00:00Let me just set the scene right, because I also understand the context view of the world we're in right
00:04now is crazy.
00:07Anthropic just filed confidentially.
00:10Ultimator led that round.
00:11I get that there's a limit to what you can say about that.
00:14But you're also on the cap table of OpenAI, SpaceX.
00:18This isn't also about IPOs in isolation.
00:22Alphabet came out and upsized an $84 billion now equity offering.
00:26So actually, my question to you right now is, what is the capital market telling you about the state of
00:32the world?
00:32Is that a good place to start?
00:34That's a great place to start.
00:35Thank you for having me, Ed, Caroline.
00:36It's lovely to be here.
00:38You know, I think at the highest level, I think you nailed it.
00:40It's like AI has gone from being a technology product cycle to one of the largest capital formation cycles, right?
00:48On one side, you've got the picks and shovels.
00:50You know, this is the compute, memory, optics, networking.
00:53You go down the list, they are selling these products to, let's call them the miners.
00:58These are the labs.
00:59Google just raised $84 billion.
01:01OpenAI has a historic raise, $122 billion.
01:04Anthropic, of course.
01:05And, you know, you look at the same week, NVIDIA just announced an $80 billion stock buyback program.
01:12SK Hynek, same thing, $8 billion of stock retired with more dividends coming.
01:16Those are like sort of contradictions in the function of the market.
01:20They're opposite.
01:21And that's the line that I wanted to highlight.
01:23On one side, you've got the folks who are helping you scale AI.
01:27These are receiving the CapEx.
01:29The others who are spending the CapEx.
01:31And I think that's the line.
01:32That's the biggest mental model I want to leave the folks with, which is as you look at these businesses,
01:37are you on the side of receiving the CapEx, which would be the compute layer, the energy layer, the memory
01:42layer, the networking layer?
01:43Or are you on the side of spending?
01:45And I think that's the biggest divide in AI right now.
01:50But at the moment, we have seen rocket ship when it comes to AI infrastructure and those who've been receiving
01:57all the capital expenditure.
01:59And then people are just so desirous to help fund those that are going to be out there and raising
02:03it, whether it be open AI, whether it be SpaceX, whether it be Anthropic, whether it be Google, Alphabet.
02:08Is there enough room for all of these large language models?
02:11There's always been this argument that they'll be commoditized.
02:13When you're sat thinking about being in block AI and SpaceX and at the same time open AI and Anthropic,
02:19should you be?
02:21You know, I think there's this narrative of companies doing a binary thing.
02:26Oh, you're the consumer AI company or you're the enterprise AI company or you're the space AI company.
02:31I think the reality is a lot more multifaceted.
02:34Let's take open AI, for example.
02:36I think the company's mission is getting AI to the market holistically across consumer.
02:43You've got just under a billion users on ChadGPT.
02:45On Enterprise, they've now got 5 million Codex users that are doing really well.
02:49And then a long tail of other bets with hardware, with robotics, with obviously the Frontier Enterprise offering.
02:57SpaceX, which is, you know, the two largest markets on the planet, space and AI.
03:03And so I think the…
03:04And it's selling compute.
03:05And now they're selling compute.
03:06Elon Web Services.
03:07I actually asked you about that.
03:09You know, there's so many directions we could go.
03:12But the space business of launching Starlink, two trillion in the TAM.
03:16Get it.
03:17Enterprise AI, you've kind of answered that.
03:20But stealthily in there, they are becoming a hyperscaler, right?
03:24If you think about the anthropic arrangement, they're almost hedging by saying in the interim, we're really good at building
03:31data center.
03:32Well, I would say it's the Switzerland, right?
03:35It's the Switzerland approach.
03:36You know, for a long time…
03:37You're the first person to say that to me.
03:39For the longest time, we were holding our breath for, hey, you did this deal with Cursor.
03:44And the big question was, well, a lot of Cursor revenues come from the great anthropic models.
03:48Then they got anthropic on the platform.
03:51Then, obviously, they're building their own models.
03:52And so I think of the XAI business as the Switzerland business.
03:56We're now selling compute.
03:57We will sell models.
03:58We'll have a coding product.
03:59We'll have other products down the line on consumer, et cetera.
04:02And, again, it's a holistic approach.
04:04And that's why the TAM is so large.
04:06Look, we love in the media and more broadly to be putting anthropic and open AI in some race against
04:12each other.
04:13But do you have, as a private market investor, any pause that maybe there isn't going to be the demand
04:19for this wall of public offerings that in many ways are going to be liquidity to your ultimate?
04:25Look, I think for the longest time, we've had a large part of the market finding ways to get access
04:31to the biggest super cycle in AI right now.
04:36And I think that that moment has arrived.
04:39IPOs are cyclical, as we know.
04:40You know, the last big cycle was four or five years ago in 2021.
04:45And based on everything we know, there's a lot of demand for these products as they go out.
04:49But in terms of the demand for the capabilities, look, I think in December last year, we crossed a level
04:54of capability with coding models in Opus 4.5 that has pushed it to the next frontier.
04:59On consumer, as everybody's already experienced.
05:02And then yet to be proven is robotics, manufacturing, the hardware device that OpenAI is working on and others.
05:08So a lot more to come.
05:09And I think more participation is better for us.
05:12Because you're teaching at Stanford, and you've kindly invited me to come and check out a class, and I will.
05:18And one of the questions that you pose in a series of guest speakers, right?
05:22You know everyone in industry and your fellow investors on the cap table.
05:26You ask them, like, what's one idea?
05:28You're long, one idea.
05:29You're short.
05:30I explain that.
05:31But why do you think that's a useful mechanism for understanding the world?
05:35Well, I like to hold the speakers to what they say.
05:38I'm going to ask them the question next year and have a scorecard of what happened to their long, short
05:42picks.
05:43It's my favorite game to play with founders.
05:45I learn a lot as founders pick their most optimistic idea, their most bearish idea.
05:50And I'll give you the theme.
05:51You know, the theme in this class was more than half of the speakers were long.
05:57Let's call them the picks and shovels providers.
06:01Compute, most common theme.
06:03Energy was the second most common theme.
06:05That is ultimately the bottleneck at scale.
06:10The most common theme on the short side was folks who were not innovating.
06:14Incumbents were not innovating.
06:17You know, it's a fun time and good to bring the conversation back.
06:20Are you short or what are you not looking at in a startup space as you're thinking about a wall
06:25of liquidity coming to Altimeter?
06:27What would you not like to invest in very quickly?
06:30Well, I'm an optimist by trade, by profession in my blood.
06:34So I'm very optimistic on a lot of things going on.
06:37I think the simple mental model that I have is if there's a business that gets hurt by the scaling
06:43laws.
06:43It's one of the scaling laws.
06:44The scaling laws are if you provide more compute, more resources, more data, more algorithms, intelligence goes higher.
06:50If you're a business that is hurt by this empirical observation that intelligence is going to go up,
06:55that's a tough place to be because we believe intelligence is going to go up a lot.
Comments