The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
00:28:48 · Jason Lemkin
[00:28:48] S0: Qwen lied. It says, I'll call no tools. And Claude will lie and Aster will lie too, right? It said, I'll call no tools. And what did it do? It called servers in Beijing and in Singapore and uploaded data. It said, I won't do it. So you really have to go deep in these open weights and really test it. And depending on how you test it, nominally what happened with Cohere only happened one in a 100,000 calls. So a bunch of folks tested it a few times and they sort of found this issue a couple months ago, but you had to run it at scale. But at Saster, we run 40,000 API calls a day. It doesn't take very long to get to 100,000. If you're GE or whoever in the Fortune 500, you're running millions and millions a day. That could be every hour, every ten minutes it's uploading data. And again, this doesn't appear to be even malicious per se. This appears to be a training error in a sense. Right. But who cares? Like if you're a CISO, who cares if it's a training error or if it's a spy or even worse, if it's a training error and some other actor exploits it, which is the job of all of our intelligence services to exploit these things, To have pagers blow up. I don't believe Alibaba itself, it doesn't look like the engineers did it as an exploit, but the government could take it over or force them to. They're just... It's... Why do you want to have this meeting? Let's just use Haiku 5.5. It's pretty good. Guys, your budget this month's $5,000,000 I just... And Anthropic, if you... The other thing, if you look, Anthropic is, it's on the GSA list or whatever it's called again. It's back in the government's good graces. It's across all the GSAs with a 40% discount. That means a couple months ago, they were persona non grata. Now every government agency can use it with a 40% discount and it's already pre listed. Why would you use Kimi? Plus, also think, and again, listen, these are great companies. I also think the fact that you have to pay Kimi, what is it, 30% of your gross... Of your revenue to use it at scale for the inference providers, there's just a lot of complexity here. It may not be worth it at some level. It may not... The juice may not be worth the squeeze for many people.
00:30:39 · Harry Stebbings, Jason Lemkin, Rory O'Driscoll
[00:30:39] S1: I'm just trying to understand, though. Last week, we said that Modal at 15.7. Oh my god. Yes. You buy. And then you're saying, no. No. No. Chinese open source is not... Is it gonna decline significantly? It's great for Anthropic and OpenAI. That is inherently bad for Modal. I'm just [00:30:57] S0: trying to for Fireworks, Modal. I I don't know how bad it is for OpenRouter. I'm just not sure. Right? Is it a... Maybe I should say this [00:31:03] S2: how You're you're making a fair point, and I I didn't participate in the investment committee because I don't. But I think to be fair, Jason, the assumption is that there is demand for an open weight model, which is obviously the demand that inference providers primarily meet. And you just got to believe that there's a bunch more alternatives, that there will be some US alternatives. There also will be people for whom China is okay, and there will be people for whom they want the poolside NVIDIA one. And provided something exists in The US, that business is fine. And I do agree with you. I think it's not as clear a run as you have when you own your own close source frontier model, but there is a role for inference provision just because there's just so much raw demand for an alternative other than the frontier models.
00:31:47 · Jason Lemkin
[00:31:47] S0: To answer the question, think Harry's got a good point. I think both are true. A while ago, Rory brought up that there probably won't be another deal like Manus, right, which was a direct investment. I mean, I'm paraphrasing you, right? But after that one got unwound, no one wants to do that deal again, right? This sort of direct investment in an Asian startup. This isn't investing in a US based folks running open weights, but not open source models is not the same as Manus. But there is risk. And I think if twenty four months ago, most VCs would not want to take that risk because it's a clear risk. You don't have to fast forward to today with the Cohere Qwen issue to know it's a risk. It's just the reason on the IC thing, the growth is so phenomenal that we're taking risks today that you wouldn't take if we didn't see growth that was jaw dropping. That's the job of venture today. And it's okay if it blows up. It's okay if fireworks or whatever goes to zero in our portfolio at IC because we have other investments. That's the job. And we're taking risks. We wouldn't take this risk if it was a triple triple double double, but it's not a triple triple double double. But it doesn't mean that it doesn't mean that the risk level didn't just go up in the last week. A lot. And these DeepSeek being blocked by the U. S. Government and this whole Qwen thing and dialing home and Anthropic being on the GSA schedule. These are negatives. These are negatives. I think what we may look back is, and I still believe in the IC, we may, this is just my sense, I think there's a 56.3% chance we look back on OpenRouter and this was just a moment in time. Was a moment in time, going back to earlier the show, when OpenAI stumbled. So Anthropic was the only alternative for state of the art coding. And then the Chinese open weight models got really good at coding. That's what Elon said. Like, it's really good. But he even said they'll catch up the fable by the start of next year. So there was a gap that these models filled. And now OpenAI and others are going to fill this gap. And the low cost models are going to fill the gap. Like they each now have a much cheaper model that's much better. We may see these open weight models go from 60 something percent of open router to 10% in six months.
00:33:43 · Jason Lemkin, Harry Stebbings, Rory O'Driscoll
And that would not be an extraordinary amount of change based on what we've seen. That would not be... This is not sticky. [00:33:49] S1: It wouldn't. But Jason, you've talked very well, I think, in the past about maiming markets in terms of kind of the Fortnite ification, the shrinking land that's available. And to Rory's, I think, eloquent point, there will still be some companies and there'll be some companies that want reflection and poolside, but it is a maiming of an open market. [00:34:08] S2: Think you had to look at it... I understand the point you're making, but I think you've just got to look at it. The inference market is still so big. I mean, we added up all the competitors, and they only had $5,000,000,000 And, you know, we just agreed that the closed source two frontier models are a 100,000,000,000 already. So as long as there's... Yeah. If 10% of the spend is willing to kind of do... Put up with the brain dead of open source to save real money, and you do save real money, then there's gonna be demand for that. Does it get to 30,000,000,000? I don't know. When the overall market is so big, 10% of it is still big. And that's the big Ohio. This is... It's the mother of all markets, so everything else works.
00:34:08 · Rory O'Driscoll, Harry Stebbings
[00:34:08] S2: Think you had to look at it... I understand the point you're making, but I think you've just got to look at it. The inference market is still so big. I mean, we added up all the competitors, and they only had $5,000,000,000 And, you know, we just agreed that the closed source two frontier models are a 100,000,000,000 already. So as long as there's... Yeah. If 10% of the spend is willing to kind of do... Put up with the brain dead of open source to save real money, and you do save real money, then there's gonna be demand for that. Does it get to 30,000,000,000? I don't know. When the overall market is so big, 10% of it is still big. And that's the big Ohio. This is... It's the mother of all markets, so everything else works. [00:34:43] S1: Jason, you teased Black October, but I do think it's significant. Oh, yeah. OpenAI models and Wiz found two very critical PHP flaws hidden for more than a decade that have been exposed thanks to, you know, where we are in the cycle and the sophistication of AI and some models. What do we take away from this, Jason? How do you respond to this? What should we know that we don't know? Help me out here.
00:35:05 · Jason Lemkin
[00:35:05] S0: Well, look. I mean, I mean, we we said months ago on this. I said that everyone's gonna be hacked. The biggest change is, know, back in the old days before the models got good, I remember my CEO said the only reason we haven't been seriously hacked is no one cared. Now anyone can be hacked in seconds. People attempt to hack faster every day. So everyone's going to get hacked. And this PHP thing actually is worse than it looks because there's so many websites in PHP that no one's going to update. It's great that a lot of stuff runs on WordPress or that Replit and Lovable and others immediately ran like gateways or barriers to protect you. I mean, they're not all running unpaid, but they immediately ran. And there's a lot of stuff that went up and people, people released patches and updates. But you know, a lot of government and other sites and nonprofits, they just took their websites down and who's going to fix it? I mean PHP is still like 40% of the internet, right? And I bet it's more of old crap. I bet it's like 80% of old crap. So like, this is like Mr. Robot Live. Like all these old websites are running on PHP. No one's patched them since this happened. No one cares. The IT team was laid off three years ago. They're not vibing nothing. And we're going to find our way into all this. It's just going to get worse. Everyone basically, including Sam, just said it's going to get worse and worse and worse and worse. Right? And I think we're still grappling with the fact that everything that is a threat vector will be exposed almost in real time.