Jason Lemkin | 20VC Summary

1. Key Themes

The Government Is Getting Enmeshed in AI — And Tech Is Inviting It In

The Fable 5 ban and OpenAI's 5% government stake proposal both point to a fundamental shift: AI companies are voluntarily stepping into a regulatory embrace that the internet industry spent 20 years avoiding. Rory drew a direct historical contrast: "Six months ago, you could ship software like a free man, and now you have to get permission from Washington before you do it... If you look at two of the biggest deals at the start of the internet... you had the Telecom Deregulation Act... Section 230... and for a long time no sales tax. All those things were basically Silicon Valley managed to have a 20-year run with the internet where the message to Washington was leave us alone and we did great." 00:05:01 Now the opposite is happening.

Sam Altman's 5% Offer Is Anchoring — Not Altruism

Jason reframed the OpenAI government stake proposal as sophisticated political communication, not naivete. "I think Sam is a very thoughtful communicator and he puts stuff out there early to socialize it... He's anchoring this idea that, hey, 5% will align us with the American people, with the federal government, with the administration... he's sensing the political winds. So I think it's anchoring 5% rather than 50% because not only does he need the alignment, he's sensing the political winds." 00:15:31 Jason further noted: "Maybe it really doesn't matter what we think because it's already happened. The decision essentially has already been made that the federal government will be acquiring a stake in OpenAI. And Sam is just anchoring it as the smallest possible stake to get ahead of this discussion." 00:16:30

Founder Dilution Sensitivity Has Collapsed — With Systemic Consequences for Investors

A quiet but structurally important shift: founders no longer fear dilution the way they once did. Jason flagged this directly: "I think in the age of AI, massive dilution has been sort of institutionalized. Founders don't care anymore... you'll look at really hot startups in the news. And if you peel the layers back and you look at the stub rounds and the up rounds and the half rounds, they've done 16, 17, 20 venture rounds often." 00:19:03 The consequence for seed investors is severe: "Harry just talked about doing a seed deal at 60, right? I think you're really doing it at 240, Harry, is the honest math today." 00:20:52

Enterprise AI ROI Is Genuinely Unproven — And Vendors Are Going to Get Caught Cutting Corners on Data

Alex Karp's CNBC comments resonated with the panel as substantively correct beneath the theatrical delivery. Rory summarized: "Corporate America is saying, I'm spending all this money. Am I getting anything? Which is the ROI comment. And then the other comment... he said corporate America is saying, am I giving them all this information? Are they training them? Are they learning my business? And are they going to be selling my business to everyone else? What's my IP?" 00:29:14 Jason tied this to a live example: "HubSpot said a week ago... we're going to pool all your data... And their customers erupted that you're sharing my contacts. They had to roll it back within a week... I think vendors overall are going to push the limits on training on your data." 00:30:40

China Is Building Its Own AI Stack — Because It Has No Choice

Jason's two weeks in China produced a sharp insight: the US has inadvertently guaranteed Chinese AI self-sufficiency by blocking access. "When you're in China, OpenAI and Anthropic will not serve you. It's not just a question of being blocked. You cannot access it... What do you expect China is going to do the second largest economy in the world? Of course, they're going to build things that are as competitive or better than we are because you can't even use Claude in China." 00:53:01 The result: "The top six models as of today on OpenRouter are Chinese." 00:52:01

Enterprise AI Adoption Will Be Slower Than Expected — A Talent Bottleneck, Not a Technology Bottleneck

Microsoft embedding 6,000 engineers in enterprise clients sounds like a solution, but Jason argued the talent simply doesn't exist at scale. "I think this is going to fail because I don't think there is enough talent to do what we want to do in the enterprise... I was told it would take three months to fix the fact that your AI is still talking about SaaStr 2026, which already happened." 01:03:24 Rory translated the structural implication: "The rate of adoption of their technology is to some extent a little bit outside their control... The biggest problem when I'm picking Exxon or Bank of America, rolling out Gen AI is not their ability to buy from Anthropic. It's the ability to do change management and application building in the enterprise." 01:08:21

Microsoft Is Becoming the New IBM — And That's Not a Compliment

Rory made a pointed observation about Microsoft's trajectory: mature companies with deep enterprise relationships, but lacking the cutting-edge product, end up as the services layer for the new platform. "Every technology company either goes bust or lives long enough to become next generation's IBM... OpenAI and Anthropic are the companies with the new incredible product, and Microsoft is the more mature company with the enterprise relationships who is going to build a large services business just like HP did, just like IBM did." 00:00:20 The implication: high revenue, but not nearly as profitable as selling operating systems once was.

Frontier vs. Open Source Models: The Real Split Is Problem Complexity, Not Price

A nuanced framework emerged from the Decagon CEO's post, endorsed by the panel. Jesse Zhang's framing: use frontier models for unknown unknowns, push to open source once the answer is commoditized. Rory articulated it: "When you're trying new stuff or you don't know the problem or you can't bound the problem, you're going to use frontier models because they're smart and they'll figure out the unknown unknowns. The more it becomes a commoditized answer where you know the answer you want to give, the more you're going to push it to open source." 00:58:09 Jason confirmed with a personal example: a problem that took 10 hours and $500 in a step-down model took 20 minutes and effectively nothing in Claude's highest tier. 00:59:04

NVIDIA Is Now Financing Its Own Demand — A Derivative Bet on Perpetual Compute Growth

NVIDIA's "Compute Now, Pay Later" model — selling chips upfront and providing put-back rights if neoclouds can't use the compute — is accounting legitimate but commercially aggressive. Rory: "They are going to recognize that hardware revenue up front... they are taking the revenue up front. So it's as legal as church on Sunday... But it is pretty aggressive... What it's basically saying is their push has been to diversify away from the hyperscalers... they are starting to expand the number of significant customers." 00:42:01 The risk: "If that slows down and there's excess capacity, these deals will look horrible because you'll not just be not growing quickly. You'll be debooking prior revenue." 00:43:21

2. Contrarian Perspectives

Giving the Government a Stake Will Make Things Worse, Not Better

Against the consensus that 5% government ownership creates useful alignment, Rory argued it triggers an uncontrollable political escalation. "You start with pre-approval. Suddenly you end up with an ownership interest. Then you end up with a board member... Bernie's already said he wants 50. And you deserve whatever happens to you, right? You deserve being regulated by the government... If half those nice middle-class people in the middle-class jobs all across this country lose their jobs to AI, it's going to take a lot more than $140 per head — which is what the 5% of our topic would be worth — to keep the wolf from the door." 00:08:11 The deeper point: the same doomsday narrative that enabled billion-dollar fundraises now makes political intervention logically inevitable.

The AI Jobs Apocalypse Narrative Is Delusional — And Its Believers Are Trapped By It

Rory argued that the catastrophist framing about AI destroying labor markets is both factually wrong and strategically self-defeating for OpenAI. "If you really are destroying labor in a $30 trillion economy, do you think the monster, the political monster is going to say, I'll settle for five? That's grand. Call it 50 billion. You've destroyed 15 trillion of labor value, but we'll settle for 50 billion." 00:12:07 But he also acknowledged why they believe it: "That belief is what gave them the self-motivation and the confidence to raise billions of dollars. That narrative is what it took." 00:12:59 They are now trapped in a narrative they needed to tell to get funded.

Open Source Token Volume Is Misleading — All the Dollars Are Still at the Frontier

The fact that the top six models on OpenRouter are Chinese open source doesn't tell you where the money flows. Rory: "The open router — all the tokens with open source — is a little misleading because all the tokens can be in one place, but all the dollars can be in the other place." 00:58:38 The economic reality is that hard, novel problems still require frontier models, and that's where willingness to pay is highest — regardless of what the token count distribution looks like.

Building Your Own Chip Makes No Sense If You're Driving That Volume to NVIDIA

Jason flatly rejected the "customization" rationale for AI labs building their own silicon. "If you're driving that much volume to them and you need a special version of a chip, they'll build it for you... For this amount of dollars, in my limited experience in the semiconductor industry, they'll do your own tape out, they'll build your own. It's so much money... The idea that it's customized for us is just soft language... it just makes no sense." 00:46:46 His real interpretation: it's purely a margin recapture play, with companies dancing around the real reason to preserve NVIDIA relationships.

Founders No Longer Fear Being Blocked by Late-Stage Investors — And This Is Quietly Changing Startup Behavior

Jason identified a structural shift in founder psychology that most observers haven't named explicitly. "No one's worried about making their last round high-priced investors' money anymore. Literally, no one is. Because I believe investors have learned to accept 1x when it doesn't work out without drama, without blocking, without threats." 00:00:15 This means founders are more willing to raise at high valuations and take existential risks they would never have taken a decade ago when blocked exits were a real threat.

3. Companies Identified

OpenAI

Leading US frontier AI lab. Mentioned throughout in context of government stake proposal, Sora shutdown, chip-building rumors, and competitive positioning against Anthropic and Chinese models. "OpenAI and Anthropic are the companies with the new incredible product." 01:05:54 Also: "The rate at which OpenAI and Anthropic got to $4 billion and $4 billion respectively in GAAP revenue [was] never before seen." 01:08:50

Anthropic

Frontier AI lab, competitor to OpenAI. Discussed as the company with Dario Amodei at 1.something percent equity, talks to build its own AI chip via Samsung, and Palantir's implicit foil. "Anthropic opens talks with Samsung to build its own AI chip." 00:45:02 Also flagged for having "opinions about how their AI should be used by the DOD," which Karp criticized.

Palantir

Enterprise AI and defense data analytics company. Alex Karp's CNBC appearance praised as substantively accurate, and Palantir positioned as the beneficiary of enterprise mistrust of frontier model providers. "If you're Palantir selling to the government and highly regulated industries, I think it's a great play. It's a great play. You can't really trust these guys not to share your data." 00:32:06 Stock went up 9% the day of Karp's appearance.

Kling (Kuaishou)

Chinese AI video generation model. Described as the most commercially successful AI video product on earth: "$500 million in Q1 ARR-wise... raised $2.8 billion at an $18 billion valuation... clearly going to go public on the Hong Kong Stock Exchange soon." 00:47:43 Contrasted favorably with OpenAI's shuttered Sora product.

Higgsfield

AI video generation platform. Harry and Jason are both investors. "Higgsfield just announced they're at $500 million in revenue, actually doing $2 million a day now just in credit card billings outside of the enterprise." 00:48:01 Described as a platform where users can run multiple video models including Kling side-by-side.

Harvey

AI legal tech company. Cited as a successful deployment model: "Every deployment they do at Harvey has an FDE and a lawyer. Every single deployment has a lawyer." 01:09:45 Identified as evidence that pairing deep domain expertise with technical resources produces successful enterprise AI rollouts.

Decagon

AI customer experience company. CEO Jesse Zhang cited for a smart framework post on frontier vs. open source model usage. "When you're trying new stuff or you don't know the problem or you can't bound the problem, you're going to use frontier models... The more it becomes a commoditized answer... the more you're going to push it to open source." 00:58:09

ElevenLabs

AI voice company. Secondary at $22 billion cited as a benchmark for the new employee liquidity imperative. "The 11 Labs secondary at $22 billion... as an employee today, why would you join something that you don't believe will have secondary options?" 01:15:23

Linear

Developer productivity tool. Praised by Harry as an example of rare capital discipline: "Carry it linear. The dude is so disciplined and so focused. He's raised two rounds of funding. He never wants to meet VCs." 00:23:37 Debated by Jason as potentially leaving value on the table if the prize turns out to be $100 billion.

Ramp

Fintech/expense management company. Mentioned as having done approximately 12 announced rounds (estimated ~24 with stub rounds), cited as an example of the new normal of serial fundraising. "Ramp's done 12 announced rounds, according to Claude. So I'm going to guess it's more like with stub rounds, like 24 rounds." 00:23:12

Databricks

Data and AI platform. Cited alongside OpenAI as an example of a company where founders maintain honest, old-school equity structures. "These are honest founders. These are actually old-school founders living in the AI age — DataBricks. They're honest, right?" 00:23:31

SpaceX

Elon Musk's rocket and satellite company. Cited as the precedent for Meta's cloud business pivot: a company that bought compute for proprietary goals, failed to fully utilize it, and turned it into a successful cloud rental business at exceptional prices. "One of them is SpaceX, obviously, and then the other obviously now is Meta. You ask yourself, what's going on long term?" 00:33:57

CoreWeave

GPU cloud / neocloud provider. Mentioned as a benchmark for neocloud business model and as a company whose stock declined ~10-15% on Meta and SpaceX entering the cloud market.

Nebius

GPU cloud provider. Mentioned alongside CoreWeave as a neocloud competitor negatively impacted by Meta and SpaceX entering the market.

HubSpot

B2B marketing and CRM software. Called out for attempting to pool customer prospecting data across its customer base and being forced to walk it back within a week after customer backlash. "HubSpot said a week ago... we're going to pool all your data... And their customers erupted... They had to roll it back within a week." 00:30:40

Cred (India)

Indian fintech company. Meta reportedly invested $900 million into Cred, with its founder Kunal Shah becoming head of WhatsApp — described as Meta effectively paying over a billion dollars to acquire a WhatsApp leader. 00:37:43

Sound Ventures

Ashton Kutcher's VC firm. Discussed in context of Kutcher departing to start a new deep-tech seed firm with Morgan Bella. "They're in some good names... we've co-invested with them in some deals they've been wonderful to deal with." 01:12:29

Clay

Sales intelligence/GTM platform. Cited as doing a tender offer at $5 billion as the minimum threshold for maintaining employee liquidity. "Clay did it. I know it's much smaller. It's $5 billion, but they did a tender offer at $5 billion." 01:16:33

Klaviyo / Shopify

Klaviyo cited as an example of giving equity to a large partner (Shopify) to ensure alignment even when economically immaterial. "You give 5% of your company to Shopify like Klaviyo did so that they don't destroy you." 00:09:50

Finn (by Intercom)

AI customer experience agent. Mentioned as recently acquired for $3.6 billion, and cited in context of the economics of AI customer support resolution costs. "Finn just got bought for $3.6 billion, right? I got to push the open source thing." 00:59:51

DeepSeek

Chinese open source AI lab. Announced building its own chips, cited as part of China's systematic effort to build a self-sufficient AI stack. "DeepSeek is developing its own chip." 00:00:31

Replit

AI coding platform. Jason used Replit (running Sonnet plus open source) to attempt to solve a complex algorithm problem, ultimately failing after 10 hours before switching to Claude's highest tier.

4. People Identified

Alex Karp

CEO of Palantir. Praised for substantively sharp commentary on CNBC despite theatrical delivery. "I actually thought he was more stable than he normally does... when you strip away all that, the two comments he made were spot on." 00:28:44 Called Dario Amodei a "world historical figure" — a Hegelian reference reflecting his PhD in German philosophy. His Palantir stock rose 9% the day of the appearance.

Sam Altman

CEO of OpenAI. Discussed as one of the most successful investors of his generation, running OpenAI with a "super startup" playbook. "Sam Altman, beyond being the CEO of OpenAI, is one of the most successful investors of our era of all time... he runs OpenAI with that playbook in a way the others don't. Okay? It's like a super startup. How he funds it, how he thinks about it, the relationships, the scaling." 00:09:23 Also identified as a skilled communicator who uses public statements to anchor future outcomes.

Dario Amodei

CEO of Anthropic. Mentioned as owning approximately 1.something percent of the company he founded, and for making statements about AI displacing white-collar jobs. "If you really are impacting a $30 trillion economy... if half those nice middle-class people in the middle-class jobs all across this country lose their jobs to AI." 00:12:07 Karp called him a "world historical figure."

Andrej Karpathy

AI researcher and former Tesla/OpenAI figure. Cited specifically for a comment endorsing vertical integration down to the chip level: "You got to own the compute. If you don't own the compute, you're screwed. A little like the crypto. If you don't own the keys, you don't own the crypto asset." 00:45:26 Rory called him "just super smart" and credited his comment with shifting his view on the chip-building question.

Jensen Huang

CEO of NVIDIA. Cited for being right that blocking Chinese access to chips and frontier models would force China to build its own. "Jensen was right about this... Jensen's point was you better let the GPUs go over there, right? Or they're going to just do it themselves." 00:53:01

Jesse Zhang

CEO of Decagon. Praised for a nuanced public post on frontier vs. open source model economics, tied to problem complexity and latency rather than just price. "Jesse Zhang, the Decagon founder, did a nice post on that just now... Basically, he said, look, when you're trying new stuff or you don't know the problem or you can't bound the problem, you're going to use frontier models." 00:58:09

Mark Zuckerberg

CEO of Meta. Discussed in context of Meta's compute cloud launch and $70B+ AI capex strategy. "He's earned the right. You've got a $100 billion cash flow business... If I was a board member, I'd be saying, I mightn't get it, but you've earned the right to play." 00:39:35 Also flagged for investing $900M into Cred to hire Kunal Shah to run WhatsApp.

Kunal Shah

Founder of Cred (India). Hired to run WhatsApp after Meta's $900M investment in Cred. "Meta invested $900 million into Cred, an Indian company with a CEO called Kunal Shah. And Kunal is now head of WhatsApp." 00:37:57

Ashton Kutcher

Actor and co-founder of Sound Ventures. Departing Sound Ventures to co-found a new deep tech seed firm with Morgan Bella. "He's so successful... very few people are in the position whereby the name is such that they don't have to worry about the firm brand. They just say, I'm a famous person who's been a brilliant investor." 01:13:29 Rory predicted he gets more Google searches than Sequoia.

Morgan Bella

Previously at Andreessen Horowitz and NFX. Co-founding a new deep tech seed VC firm with Ashton Kutcher. 01:11:39

Jack Altman

Raised a large solo GP fund and subsequently joined Benchmark. Cited as an example of moves that "make sense in 2026" but would have seemed strange even recently. 01:12:13

Karri Saarinen

Founder/CEO of Linear. Praised by Harry as exceptionally disciplined — raised only two rounds, refuses VC meetings, focused on returning the fund one multiple. "Carry it linear. The dude is so disciplined and so focused. He's raised two rounds of funding. He never wants to meet VCs." 00:23:37

5. Operating Insights

The Harvey Deployment Model: Pair an FDE With a Domain Expert on Every Enterprise Rollout

The most concrete operational finding in the episode came from SaaStr: Harvey deploys a lawyer alongside a field deployment engineer on every single enterprise implementation. Jason: "Every deployment they do at Harvey has an FDE and a lawyer. Every single deployment has a lawyer... If you have a lawyer and a very experienced technical resource deploying Harvey, which has a high price point, you can afford it. That might be what you need to have a successful deployment." 01:09:45 The implication for any enterprise AI company: the unit of deployment is not just a technical resource but a two-person team — one technical, one domain expert. Most companies are not doing this, and it's why deployments fail.

The Anchor Before the Ask: Float the Smallest Possible Number Before Regulation Forces a Larger One

Jason's read of Sam Altman's 5% government stake proposal has direct operating applicability: when you sense that an external party (regulator, large partner, distribution gatekeeper) is going to extract something from you, name the smallest reasonable number first to set the anchor. "He's anchoring this idea that, hey, 5% will align us... he's sensing the political winds. So I think it's anchoring 5% rather than 50%... Sam is just anchoring it as the smallest possible stake to get ahead of this discussion." 00:15:31 This works because the counterparty rarely has a number of their own; whoever names first sets the frame.

Shower Early-Stage Startups With Infinite Love — The Lock-In ROI Is Extraordinary

Jason argued that the single best long-term investment any platform company or supplier can make is capturing startups in their first 24 months. "Every leader should be showering startups with infinite love their first 24 months. It's the best long-term investment you can get. If there's any lock-in or anything at all, shower them with love." 00:41:43 NVIDIA's compute financing program is the institutional version of this — subsidize early to lock in customers before they have the leverage to switch.

Small Equity Stakes to Strategic Partners Create Disproportionate Alignment

Jason observed — against his initial instincts — that selling even a small percentage to a large strategic partner creates surprising board-level engagement. "I'm constantly shocked how much that brings you into the boardroom... You sell 5% of your company to a $100 billion partner. It don't matter — it is immaterial. But I'm constantly shocked how much that brings you into the boardroom." 00:10:17 The lesson: for critical distribution or regulatory relationships, a 5% stake that's economically trivial to both parties can produce political and operational alignment worth far more than the dilution costs.

6. Overlooked Insights

The Chinese Great Firewall Has Made Chinese AI Independence Structurally Inevitable — And Now China May Reciprocate by Restricting Western Access to Its Open Source Models

This point was mentioned briefly but carries enormous investment implications. Jason noted that even in Hong Kong — far more open than mainland China — he simply could not access ChatGPT, Claude, or their APIs. "What do you expect China is going to do the second largest economy in the world? Of course they're going to build things that are as competitive or better than we are because you can't even use Claude in China." 00:53:01 Rory then added a second-order twist that was almost entirely glossed over: the Chinese government may now restrict overseas access to Chinese open source models. "It would be very significant in terms of competition, the competitive environment if Chinese open source models were removed as an alternative going forward. I think that would be obviously pretty excellent if you are a US frontier model." 00:55:47 If true, this is a bifurcation event: Western frontier models (OpenAI, Anthropic) gain pricing power, while the competitive pressure from DeepSeek and similar open source models — which has been a core reason enterprise customers push back on frontier model costs — evaporates overnight. This is not priced into any current discussion about the competitive landscape.

AI Customer Support Is the One Enterprise AI Category That Has Already Crossed the ROI Chasm — and Its Economics Are Quietly Plateauing at a Specific Price Point

While the broader enterprise AI ROI debate raged, Rory slipped in a precise observation that reframes the entire customer experience AI sector: the industry is quietly converging on $0.50 per resolution as the price point, with LLM costs needing to come in below $0.25. "We're kind of standardizing this industry around 50 cents per resolution... if you can charge 50 cents for resolution, what do your LLM costs have to be? 25 cents? Maybe less." 00:59:51 This is not a general AI insight — it is a specific unit economics benchmark that reveals: (1) which AI CX companies can survive and which cannot, (2) why pressure to use cheaper open source models is so intense in this category, and (3) why Jason observed plateauing quality at the 40% true resolution rate when cost pressure forces step-down models. Any investor evaluating AI CX companies should be running actual LLM cost-per-resolution numbers against this $0.50 benchmark before making a decision.