IPO卷土重来：科技巨头为何终于上市 | All-In流动性IPO圆桌

Hey, 2026 could be an all-time record for IPOs. The AI IPO of the year so far. That company is Cerebras. Cerebras Systems founder and CEO Andrew Feldman. We are participating in something extraordinary on everything we do. We are the fastest, bar none. Will Marshall is the co-founder and CEO of Planet Labs. Space and AI are really a match made in heaven. They're getting married. In fact, just like Google figured out how to index the internet and make it searchable, we are indexing the Earth and making it searchable. He's got his glasses, the famous red glasses. Brad Gerstner is here, founder and CEO of Altimeter Capital, a leading tech investment firm. I believe that the wave is the biggest wave in the history of technology, will be incredibly beneficial for America. I'm rooting for all of them because I'm rooting for America. Ladies and gentlemen, please welcome Brad Gerstner, Will Marshall, and Andrew Feldman. On the couch, we switch on the couch.

Nice to see you, my friend.

Hey, big boy.

Nice to see you. Last time I saw you, we were in Davos.

Yes. We were in Davos causing another drop. Another J.L. Do you hear that little Davos?

We were just—it was pre-IPO. We're chopping it up with Davos.

We're in Davos. Hanging out at Davos.

Well, no. Listen, everybody knows the story. I'm supposed to go on my yearly Japan ski trip. Sacks calls me with Tucker.

Yeah. Well, anyway, we don't drop that name, but I'll pick it up for you. Put it over here.

Tucker. Anyway, so I cancel on Tucker. I cancel because Sacks calls me. He says, “Listen, POTUS needs you, the world's greatest moderator in Davos.” I said, “No problem.”

I said, “Sacks, POTUS, and Davos.” So I said, “When?” He says, “In 3 days.” I say, “You got it.” I go, and they give me a badge. It's this special green badge, and they buzz you through security. I look at the monitor, and it says, “Jason Calacanis with Donald J. Trump.”

Oh, wow. How did you feel?

I thought it was hilarious. So then I went, and we did a great interview there. We did 6 or 7 of these great All-In interviews, and it was fun.

Let's start this, because the 2 of you guys run 2 of the most interesting and consequential newly public companies in the stock market. Andrew Feldman is the founder and CEO of Cerebras. Will Marshall is the founder and CEO of Planet Labs. But you are also the insight and a gateway for all of us to understand these 2 big trends. One is in AI silicon; the other one is in space data centers. I think it would be a really interesting thing to—

And emerging. Yeah. But let's just take 1 step back. You just heard the last conversation about being public, going public early. Let's just talk about that, because I'm very curious. How's it been? It's been 3 weeks or so for you. It's been about a year and a half or 2 years for you. Was it everything that you thought it would be?

What's clear so far is I need to upgrade my name-drop game. That was a tour de force.

But by the way, you were in Davos with—

Look, I think you do all this work, and I think it's really difficult to overestimate the amount of garbage that's involved in going public. The number of meetings where you look on Zoom and there are 130 attendees, and the amount of times you review these documents and the commas move—there's just no value added.

You go there, and you have this enormous event, and the next morning you've sold no more stuff. Your engineering projects have made no progress since the day you weren't public. You go back to work, and you have some new constituents that you have to address and communicate with, but the core parts of your business—you have more money in the bank, but not a damn thing changes in the important parts of your business.

If you still need new supply, or if your relationships with your vendors are bad, they're still bad. If they're good, they're still good. What we've seen is your employees have a party, everybody's really excited, and you put your head back down. You high-five, and you go back to work.

Right.

Can I just give a little context, and then I want to hear from Will? Andrew, we were investors in Cerebras. I was on the board a year earlier, when we were trying to go public. Aside from just being a warrior who weathered a decade's worth of storms that would have taken out any normal human being, the path to going public for Cerebras was a particularly challenging one.

One of their investors was the UAE, so there were questions about CFIUS. Under the Biden administration, it was challenging to get public. My observation, outside looking in, is that everything was really hard until it got really easy—like 9 and a half years of really hard, and then 12 months of really easy, where everybody wanted to get in.

They priced the IPO at $18.50, which was up. The range was taken up 2 times. The stock opened at $32.00 a share, I think. Today, it's at $23.00 a share, a $5–6 billion market cap for a business like this. Andrew is just one of these people who says, “Let's get back to work and build.”

But my just add-on question to that is, from an employee-morale perspective and a distraction perspective, you got a lot more capital and a lot more profile over the last 3 weeks. Presumably, it's easier to sell to enterprise customers today. Net-net, if you were advising me, if I was in a similar position, would you say, “Go public”?

I think the first thing is, a lot of people asked us about how we got the timing right.

I think the answer is by getting it wrong for a decade. That's really the right way to get timing right. We've been at this for more than a decade, and we brought everybody who'd been with the company for more than 9 years to share it, and we brought their families.

First, I learned that engineers owned ties. I didn't actually know that. Second, I was surprised at how big a deal it was for them and their families. They were really proud, in a way that their parents might have heard of it, or that somehow this was like a bar mitzvah or quinceañera.

You had these children of immigrants. One of our leaders, whose father was a Chinese immigrant, said, “I thought it would have happened faster.”

But I think we are, by nature, people in the trenches. We love solving hard problems. When we had this excitement, everybody went and they were so excited, and we had a party. I think it gave external validation. Then everybody turned around and said, “Now what? Are we now back to work?”

And so you started off kind of bang right out of the gates. Will, you had a little bit different experience in terms of your entry to the public markets, but over the last 12 months, your stock has gone from $5 a share to $50 a share—some 10x move in the public markets. Talk us through the other side of this, where you come public, nobody really notices until they notice.

Well, we were one of the first space stocks, and I think people just had no idea what on earth was going on in space, how it was changing everything. They were just like, “What the heck is that?”

But I have similar opinions. In the end, you've just got to get on with executing the business. Going public gives you access to liquidity for early shareholders, whether that's early employees or early investors, and that's great. It gives you cash for the company, and that's great.

I do think it helps your business as well, because the maturity event gives you more credibility with various customers. For us, we work with the biggest agricultural customers, big governments, civil governments, defense, and intelligence. All of those actors want to know you're going to be around.

We have countries that are fully dependent on us giving them information. They don't want us to just disappear. They really care that we're going to be around, and being a public company gives you the kind of force in the world that makes people say, “Okay, you're here to stay, and you have access to capital if you need it,” and so on. It's legitimizing.

You know where the stock is at any one day. We're not focused on that day to day. We're focused on how we build long-term value for our shareholders. The market, I think, has started to really understand where space is going and why it is changing the world.

People forget how space is part of your everyday life. Every time you use a phone, you're using communications satellites, or GPS satellites, or satellite data in some way or another. That's integrated into your lives. You may not realize it, but it's just booming now.

And the story's changed as well, obviously, with SpaceX going public. Has the framing of Planet gone from a data source for people who need data from space and maps to, “Hey, this is a tool to accomplish tasks in the military,” post-Andrew's success?

So is that framing what's driving a lot of it?

I think it's a bit more nuanced than that. Firstly, for the audience's benefit, what Planet does is satellite-based Earth imaging. We have the largest Earth-imaging fleet, with about 200 satellites. They image the entire Earth every day.

Think of it like the Google satellite layer on Google Maps, except it's today's date rather than 3 years old, and we have every day going back. So it's a time-series analysis of everything going on on Earth. That's useful for farmers, energy companies, and civil governments dealing with flooding and fires. It's also useful for security applications, like you're getting at, and it's a wide variety of use cases.

I think what we're seeing is that AI is now enabling—or basically reducing the barrier to entry—so that more people can get access to this. There's a lot more to say on that, but AI is only as good as the data it's trained upon.

What percentage is military? I'm curious. Sorry.

In terms of percentage of revenue and customer base, security is part of the initial thing that we said we would do out of the gate, but it's true that there's a bigger fraction today than perhaps we would have guessed. The needs of the geopolitical situation right now demand what we're doing.

Just as an example, what this does is enable them to see threats around the corner and give them weeks or months of advance warning of things. That enables them to be more likely to do things that stop conflict, so we believe this is really better for the world.

Are you reticent to be perceived as a military company?

Not really, but I wouldn't say we're limited to being perceived like that. We're helping farmers, energy companies, and civil governments. We work with NASA and with a wide variety of other organizations, so it's a bigger play than that.

But back to the space piece of it, rocket costs have come down about 4 or 5 times over the last 10 years, which has helped tremendously. A thing that people don't know, and that is perhaps more important, is that we've had a miniaturization of satellites.

The same satellite that used to cost $1 billion and weigh 20 tons now weighs a few kilograms, or a few tens or hundreds of kilograms, and can do just as much stuff, if not more. It's the same as the mainframe-to-desktop computer revolution for space. Just as moving from mainframes to desktops unlocked loads of applications, this is unlocking loads of applications. Both things are going in combination: launch costs are coming down, and satellites are getting smaller.

Let's build on this. I'd like you to take a few minutes, and then I want to talk to Andrew about the same question. Both of you guys are at the foot of what are probably huge secular trends in technology.

I would frame this as: We are rebuilding the data-processing infrastructure that has existed on Earth in the sky. First you do the satellites, but I would love for you to explain space-based data centers because I think everybody's hearing about that. Are they really viable? What are they? How will they work?

And then, Andrew, this is the rebirth of silicon. We're going to find the next version of Moore's law, which I think is more time-bounded, not transistor-density-bounded. We now hear a lot about domain-specific architectures. Your chip was a complete transformation in terms of the design principles. At Groq, we took a very different approach, NVIDIA has taken a very different approach, and you took a big, pizza-shaped die and said, "YOLO, this is it." And you were right. Just explain where we're going in silicon. Maybe, Will, you start, and then Andrew, you start.

What we're seeing first in space is all these new applications based on data and AI. We're collecting vastly more data about the planet, and with SpaceX, Starlink, and OneWeb, we're transporting far more data around the planet. As you say, we're changing the nature of data using satellites. That's basically doing what was once the province of governments only and giving everyone else access to satellite capabilities.

I estimate there's a $75 billion to $100 billion market just on Earth observation—the kind of data we collect—and AI on top of that, unleashing all those applications. That's the near-term opportunity: applying large language models to Earth-imagery data and unlocking agriculture, energy, civil government, permitting, and countless other applications. This is going to make everything more efficient.

Where we're going is indeed space. We did a study with our partners at Google about 8 or 9 years ago, looking at the costs of data centers on the ground, what it would cost to put them in space, and when it might make sense to do it non-terrestrially. We figured out that when launch costs come down to about $200 to $300 per kilogram, it would be cheaper—simply cheaper—to put the data centers in space.

Today, we're at about $1,000 per kilogram, just over that. But that's come down about 10 times in the last 10 years. On the current trajectory, with Starship in particular, I would expect launch costs to come down to that level in 2 or 3 years. Elon might say it's next week, but realistically, it's at least a couple of years. We're not far away from it literally just being cheaper.

The intuition that helps people understand this is that you would naturally use solar panels for data centers. Data centers are a power problem; it's a power game. You would normally use solar panels because that's by far the cheapest way to get a watt today. But you don't want intermittent power, so then you have to have batteries, gas, or nuclear, and it gets really expensive.

In space, you can put a solar panel in a sun-synchronous, dawn-dusk orbit where you're looking at the sun 24/7. You can have a solar panel that collects and gathers 5 times more energy than a solar panel on the ground, and you don't need batteries or anything else.

The infrastructure for compute in space is literally just solar panels, the chips, and the RF signals up and down. It's actually quite simple. It was just a question of when it would be cheaper to launch all those solar panels and chips into space than to put them on the ground. It turns out that's going to be in a few years.

We're partnering with Google to launch some of its TPUs into space. We've already launched some of NVIDIA's GPUs into space, and we're launching Google's TPUs into space on an early test. There's lots of technology to figure out.

Let's have a conversation.

It's early days, but I think there's no question that within 10 years, most compute will be put in space. To give you a sense, that's a lot of money—trillions—and it will be bigger than any of the other space businesses today, including Earth imaging. This is why we're getting into this.

Do you believe this? Do you believe sending data centers to space makes more sense?

Can you have him explain the business first and then—

Oh, yeah, of course. I think there are 2 parts. Your first question was around the rise of silicon in general, and I think what AI did—and it's rarely framed this way—is allow computers to address a class of problems that, before AI, computers were bad at.

We were bad at images for almost the entire history of compute. We could store them, and that's about it. We were bad at language. We could store it, but that's about it. We could transform numbers; we were magical with numbers.

What AI did, starting in about 2015 or 2016, is open the door—the aperture—to say, maybe we could use computers on images. Maybe we could find insight in images. Maybe not only could we store language, but we could generate it. Maybe we could understand it rather than storing it and regurgitating it.

What this did was open up huge areas of compute that were previously foreclosed, and at the same time, we were adding to those areas.

Between—

Exactly—between them. We're not good at building the clusters in space necessary for the communication between them.

We're not good at doing it on the ground.

We're not good at doing it on the ground. We're really not good at doing it in space. I think this is an extraordinarily important and interesting problem, and one we should be spending money on and attacking.

I've got it in a slightly different time frame, but it's certainly something that will occur. The hard part is whether it's one of those problems where the last 10% is 80% of the time.

Self-driving was a problem like that, right? The last 10% proved to be a decade's worth of work, and just now we're over the hump. We don't know yet, but I think the interesting work they're doing at Planet is really important. The fundamental driver to experiment—to even get insight into whether I'm right or not—is to get down the cost of launch vehicles. Then you can start doing experiments, getting it wrong, fixing it, and figuring it out. Until then, it was unpayable.

For the foreseeable future, you're going to be terrestrial. Explain your business, how you made these critical decisions that took you down a different path—Groq versus NVIDIA versus AMD—and what you think the future of AI silicon looks like.

I think there are 2 parts. Your first question was around the rise of silicon in general, and I think what AI did—and it's rarely framed this way—is allow computers to address a class of problems that, before AI, computers were bad at.

We were taking vastly more images, terrestrially and in satellites. What this did was simultaneously open up this entire area and allow compute to attack it. This is what's underpinning both NVIDIA's growth and all the growth you're hearing about in AI compute. As a processor builder and a hardware builder, suddenly our tools could attack more and different parts of knowledge, and that was the first part to answer your question.

Now, how you do that—there are lots of different strategies and tons of different ways to skin cats. What we saw in 2015 were several things. First, we saw that AI would be an enormous consumer of compute. Historically, for computer architects, new workloads were an opportunity for share to change.

Share changed when the rise of graphics emerged and you got the dedicated GPU. That's how NVIDIA was born. Share changed when cell phone compute emerged, and Intel and AMD, who had fabs and the best architects, got zero share; it all moved to ARM. Share changed in the late '90s when Nortel and all these companies we've forgotten about couldn't build chips and couldn't do data networking, and what you got was Cisco, Juniper, Arista, and this collection of new companies.

So we knew that this new problem would present an opportunity for massive change. We made 2 bets. The first was that dedicated silicon would be the answer. The second was that it couldn't look like a GPU.

Our view as computer architects is that if you want to be 20 times better than somebody, your architecture can't look like theirs. They have enjoyed and eaten all the low-hanging fruit. If you build a GPU, the odds that you're better than NVIDIA, in our view, are approximately zero. That led us to a fundamentally different architecture.

The hard part here is moving data from memory to compute. This is the fundamental problem in AI, and we solved it in a way that very few others had even attempted: We built a very big chip and put memory right next to compute. By building a big chip—a chip the size of a dinner plate, whereas most chips are the size of a postage stamp—we could use a different type of memory.

By using a different type of memory, one that was vastly faster, we opened up all sorts of opportunity. So when OpenAI uses us, we're 15 or 18 times faster than a GPU. That means your answers are delivered more quickly. It means your engagement with the AI is more enjoyable. It means you can use the AI to solve harder problems and not wait.

The way to think about this is to ask yourself the counterfactual question: How big is the market for slow search today? Zero. How big is the market for dial-up? Zero. How long do you wait for a website to resolve before you click away—3 seconds, 5 seconds? You will not wait for AI. We have to deliver it to you in real time. That's what we saw, and that's what we built.

The panel's on going public. There are a lot of LPs in the room, and they need to get liquid. I'm curious about the journey for your investors. You guys went public what year?

2021.

2021 by way of a SPAC. And your VCs were who?

Draper Fisher Jurvetson was one of the earliest. Capricorn, Peter Thiel's Founders Fund, and then Yuri Milner's DST.

Your investors came in, and you went public at $2 billion via a SPAC. Now we're 4 years later. Really, it wasn't until year 3 or 4 that 90% of the value was created. Did those early investors capture this 90% move? Did they stay in it?

Most of them did. Most of them did, which is really smart on their part. Obviously, I think they should hold on even more. I'm a little bit self-interested. No, but really, they did. Google hasn't sold a share; they're our largest single investor. Capricorn didn't until very recently. Basically, most of them stayed really well in, and they got all of that upside. Good for them.

The reason I think this is so important is that there are a lot of LPs in this room who are like, "When a company goes public, give us the shares." "No, no—give us the shares." This is a counterexample, right? This happened to us 10 years ago. We invested pre-IPO at $1 billion. We distributed the shares, I think, at $3 billion or $4 billion, and then it went to $50 billion over the course of the next 24 months. We had people who called us and said, "Why didn't you hold on to the shares?" And we're like, "Because you're pounding on us to distribute the shares." So you're an example.

Now, in your case, Andrew, you have an innovation, right? You're just now public, so all of your investors are still under lockup, like Altimeter. But you guys have innovated with the banks on what I call a dribble lockup. Over 6 months, the shares can be dribbled out according to a bunch of performance hurdles, and SpaceX is going to have a very similar structure.

When did we start this process of the dribble?

The dribble. (laughter) The concept—you started it years ago, but—

Yes.

With respect to the lockup, I think this is the most innovative approach. Andrew, for your investors, if you were talking to LPs right in the room, should Altimeter be distributing the shares when they come out of lockup? How do you think about your VCs holding on to the shares post-lockup?

I think historically more money's made after IPO than before.

Yeah. I think every single study shows that there is more money to be made, both in percentage and in what we care about, which is absolute.

The amount of money that it's possible to put to work in most venture companies is very modest. There are 2 or 3 or 5 outliers, but for the most part, you can only put a relatively little bit of money to work. By the time we get public, there's a lot more money there if things are going well, and the opportunity to make vastly more is after IPO, not before.

If I could just add on that, one interesting question is what's going to happen with SpaceX on this, because a lot of the value is in the future, right? Most of the big tech companies went public at a few billion, not a few trillion. There's a lot of zeros in between those, right? You've got all this upside afterward.

For the equivalent liftoff, SpaceX would have to be aiming at quadrillion valuations. I know Elon has those sorts of ambitions, but you really have to believe in that.

This is kind of the point I'm getting to, right? We have 3 mega IPOs that we keep talking about that are multitrillion-dollar companies. All of that value accrued to private-market investors.

Planet Labs is a great example of venture capital in the public markets, where the 10x has occurred in the public markets. We're all advocates of these companies coming public sooner. Had Andrew had his way, he would have been public 18 months ago, probably at $10 billion rather than $50 billion. That 5x over the course of the last 2 years would have gone to public-market investors. So go ahead.

Way better to be lucky than good.

Yeah.

I hear a lot of people thinking that Anthropic, OpenAI, and SpaceX are the new normal. I actually think the public markets may be shifting back in this direction, and a lot of the companies in our portfolios are now thinking about going public at $1 billion, $3 billion, or $5 billion.

We had this period of a decade where Andreessen was really pushing "stay private forever," and I see the pendulum swinging back. Companies are like, "I want to be like Planet Labs and get public," right? They want to play in the big leagues and do it in the public markets.

Here's what I'll say, maybe just to the 2 of you: Both of you have had enormous pressure because there's visible competition that's always sort of in your periphery. But I do think that getting public sooner, having the scrutiny of public markets, and having the scrutiny of having to deliver sharpens the focus. Steel sharpens steel; iron sharpens iron.

I think innovation tends to get better. The idea that you allow everybody to participate, but you also put yourself in the spotlight, to me is where great things happen.

I agree.

I just wanted to say to both of you, as we wrap, that you guys are an incredible testament to entrepreneurship. We've been talking literally since day 1—me and Andrew, because we went in different paths and then reconverged—and then the same with you, Will.

I'm happy it worked out for you, Chamath.

It's worked out for both of us, so it's fine. You guys are an incredible testament to entrepreneurship, and I just want to say thank you for everything you guys are doing. The next few years are going to be really spicy.

If I could just spend 30 seconds on the next few years, I think it's going to be so exciting with AI and space merging together. We're going to see a takeoff of AI applications. All the cool stuff that we're doing with LLMs now is really based on just the text of the internet being absorbed into these models, which is incredibly powerful already, but they don't know about the real world.

I call them blind. They don't know about that farm field, that flood, or that security situation around the corner. If you give them real-world data, then they can answer real-world problems. That's going to open up gazillions of applications for these AI models.

I call them, instead of large language models, large Earth models. Or, instead of AI, planetary intelligence, where you have planetary sensing systems in space and planetary compute systems in space. We can agree or disagree on the exact time frame, but I think it's going to happen. Then that's going to enable a huge economy. So, it's an exciting time in the next few years.

Will, Andrew, thank you guys very much. Well done.

Thanks, guys. Okay. Thanks, buddy. Appreciate you. Really appreciate it. Great seeing you, brother. Congrats.