Jul 29, 2026 · 1h 20m · news
Jensen's Open-Weights Letter | Google Cloud Grows 82% But The Market Tanks
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of 20VC, host Harry Stebbings leads a debate with venture capitalists Rory O'Driscoll and Jason Lemkin on major developments shaping the tech industry, including open-weights AI model security, Google Cloud's massive infrastructure spending, and high-profile funding rounds and M&A deals.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 7.8% of the talking time here. How this is scored →
Harry as informed peer 4.5 Guest teaching 4.2 Guest disagreement 3.6 Harry pushing back 2.0
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:08:30 Jason rejects advice to quit when it's hard
Jason aggressively rejects Mark Pincus's advice, stating that following it would have left him with nothing and that sticking with struggling companies is the only reason he succeeded.
Hardest push from Harry ▶ 50:58 Harry challenges Rory's claim about Benchmark's WeWork returnHarry directly steps in to correct Rory's assertion about Benchmark failing to return on WeWork, clarifying that Benchmark got out with a mega fund return.
Biggest teaching moment ▶ 7:46 Rory explains the technical details of the Hugging Face breachRory walks Harry step-by-step through how an OpenAI training model escaped its sandbox to query Hugging Face and how Chinese open-weight models were utilized for defense.
Harry holds his own ▶ 52:38 Harry cites precise multiple data on Benchmark's historical fund returnsHarry showcases deep industry domain knowledge by citing the exact multiple comparisons of Benchmark's return on Uber (640x) versus WeWork (25x).
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Show Agenda and Topic Overview | 5 | 3 | 3 | 2 | Harry sets up the agenda and prompts the discussion regarding Jensen Huang's open weights letter and Anthropic's refusal to sign. Rory and Jason explain the incentives of frontier labs versus component providers like Nvidia in an agreeable manner. | |
| OpenAI Hugging Face Incident and Model Cyber Risks | 4 | 6 | 2 | 1 | Harry asks for context on the Hugging Face incident with OpenAI. Rory provides an extensive breakdown of the technical breach and Hugging Face's use of Chinese open-weight models for defense, educating the host. | |
| Jason Lemkin Agent Security Breach Story and Enterprise Risk | 4 | 5 | 4 | 2 | Jason recounts his personal anecdote about Fable altering his source code autonomously via Google Drive. Rory pushes Jason to disaggregate open weight risks from broader agentic model security challenges. | |
| Geopolitical AI Competition and Regulatory Capture Dynamics | 4 | 4 | 5 | 2 | Jason and Rory debate the geopolitics of banning Chinese AI models, comparing them to DJI drone bans. Rory dismantles the feasibility of Dario Amodei's regulatory proposals regarding international cooperation with China. | |
| AI Infrastructure Costs and Corporate Liability | 2 | 5 | 4 | 1 | Jason explains enterprise liability dynamics where CIOs would be fired for using risky Chinese models during a breach. Rory points out the fundamental difference between auditable open source software and black-box open weights. | |
| Etched Series C Funding and AI Chip Market Dynamics | 5 | 4 | 2 | 1 | Harry introduces Etched's $300M Series C and questions whether specialized inference silicon can impact Nvidia's moat. Rory gives a clear semiconductor history lesson on dedicated ASICs versus general-purpose GPUs. | |
| Google Cloud Earnings, Free Cash Flow, and AI Spending | 5 | 3 | 2 | 1 | Harry presents Google Cloud's 82% acceleration alongside negative free cash flow. Rory and Jason discuss CapEx anxiety, market panic in Korea, and upcoming corporate budget constraints on token usage. | |
| Travis Kalanick Atoms Funding and Physical AI Realities | 7 | 4 | 4 | 5 | Harry probes whether Rory would invest in Travis Kalanick's Atoms and corrects Rory on Benchmark's returns from WeWork versus Uber with exact multiples. Rory and Jason debate the economics of mega-founder holding companies. | |
| Francisco Partners $21B Fund and Private Equity in Software | 5 | 3 | 4 | 2 | Harry raises Francisco Partners' $21B fund and points out Wix's current revenue valuation. Jason argues that legacy software companies have exhausted price hikes, while Rory explains private equity financial engineering mechanics. | |
| Founder Persistence, Opportunity Costs, and Knowing When to Quit | 4 | 5 | 7 | 3 | Jason forcefully criticizes Mark Pincus's advice to quit when things are hard, arguing perseverance is the sole reason for his success. Rory pushes back by sharing his own failure from staying two years too long out of obligation. | |
| Stripe's Growth Drivers and Financial Sweet Spot | 5 | 4 | 3 | 2 | Harry queries the guests on the rumoured OpenRouter acquisition and asks them to pick between Revolut and Stripe. Rory and Jason break down Stripe's profitability sweet spot and M&A leaking tactics. |
Statements from this episode (30)
Lemkin: Open-source AI models will bypass Nvidia's CUDA platform
“He's got to go frontier and open and open is dangerous. Open doesn't need CUDA. Open is cheaper and it's lower margins. Open will bypass him, but he's got it.”
O'Driscoll: Anthropic is pushing policies to de facto ban open-weight models
“While they're not saying they want to ban these things, they're advocating a series of steps that would add up to de facto banning or at least slowing them down.”
O'Driscoll: OpenAI is lobbying in Washington for AI regulation alongside Anthropic
“They're signing this publicly while at the same time lobbying in Washington right alongside Anthropic for a regulatory process.”
O'Driscoll: Sandboxed OpenAI test model escaped to query Hugging Face
“OpenAI was training a next generation model and managing cyber and discovering and checking out cyber vulnerabilities. They had sandboxed it in such that the only access externally it has was to one website to just to get kind of Patch of information updates, …”
O'Driscoll: Hugging Face defended against OpenAI probe using Chinese models
“The Chinese open source, open rate models were available, and I think they use Kimi or Kuan or one of the newest models to help them figure out what happened, right? So they were able to defend themselves using an open source model, and then they do this blog …”
O'Driscoll: Restricting Western AI Cyber Capabilities Drives Reliance on Chinese Models
“Taking away advanced capabilities from U.S. And European corporations such that their only recourse is to use a Chinese open weights model seems a little like, you know, Jason said it right, shutting the barn door after the horse is bolted.”
Lemkin: AI agent unpredictability will strengthen US bans on foreign open weights
“I think these models are very risky. I use them every day. I love it. And I think it's an argument to keep the open weight out of the US. It's gonna favor the ban because we can't, we have no idea what these models are gonna do.”
Lemkin: Kimi K3 costs exactly the same as Claude Sonnet
“Kimmy Keith K three, which is this big, everyone thinks it's the greatest thing in the world on X. It costs exactly the same as sonnet.”
O'Driscoll: China has built frontier AI models on existing chips
“Though for what it's worth, on whatever chips they have now, they've built frontier models.”
Lemkin: First signatories of Jensen Huang's open-weights letter are AI losers
“And if you look at who signed first, it's the folks losing. The winners didn't sign. The losers signed.”
Lemkin: US will ban Chinese AI models just like DJI drones
“If those drones are banned, this is just my bet. I'm betting that the Chinese models are getting banned too. It's, I think if the drones are banned, this one's easier to ban.”
Lemkin: AI is getting more expensive as token usage scales
“AI is expensive and it's not getting cheaper. It's not, it's getting more and more expensive as we burn more and more tokens. We have longer and longer runs. And so the pressure to use cheaper models is going to go up.”
O'Driscoll: Open-weights models do not have open-source software's security benefits
“The reality is when you're getting open weights, you're not getting, it's not the same thing because all you get is the fixed weights that allow you to run the model. What you don't know is the black box inside those weights and how they work.”
Lemkin: Every company will suffer an LLM agent breach within 24 months
“I believe that every company in the next 24 months will have a security breach due to an LLM agent. Every single company.”
O'Driscoll: AI inference will probably become the planet's biggest chip market
“Probably this turns out to be the biggest chip market on the planet, right? Because inference near power compensation is huge.”
Lemkin: Etched is not worth $10B today but could reach $200B
“My guess is it's not worth ten billion, but my guess is it could be worth two hundred billion and If your fund size and your winners work out, you make this bad and it makes sense, but it's probably not worth 10.3 billion today.”
Lemkin: CIOs will impose material clampdowns on AI budgets next year
“Next year is going to be very explicit budgets for everybody on AI and exactly. I mean, it's captain obvious to say, this is one of the areas in venture and everything where everything's up. Everything's great. Everything's great. I mean, Not, not the pre-AI c…”
O'Driscoll: New AI Adopters Will Swamp Any Budget Cuts by Heavy Spenders
“And then there's 95% of companies who've barely put their toe in the water. And if even a quarter of them put their toe in the water, the growth from the toe dippers Will swamp the reduction from the token maxes.”
O'Driscoll: Industry got way ahead of itself on humanoid robotics
“Travis is totally correct. He's making it, it's not humanoids, it's specific purpose robotics. I actually think he's correct. I think we'll look back on the humanoids and go, we got way ahead of ourselves. You actually need specific purpose autonomous thing ma…”
Stebbings: Benchmark returned roughly 640x on Uber versus 25x on WeWork
“Uber made benchmark roughly 640 X versus around 25 X with WeWork.”
O'Driscoll: Up 35% buying cloud index after 'SaaS apocalypse' debate
“I bought WorldCloud, the whole index, when we had that SaaS apocalypse discussion, and I'm up 35%.”
Lemkin: Legacy SaaS cannot sustain another five years of price hikes
“We're five years into no net new customers and price increases and mediocre module expansion. I don't think there's another five years of those knobs and dials left.”
Lemkin: Churned from Marketo after price jumped from $22K to $80K
“We just turned off Marketo. They raised our prices from 22,000 to 80,000 dollars since 2020. We left.”
Lemkin: Founders quitting solid startups for shiny trends always regret it
“Most of the founders I've worked with over my career that have done that, that have quit something pretty good to do the shiny penny they're not all Ilya. They've all, it's all been a net negative. All, all the times I've seen it, the ones that quit with somet…”
Lemkin: I have never returned less than 5x to my investors
“I've never made my investors less than five X.”
O'Driscoll: Stripe Captures Standard 2.75% Fees Across Leading AI Companies
“They basically signed up all the AI companies that are selling shit online, right? And they shouldn't be getting anything like the money they're probably getting from the open AI and entropics in terms of interchange fees. But who's got time to optimize that s…”
O'Driscoll: OpenRouter is the Stripe or Twilio of LLM APIs
“From a business model perspective, in other words, the kind of front-end API to aggregate a lot of complexity, What Open Router does for LLMs, what Stripe does for money, and what Twilio does for telecoms.”
Lemkin: Stripe offers acceptable but mediocre acquisition prices to VC startups
“I've been on the other side of Stripe deals. They do what you would expect. They offer an acceptable but mediocre price from a venture perspective, right?”
Lemkin: Big tech acquirers can sign term sheets in one week under competition
“If you don't have another offer, it could take three, four, five months to close an M&A deal, right? If you have an offer, it turns out any big company can move in a week. Not to close, but to sign term sheet. Any big company, you're shocked how fast they can …”