00:00So, where it's the first innings, how much more can actually the economy be able to support an infinite demand,
00:07as it seems, for compute and for the latest AI agent?
00:12It's great to be here. Thanks for having me.
00:15It's great to be here. Thanks for having me.
00:19Look, we are just at the beginning stages. I say we're not even in the first innings yet.
00:24The biggest constraint today is data centers and gigawatts of compute.
00:31But in terms of demand, I view it as infinite.
00:34We haven't even scratched the surface of robotics starting to infiltrate our workplaces, our homes.
00:42And we're just beginning to have agents in the enterprise that can do things across the enterprise at the intelligence
00:49level and at the speed of humans.
00:51And so we're just at the very beginning.
00:53Just at the very beginning of the pushback, as well, in many ways.
00:58And I'm interested, actually, as to first whether we push back against some of the valuations that we're seeing at
01:03the moment, Ethan.
01:04I got to speak with Gradient yesterday, who'd just been raising a new fund, talking about some of the numbers
01:09involved in the latest funding rounds.
01:11Darren Shiraz, just hear what he had to say.
01:14We actually think that there is a bubble in certain parts of AI.
01:17There are bubbles, there is a bubble in foundational model companies, contenders to OpenAI, contenders to Google, contenders to XAI
01:25and Anthropic.
01:25When it comes to a lot of these other companies that are raising a billion dollars here or there, it
01:30seems as though those are a lot of venture capitalists or investors that want to chase the next OpenAI.
01:36But realistically, the moat at the model layer is just capital.
01:41How much money can you raise?
01:43Coz was backed OpenAI.
01:45Are people in VC still trying to chase investments right now?
01:49Are the rounds too high?
01:50Some of the foundational models and even the applications?
01:54Look, it's clear that we have clear leaders at the LLM level for the foundational models.
02:00And today here in the U.S., it's really full companies, OpenAI, Anthropic, XAI and Google Gemini.
02:07But there's a lot of research being done on novel model architectures, things like state-space models, neurosymbolic models, math
02:16-based models.
02:17And there is a lot of optimism that there will be some kind of leap ahead of the transformer architecture
02:23that was so fundamental to chat GPT and GPT-3 that caused this AI boom.
02:30And so, you know, to some degree, it's justified that, you know, VCs like us and others are going to
02:39back teams that are going to take a crack at building the next OpenAI and Anthropic.
02:44And if there's a company that has that potential with a research team that has the caliber to produce a
02:52novel model architecture, it's the kind of bets that we should be making as an industry.
02:57Now, are the valuations too high?
02:59You could argue that.
03:01But we have to work backwards in some ways.
03:03These teams need to hire very expensive researchers.
03:07They need to have gigawatts of compute to train models.
03:12And that's very, very expensive.
03:14And as you think about the cap table, we have to make sure that the founders and the team are
03:19incentivized.
03:19And to some degree, if they're going to raise hundreds of millions or a billion dollars out of the gates,
03:25it means working backwards.
03:26If we're going to buy 20% as VCs, then the math kind of takes care of itself in terms
03:32of the valuation.
03:33So that's where we are today.
03:34Somewhat justified, a little overvalued.
03:37Ethan, can Silicon Valley venture, can American venture, can developed world venture support the sort of numbers that are involved
03:45if Middle Eastern money becomes harder to access at this moment?
03:50Yes.
03:52Look, today, there's a lot of debate about whether the model companies should be profitable and self-sustaining.
03:58I don't believe that's the right approach.
04:01We're just at the very beginning.
04:02And there's a lot of R&D left to do.
04:05There's a lot of business model transformation that's happening today, obviously, with the SaaS-pocalypse.
04:11Fundamental existing business models are being disrupted heavily.
04:15And so it's a time where we are, as an industry, still figuring out what to do in terms of
04:21how to fund all these investments.
04:24Today, at least, globally, the foundational model companies have been able to raise capital.
04:30But at some point, I do believe it becomes somewhat sovereign.
04:34And at Coastal Ventures, we have been investing behind themes in companies like Savum and Sikata in Japan, where they're
04:42raising money and being funded by governments.
04:45And there is a question, and we're seeing some of this play out with the D.O.W. between OpenAI
04:50and Thropic.
04:50Does the government step in and, through either big contracts, continue to fund some of the developments and the investments
04:58that need to be made into compute and to the R&D?
Comments