a 40% plus or minus revenue grower, very strong free cash flow." 00:54:26
Jason, on a path forward: "I would go all in on agentic design... if we don't do this, we'll be like Adobe." 00:57:14

CrowdStrike and Palo Alto Networks

Consummate security acquirers buying agent-security startups (~$400M-range deals).

Rory: "I think CrowdStrike and Palo Alto are consummate acquirers." 00:43:55

Dialpad and Ada

Dialpad acquired Ada (AI customer service) for $400M in stock at a '21 valuation.

Rory: "If you're Ada, you are a very good pre-LLM chatbot, services chatbot... your ability to now compete with Sierra, maybe Decagon, is just hard." 00:45:18

Fireworks

AI inference provider breached twice in a month by API-key swarms.

Jason: "Fireworks has had two major breaches in this month or in the last 30 days." 00:40:27

Lovable and Replit

AI app-building platforms, both cited at $1B scale, used as unit-economics comps for agent products.

Jason: "That's what it costs to run Lovable or Replit... it's a good business." 00:31:49

CoreWeave and Nebius

Neocloud companies mentioned as the main public datapoints for AI economics.

Rory: "There might be some companies... that get ahead of themselves like CoreWeave, but it'll all be fine." 00:22:26
Jason: "We've had to sit around trying to figure all this crap out from CoreWeave and Nvidia and Nebius." 00:28:45

Wiz and Unit 42

Security research sources cited for swarm-attack statistics. 00:43:22

Other companies named in passing

Dollar App, Odoo, Bolt: Sequoia's European investments cited by Harry as evidence of European quality. 00:09:21
Anduril, SpaceX, Stripe: Founders Fund growth-fund holdings. 00:13:20
Altos Ventures: Ho Nam's firm, cited for continuation funds as a generational-hold structure. 00:07:33
Monocle: Where Sam Blond now is. 00:12:46
Megaspeed: Chip-reseller linked to the Nvidia compliance issue; Nvidia reportedly will not sell it anything. 00:16:35
Cooley: Law firm whose CEO sits on Nvidia's board and leads the investigation. 00:20:08
Synopsys: Its CEO sits on Nvidia's board. 00:20:08
Harvey: Legal AI company cited as dependent on inference providers. 00:42:32
Canva: Cited as too slow. 00:47:17
Bending Spoons: Mentioned as a comp for a different type of acquirer. 00:47:17
Atlassian, Monday, HubSpot: Seat-based SaaS names under pressure. 00:53:15
Adobe: Cited as the cautionary path for Figma. 00:58:29
Notion: Design-quality benchmark. 00:58:29
Google: Pre-committed capital into Anthropic's round; also a Sequoia holding. 00:24:18
Robinhood: Offering retail access to Anthropic. 00:26:48
Walmart: Zipline's anchor customer. 00:38:01
Waymo: Comp for autonomous delivery. 00:35:10
Citadel: Reference for pod-shop drawdown limits. 00:23:29
MongoDB: Source of "no loyalty with developers" (via Dev). 00:31:32
Rippling: Parker Conrad cited on California exodus. 01:03:15
Wix: Acquirer of Base44 at $200M. 00:33:19
Cisco: Example of a Sequoia IPO hold. 00:07:33
YC and Speedrun: Accelerator benchmarks. 00:13:13

4. People Identified

Jensen Huang

Nvidia founder and CEO since 1993, the longest-tenured tech CEO, under board investigation.

Rory: "this is a world-class executive who's... the longest-tenured tech CEO period full stop." 00:14:31
Jason: "The founder CEO odds are will have the longest lifespan in tech." 00:08:35

Nik Storonsky

Founder of Revolut. Named by Harry as "the best founder of our generation" with key-man risk.

Harry: "I want security and I want safety and I think Revolut is the best asset by far here with by far the best founder of our generation." 00:36:36

Greg Brockman

OpenAI co-founder, increasingly the public voice of the company.

Harry: "I think it's interesting also Greg's further prominence into the spotlight. Greg is the one who's saying a lot of these statements now." 00:29:26

Mark Stevens, Tench Coxe, Brooke Seawell

Long-tenured Nvidia board members. Stevens and Coxe are original VCs, with Stevens the Sequoia partner on the board since 1993. Seawell was the original audit chair.

Rory: "Two of the original VCs, Mark Stevens and Tench Coxe, and the original audit chair from '93, Brooke Seawell... I bet you he's held all his stock, and I bet you he's worth a couple of billion bucks." 00:20:08

Mati Staniszewski and Anton Osika

Founders of ElevenLabs and Lovable. Jason cites both for their presence at elite Bay Area events.

Jason: "Every elite event or dinner I'm at, Mati's there. Half of them, Anton's there." 00:09:48

Dylan Field

Figma CEO. Jason empathizes with the long-term IRR hit after the failed acquisition.

Rory: "Dylan has done an amazing job, grow your damn company 40% a year, get it to free cash flow positive, get the damn thing public." 00:54:26

Ho Nam

Altos Ventures founder, known for generational-investment thinking and continuation funds. 00:07:33

Sam Blond

Former Founders Fund partner, now at Monocle. 00:12:46

Orlando Bravo

Thoma Bravo founder. Harry jokes he should hire Jason to teach CEOs to be AI-first.

Harry: "I would be paying 50 million bucks a year to Mr. Jason Lemkin to come in and talk to our CEOs." 00:48:28

Marc Andreessen

Mentioned for a clip on AI-native versus non-native startups. 00:46:15

Parker Conrad

Rippling founder, cited on the California billionaire exodus. 01:01:34

Sergey Brin and Google founders

Credited for funding the opposition to California's wealth tax at roughly 10:1 over unions.

Jason: "Sergey and friends did outspend the unions 10:1." 01:01:24

Marco Rubio

Referenced as having received a report on the Nvidia matter and told Trump. 00:18:57

Others

Jeff Lawson (former Twilio CEO), whom Jason speculates would make Twilio better. 00:52:27
Shyamal (referenced as "Shyamal Maguire"), who posted urging that Nvidia not be derailed. 00:15:59
Elon Musk, whose move to Texas Jason calls a modest loss for California. 01:03:15
Mike Cannon-Brookes, Atlassian CEO, referenced for UX views. 00:52:54
Steve Jobs, the backdating precedent. 00:16:35
Peter Thiel, invoked on picking concentrated bets. 00:36:14
Craig Walker, Dialpad CEO. 00:47:17

5. Operating Insights

Who you hire to investigate determines the output

Jason, who has been through two board investigations of CEOs, offers a practical lesson.

Jason: "You kind of get who you hire in the investigation... the input of who investigates it, who you hire, how objective they are, are they biased in favor of an outcome? They're paid... just be very thoughtful... that the output may already be decided when the input begins." 00:18:57

Rory's counter-lens: scrutiny from Congress or the DOJ raises the cost of a whitewash. "When you're doing an investigation that you know is going to be second-guessed by a different investigation... the scrutiny level goes way up." 00:20:08

Deploy early capital diffusely, late capital concentrated

Rory distills Founders Fund's pattern: "You have to be somewhat diversified early on, but you have to be widely concentrated later on... they do a lot of little experiments... and kill them quickly. And then on the growth fund, they concentrate down from I think 20 investments in the first fund to seven or eight." 00:13:48

Large checks require conviction, not headcount

Jason: "It seemed like too much capital for too few partners... But when it works, you have to have the ability and conviction to write large checks into winners." 00:11:36

Retention after a valuation collapse needs deliberate intervention

Jason on the malaise at companies whose stock fell after a failed exit: "You can fix it with RSUs and you can fix a little bit with additional grants and targeted stuff, but there is just this malaise... it's hard to attract new people, right? So you double down on the existing folks." 00:55:26

Watch real users do the task to find product gaps

Rory describes observing designers perform the same task in Canva, Figma, and agentic tools. Designers consistently found agentic tools easier because "they just type in what they want and it shows the node display." 00:58:42 Founders should see competitors through actual customer workflows rather than feature matrices.

Set spend limits on API keys

A directly actionable lesson from Jason's experience: his default Claude spending cap was $100,000, which is a theft magnet. Audit and lower defaults on any key exposed to third-party platforms. 00:41:06

6. Overlooked Insights

The most liquid growth businesses may be handing the best employees an exit that overwhelms a state tax base

Buried in the wealth-tax discussion, Jason notes the per-person numbers at Anthropic: roughly $41M average per employee, founders each selling $264M, and "216 billion going to current employees and 160 billion in the pool." 01:01:34 This is a tax-base and talent-migration story as much as a wealth story: the people who would leave under a wealth tax are not passive holders but active builders, and Jason says two quiet billionaire founders told him they were already planning to leave. 01:01:34 Rory's counterpoint, that "even if the company doesn't move, people can move when they're making their tax decisions," suggests the real policy risk is individual residency choices after liquidity, which is a vector most people tracking corporate relocation would miss. 01:03:29

The "last 20 billion" of Anthropic's round was partly Google's repackaged pre-commitment

Jason mentions in passing that the round could not be fully filled and that "some of it was Google's pre-commitment repackaged... they had to lower the size of the round and lower the valuation." 00:23:30 Rory contests the oversubscription read since pricing hasn't happened, but the detail matters for investors modeling demand: a meaningful portion of apparent demand may be strategic and circular rather than independent price discovery. Combined with Harry's allocator comment about the "next 100 billion" being thin, it suggests the real test is organic public-market demand for the first float, and signals that strategic investors may be a larger share of the book than headlines imply. 00:26:49