Jul 23, 2026 · 1h 27m · news
Should the US Ban Chinese Open-Source Models | OpenRouter's Chance To Sell | Stripe Buying PayPal
Clips from this episode (5)
Short vertical cuts produced from the tape, captions burned in. Where a cut lands on a statement from the ledger, its card says so.
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 the 20VC podcast, host Harry Stebbings and venture capitalists Rory O'Driscoll and Jason Lemkin discuss the impact of Chinese open-weight AI models, AI middleware and routing economics, Stripe's acquisition bid for PayPal, and late-stage venture capital dynamics.
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 11.8% of the talking time here. How this is scored →
Harry as informed peer 3.4 Guest teaching 3.1 Guest disagreement 2.3 Harry pushing back 2.6
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 25:27 Jason rejects VC-centric framing
Jason forcefully interrupts Harry's line of questioning about selling at a $6 billion valuation, declaring that the question triggered him because it represents a narrow VC return perspective that ignores founder incentives.
Hardest push from Harry ▶ 12:06 Host challenges guest on on-prem AI securityHarry directly refuses Jason's premise about Chinese open model security risks by raising the standard enterprise argument that hosting models on-premises completely eliminates backdoors.
Biggest teaching moment ▶ 25:27 Jason educates host on founder vs VC financial logicJason dismantles Harry's selling scenario by explaining that a 3x exit does not change a founder's life stack risk profile, requiring a 10x trajectory to justify continuing.
Harry holds his own ▶ 1:05:56 Host demonstrates market expertise on valuation multiplesHarry shows deep domain knowledge by breaking down actual market data, pointing out the irrationality of paying $300M for $2M ARR at Series A compared to paying 10x ARR for proven scale at Series B or growth rounds.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Impact of Chinese Open-Weight AI Models | 2 | 2 | 2 | 1 | Harry introduces the topic of Chinese open-weight models Kimi and Qwen catching up to frontier Western models. The guests lead the conversation, discussing evaluation accuracy and the political fallout from OpenAI policy staff tweets. | |
| AI Model Policy, Regulation, and Exfiltration Risks | 3 | 4 | 3 | 4 | Harry challenges Jason on data export fears by pointing out that CIOs are told on-prem deployments remove backdoor risks. Jason pushes back forcefully, arguing that complex LLM behavior cannot be fully audited or trusted regardless of on-prem hosting. | |
| OpenRouter M&A Rumors and Model Routing Infrastructure | 3 | 6 | 5 | 3 | When Harry asks if a board should sell OpenRouter for $5B to $6B, Jason calls out the question as a classic VC-centric mindset. Jason educates the host on how founder economics differ from investor fund returns. | |
| Fireworks AI Funding and Infrastructure vs Application Value | 4 | 2 | 1 | 2 | Harry demonstrates strong topic knowledge by presenting the detailed deal terms and operational metrics for Fireworks AI. The guests agree and elaborate on inference unit economics and capex requirements. | |
| Custom Micro-Models, Data Labeling, and Model Economics | 3 | 2 | 1 | 2 | The conversation moves to fine-tuning, domain-specific micro-models, and data labeling economics. Jason shares hands-on experience building an agentic app and explains how domain expert labeling vastly outperforms generic foundation models. | |
| Stripe's Strategic Acquisition Bid for PayPal | 3 | 3 | 2 | 2 | Harry prompts the guests regarding rumors of Stripe making an acquisition bid for PayPal alongside Advent. Rory and Jason explain public company board dynamics, Delaware law fiduciary duties, and the standard M&A premium negotiation dance. | |
| Venture Capital Market Dynamics and Tranche Financing Structures | 6 | 3 | 3 | 5 | Harry takes an active expert role, highlighting a sharp market disconnect where early-stage Series A valuations command 50x-100x revenue multiples while scaled growth companies sell for 10x forward revenue. He defends the risk-adjusted value of growth stage investing. | |
| Private vs Public Tech Markets and Hardware Supply Chains | 3 | 3 | 1 | 2 | The panel discusses private versus public tech market inversions, highlighting how pre-revenue speculative companies SPAC while cash-flow monster companies stay private. Rory explains the contrasting dynamics between collaborative chip supply chains and predatory DRAM memory pricing. |
Statements from this episode (35)
O'Driscoll: Massive AI infrastructure spending will require paying applications eventually
“Some point, the people spending a trillion dollars a year are gonna want some apps to pay for all this.”
O'Driscoll: 10x annual growth with positive margins covers all overhead
“If you're growing 10 X year on year, and you have any kind of positive and improving gross margin, it just covers all the nut.”
Lemkin: Half of OpenRouter traffic is for Chinese-created AI models
“If you look at you know, if you look at open router data, half the traffic's through China, China created models”
O'Driscoll: Chinese AI models lag Western frontier models by 6 to 9 months
“They're not state of the art compared to the frontier models, but they're, you know, six, nine months behind, depending on how you measure it.”
O'Driscoll: Chinese AI model Kimi K-3 has 2.8 trillion parameters
“Conversely, Kimi K-III is, I think, a two-point-eight-trillion-parameter model.”
Lemkin: The US government will never use Chinese-built AI models
“It's difficult for me to imagine. The federal government's ever going to use a China built model at this point in the U S right.”
O'Driscoll: Chinese open-weight AI startups reach $50B–$70B valuations
“Because, you know, the open weight models in China are getting 50, seventy billion dollar valuations.”
O'Driscoll: Chinese AI models rely on distillation illegal in the US
“Maybe it's because the dirty little secret is a lot of their advantage is distillation, which you can't legally do if you're US-based.”
Stebbings: Offered $5M SPV allocation in Chinese AI startup Kimi
“I was offered Kimmy today, by the way, Rory at 20 Berlin. This was the, it fell into my inbox. I have an SPV for you. Do Kimmy at 20 Berlin. We're oversubscribed, but we'll make room for five million for Harry.”
O'Driscoll: VCs cannot force a startup sale without core founder buy-in
“Even if, as the VCs, we have board control, 70% ownership, and a drag along for the founders. Jason knows this. If the founders who are core to the business don't want to sell, it's not going to happen.”
Lemkin: Founders shouldn't reject major buyout offers unless 10x upside exists
“I think for a founder, when you start to get into nosebleed offers in absolute terms, it has to be 10 X to go for it. It's not worth it for three X. It is not.”
Lemkin: Ideal startup sale timing is right before market recognizes commoditization
“And the perfect outcome is to sell the moment it becomes commoditized, but before everyone fully realizes it, that's when they'll give you the money, but that's before the value decrease rather than it creates in it.”
O'Driscoll: AI inference providers will become significantly more CapEx-intensive
“If I'm buying data center compute and then effectively selling data center compute with hosted LLM, at some point I'm going to want to own my own data center assets to have more control of my destiny, which means vertically integrating downwards, which also me…”
Lemkin: the AI application layer era is not here yet
“All the good investments sure seem to be in the infrastructure. So I'm waiting for the era of the application Layer in AI and making bets and seeing some good stuff, but I don't believe it's here yet. I actually don't believe the application layer is here yet.”
O'Driscoll: AI infrastructure spend hits $900B while AI apps struggle past $50B
“Zooming out a million miles, you know, my mental model is, I divide the AI world up into three buckets. It's the making AI, the infrastructure layer, right? And you're right, the spend there is eight, nine hundred billion dollars a year. Then there's the two f…”
O'Driscoll: OpenAI and Anthropic training data spend exceeds non-Cursor AI app revenues
“There's probably more money being spent on training data for the foundation models, you know, the Mercosurge and that, than pretty much any app company outside of Cursor. In fact, probably the sum of all the apps companies outside of Cursor, right, you know, a…”
Lemkin: Generic AI model usage will drop strictly to prototypes
“Some version of this, I think folks that are going to want to use at any application level, folks that are going to want to use just the generic models is just going to decline to, Prototypes, right? Prototypes and proofing.”
Lemkin: Generic LLMs are mediocre compared to expert-labeled domain models
“The generic LLMs, which are a sea of mediocrity combined into one giant LLM. Okay. Every mediocre history professor, every, every mediocre doctor that doesn't even know what caused your runny nose is in the LLM. But if you get the best people training it on th…”
O'Driscoll: Hyperscaler RPO depends entirely on foundation model commitments
“Because, let me be clear, if you know the answer to that question, that one question, you know the answer to the entire direction of the U.S. Stock market for the next two years. Because all the hyperscaler RPO, all of it, is a function of the commitments they…”
O'Driscoll: 20x-revenue AI valuations demand Microsoft-level 40% margins
“But as you're pointing out here, there are real, even at the margin, even after you've fully paid for your training costs, there are real physical costs to serve these models, and you've got to cover your nut. You got to cover the marginal cost of the model of…”
O'Driscoll: Stripe and PayPal Each Process $1.8 Trillion Annually
“They both process kind of 1.8 trillion dollars a year, right?”
O'Driscoll: Stripe Generates Around $6 Billion in Net Revenue
“Stripe supports revenue net, which is around six billion plus or minus.”
O'Driscoll: Stripe's Annual Growth Rate Is Between 20% and 30%
“It's between 20 and 30, so it's not that much bigger, Jason.”
Lemkin: PayPal's Board Will Ultimately Accept Stripe's Acquisition Offer
“The board rejected it, right? And the fact that the board rejected it means to me that they're going to accept it.”
Lemkin: Stripe-PayPal Acquisition Premium Will Land at Around 35%
“So it's probably just a dance from 28 to 35. And then it gets parked with advent to while they figure out antitrust and capital issues.”
Stebbings: Hot AI Series A valuations hit $500M on $5M revenue
“Is I'm seeing Series A is priced at three to 500 on two to five million in revenue, but I'm finding the B at a hundred million in revenue priced at one to 1.5. It's a three X price increase for a 50 X revenue increase.”
Lemkin: Series B VCs now need seed-level picking skills to survive
“I think the real pressure is, it means anything below that growth stage, you better be a damn good picker. Because it used to be when Rory and I met, Series B, even into Series A, you actually didn't have to be a good picker. You just had to be good at math an…”
O'Driscoll: 40% of early 2025 unicorns get marked up within 18 months
“We looked at every, I mean, if you look at all the unicorns that were minted in Q-one or Q-two of 25, by the end of Q-two 26, at least 40% of them will have had a subsequent markup.”
O'Driscoll: tranche structures price away the growth-round excess return, and that is how the trend ends
“So now what you're seeing with these tranche deals is, You know, nature abhors a vacuum, and sequoia abhors leaving a dollar on the table, so what's happening is people are realizing everyone wants these growth rounds, and this is how these trends end. Everyon…”
O'Driscoll: Heavy Structure and Excessive Debt Are Worse Than Two-Tranche Rounds
“And I will say they're silly, but they're not horrific. I mean, if you look at that versus other alternatives, like, you know, taking a high price but with a ton of structure, real structure, that's a worse mistake. If you look at it, you know, not raising mon…”
O'Driscoll: SPACs take venture-grade risk public while the best venture assets stay private
“If you think about the kind of companies that should be private, and the kind of companies that should be public, companies trying to do, you know, next generation nuclear products should probably be private. As yet, there's three of them that are public, they…”
Lemkin: C Squared IPO'd at $3B Market Cap with $1B Run Rate
“They're doing a billion dollar run rate growing 16%. Right. And they IPO'd at, with a three billion dollar market cap.”
O'Driscoll: Nvidia and TSMC Have No Written Contract Despite 30-Year Partnership
“You have the NVIDIA relationship with TSMC, which famously, they don't even have a written contract. They've dealt with each other for 30 years.”