AI投资人圆桌：聪明钱在AI领域究竟流向何方｜EP 219

How will we fund the global AI revolution?

All the rules are being rewritten about how you fund growth because we just need all the capital we can get.

What is the main thing? It is AI. Where does the next NVIDIA-style growth come from?

The compute has gotten so expensive.

They are going to dedicate massive amounts of capital to this space. I'm the old-fashioned stock exchange. I think our common challenge will be to make sure that we find as many ways as possible to match the capital with the opportunities.

The amount of capital going into the sector way outstrips the venture funds. That trend is now drawing in a huge amount of money, which is why we're talking about it on this stage in Saudi Arabia. The untapped but mobile capital is here in this room, and if it jumps on the opportunity, it's an opportunity I've never seen before.

Welcome, everybody, to our AI mini summit brought to you by Link Exponential Ventures. It's a pleasure to have you. We're going to be having a series of 30-minute conversations that look at AI investing where the next trillion-dollar companies are coming from. We'll be having a session of our moon shot summit. And I'd like to open with our first session: How will we fund the global AI revolution?

To enable this conversation, it's a pleasure to bring on stage 3 leaders in this field. Dave Blundin is my business partner. He's a serial entrepreneur and the managing partner of Link Exponential Ventures, with 23 startups under his belt, a long track record of a 44% IRR, and a little over 1 billion dollars in AUM, based on the campus of MIT and Harvard.

Bonnie Chan is CEO of Hong Kong Exchanges and Clearing, or HKEX, since March 2024, bringing over 30 years of global capital markets, legal, and listing-transformation experience.

And finally on our panel this morning is Anjney Midha, a partner at Andreessen Horowitz, a16z, investing in frontier AI and open source—the man who's backed Anthropic and is on the board of Mistral.

Take a load off, Peter.

So, how will we fund the AI revolution, guys?

When I think about it, we're seeing today, at least in the United States, 1 billion dollars deployed per day into AI. The expectation is that we're going to see that grow to 3 billion dollars a day by 2030, and I expect it's going to blow through that. In fact, I'm seeing capital flowing to the exclusion of a lot of other things.

Let's open with some opening thoughts around that. Anjney, you're at one of the largest venture funds on the planet. What percentage of a16z is flowing toward AI? What are your thoughts about the capital availability to fund this infrastructure—what we call, on the Moonshots podcast, tiling the Earth in compute?

How much capital is flowing into AI? Basically, all of it, and it's still not enough.

The firm was founded to be a verticalized firm. We have an infrastructure fund, an applications fund, and a health care fund, and all of those are now AI funds, because AI is a cross-stack thing, whether you're working with teams that are training foundation models or building applications. I don't think anybody is not an AI investor anymore.

On the other hand, what's also insatiable is the need for these AI businesses, especially ones that generate tokens and leverage the latest generation of reasoning models, which generate 10 times more tokens than traditional generative AI models before reasoning.

We're living through Jevons' paradox every day. No matter how much infrastructure buildout we do, no matter how many algorithmic efficiencies there are, we somehow just need more compute and more infrastructure to serve the state-of-the-art demand in text, code, image, and video. It's just this insatiable explosion of use cases.

I don't think we've figured out how to change the traditional venture capital stack to fund all this growth. That's why you're seeing us try to fund entrepreneurs as much as we can, but then we've got to pull in all the friends we can, whether that's NVIDIA as a strategic investor, investing directly on the cap tables alongside us, or a data center provider.

The rules are being rewritten about how you fund growth because we just need all the capital we can get. Whether it's Satya doing a billion-dollar investment into OpenAI as a nonprofit 4 years ago, or Amazon and Google investing in Anthropic, all the rules are being rewritten.

Bonnie, when I see an offering being made by Elon for xAI, or by Anthropic or OpenAI, instantly it's filled. People are fighting to get into these deals, and no one's asking whether the valuation or the deal is going to make sense. They're just throwing capital at this. How are you seeing it from your perspective?

First of all, I do agree with the comment that Anjney made, which is the insatiable demand. Everyone wants to pour money into it.

But I must say, Peter, it's very interesting how you put together this panel, because, as I see it, I'm stuck in the middle of these 2 gentlemen. You represent the private side, shall we say, the VC/PE community. I'm the old-fashioned stock exchange: I do public offerings, and I do IPOs.

So, how are we going to fund it? I think there are many different ways, but suffice it to say that, given that the Hong Kong Stock Exchange is obviously in Asia, and given the demographics, there is an emergence of a big population of retail investors. We tend to now call them pro-retail investors.

With technology, everyone has their own trading theories and strategies, and they can execute in a rather sophisticated manner. From my vantage point, I still think that whatever ways are available to bring as many different pockets of demand from investors at all corners of the world will probably be a good way to support the development of AI on the one hand and really quench that insatiable demand on the other hand.

To put things in context, we've done quite well this year in the IPO space. In fact, Hong Kong is now number 1 on the global IPO league table this year. We have 300 deals in the pipeline waiting to get done. We've already done about 80 year-to-date, and I would say that of the 80 that have been completed and the 300 that are still waiting in line, about half of them have something to do with AI.

There are different manifestations, but I would say especially with companies in mainland China, these days, if you're not already doing something with AI or at the very center of AI development, you're probably quite unable to compete and be successful in your business.

That's my answer to your question. Given how much capital is needed to support the growth, whether it's private, public, credit, or equity does not matter. I think our common challenge will be to make sure that we find as many ways as possible to match the capital with the opportunities.

Dave, at Link, you're seeing and investing in companies as the first check—companies born out of MIT, out of CSAIL, and out of Harvard. What are you seeing as the growth of companies going into AI that's feeding the pipeline at the early stage?

I'll tell you, there's a reason Bonnie's on this panel, sandwiched between the startup guys, because the amount of capital required coming into these companies is enormous. Like you said, 3 billion dollars a day is coming in. U.S. venture is 200 billion dollars a year, so it's not even close. Five times more money needs to come from somewhere.

As Anjney said, some of it comes from NVIDIA, and some of it comes from corporate venture. But these companies, like Mercor, one of the ones in our portfolio, saw their valuation go from 30 million dollars at founding to 300 million dollars, 2 billion dollars, and 10 billion dollars.

And in what time?

2 years.

So, first of all, the 10-billion-dollar number is unprecedented in 8 or 10 years. What used to be incredibly rare is now incredibly abundant, but the amount of capital going into the sector way outstrips the venture funds.

What we generally see is that the corporate money—the NVIDIA money—comes in to fill the void. But the people working there, they say, "Well, this is really fun. I'm glad I made that Anthropic investment, but I'm going to go do my own fund."

So, the talent tends to eventually come out of the corporations and go into the 2-and-20 private sector to fill the space. I think that trend is now drawing in a huge amount of money, which is why we're talking about it on this stage in Saudi Arabia. The untapped but mobile capital is here in this room, and if it jumps on the opportunity, it's an opportunity I've never seen before.

Can we talk about the 2 sides of AI? One is the buildout of AI infrastructure, and the other is AI applications. With the buildout of those applications, where do you see the capital split between those 2 areas, and how attractive are they to venture funds or public markets? Anjney?

That's a really interesting question, because the last few years—basically 3 or 4 years—were dominated by the infrastructure buildout. Most of the capital that was going into startups was being converted directly to GPUs.

What’s interesting now is that you have a whole category of super exciting application businesses. I’m just right here building one in the coding space. To build application businesses like that, sometimes you need GPUs, but other times you need tokens from other foundation models. That’s now a raw ingredient as well.

So the capital stack was just raw cash. Then you’d convert raw cash to GPUs, and the foundation model teams converted the GPUs to tokens. That’s an input now into application developers, which, if you think about it, are more of a scarce resource. High-quality tokens from foundation models are a much more scarce resource than raw GPUs, and GPUs are a much more scarce commodity than raw cash. That’s the preference stack, I would say, of compute.

Do you see the demand for infrastructure buildout continuing and accelerating, or topping out?

Accelerating, and not being able to accelerate fast enough, because now the fundamental constraint is energy. We literally just don’t have enough power density in most of the legacy data centers in most regions of the world, and you’ve got to retool these data centers for GPUs.

If you look at the new Blackwell from NVIDIA, all the research scientists I talked to are really excited because it’s got the NVL72 networking stack, which means you can do a bunch of great, big, memory-intensive training runs, like video models. Then you get down to the brass tacks of when that data center can actually go live, when we can get it cabled, and when we can get the energy permits. That’s way after the chips can actually get there.

The infrastructure needs are largely driven by demand forecasting. As we discussed earlier, demand is completely uncapped. Meanwhile, the compute supply chain has caught up, but the energy constraint hasn’t. The energy supply hasn’t.

What we’re living through right now is this frenzy for energy contracts, where compute providers are trying to outbid each other to buy literally just energy supply. Depending on which part of the infrastructure stack you’re talking about, I don’t see things slowing down from a funding perspective. The CAPEX going into infrastructure is not slowing down, but what we may be faced with is a hard wall on energy scaling. We just don’t have enough electricity to power the chips.

Bonnie, what are you seeing in the public markets in terms of energy, data center buildout, chip buildout, and application companies?

Well, it is all of the above, right? But I do want to make a slightly more nuanced point. I think at the moment, the money that has been put into AI—$2 billion a day—a lot of it is probably put into these different opportunities on the premise that there is a promise that somehow it’s going to translate into things that are much easier to evaluate.

At the moment, people just want these use cases of AI. They don’t care whether it’s infrastructure, applications, or energy. It does not matter. But eventually, as the journey continues, I see a point where people will start to be a little more focused in terms of how we put a value on all these different opportunities.

From my vantage point, for example—and I think you raised a very interesting point—the energy bit is the million-dollar, billion-dollar, most a billion-dollar question, because without that, you really cannot go that far.

Right. Therefore, if I look at my pipeline, for example, I think China, as a lot of you know, has been quite advanced in terms of coming out with new energy solutions. It’s not only generating that new energy; it’s storing it. China is a massive country, right? So how do you make sure that you have all the grids talking to one another, and then you can generate energy in the western and northwestern parts of China, where there’s an abundance of sunshine, wind, and everything?

You have the geographic conditions to help generate that green energy. How do you make sure that you can disperse that to data centers at every corner of the country so that you can support all the data centers, the infrastructure, and all that? With that as the building block, you therefore can proceed to the next level and talk about compute, applications, and all that.

Again, I would say that China has an advantage because it is still a very big and dominant manufacturing hub. With that, it’s actually quite easy to think about possible applications and how you embed AI into production processes. I would also say that where I’m seeing a lot of activity is really the data-intensive sectors.

Just to cite an example, we are now beginning to see a lot of companies in the drug discovery business embedding AI. As you could imagine, the traditional way of drug discovery requires you to go through clinical trials, select samples, and do all that. It is data-intensive, but if you can speed it up with AI, you can imagine that you’re going to accelerate the pace of drug discovery so much.

You have a friend of mine going public on your exchange in Silicon Medicine in the next—

I’m not allowed to comment on any specific—

Yeah, well, anyway, I think you see my point there, right? Any data-intensive business will be a darling in this regard.

Dave, you’re seeing companies at inception. You’re seeing brilliant entrepreneurs. I think you’ve commented that the number of startups coming out of MIT and Harvard in the AI world has quadrupled in the last few years. What kind of distribution are you seeing? Where are they going into—application layers, compute? What are you seeing as the categories?

The companies coming out of MIT and Harvard are overwhelmingly going into vertical use cases, and then also some foundation model companies. Liquid AI will be on stage right after this. So there are a few of those, but many, many more vertical use-case companies.

The success rate of those is near 100%, and they’re attracted to them because, first, they’re not super capital-intensive.

100%.

Well, so far, for us, MIT and Harvard teams that fit a profile are 100%. I’ve never seen anything like it before, and it’s because the use cases are so abundant relative to the talent pool. If you have the talent, you’d have to be crazy to go after a bad use case right now. You can use AI for so many things.

It’s very, very different from crypto, which was the last wave, and more similar to the internet. The internet is incredibly flexible. You can use it for many, many things, and you saw, when I started investing in the late 1990s, everything you invested in succeeded. Why? Because the internet can do almost anything.

Unless you’re insane and going after something really dumb, you’re going to succeed. I haven’t seen that again in my lifetime until now. Now it’s the same thing, and the value is enormous. The teams are thriving every single time, but they’re really attracted to the vertical use cases because they’re not as capital-intensive as building out an entire data center.

There, Chase Lochmiller is doing Stargate, so there’s one guy who’s an exception to that. There’s a $500 billion buildout, but that’s relatively rare. Most people go after the use case.

How quickly are you seeing the valuations in those kinds of companies scale?

In the companies that are doing the vertical use cases, typical entry valuations are what they’ve always been—maybe $20 million to $30 million. The first funding will be $100 million to $300 million, and then within 2 years, if you’re going to be a unicorn, you’re going to get there in 2 years now.

That means the founders now are still 23 or 24 years old. That’s a new thing in the world, too. I think about my entire lifetime of investing. I can name 3 or 4 people that I knew or invested in who hit billionaire status under the age of 30. Now I can name 8 that we’ve invested in just in the last few years.

So there’s this new class of person roaming around who barely has a driver’s license but has $1 billion in liquidity. We have to kind of adapt to that.

Being a billionaire was a big deal. Now we’re just going to wait for the trillionaires to start.

Well, we’re all born in the wrong age.

Yeah, yeah, yeah. I want to understand what you guys consider the biggest risks over the next year. Is it compute-cost inflation, talent scarcity, or regulatory intervention? We’ve been on this incredible inflationary and exponentially growing curve on all things AI. Just like you used to add “.com” to the end of your company, now it’s like, “Oh, we use AI.”

Anjney, what are you seeing as the risks?

On fundamental progress of capabilities, we already talked about energy, which I’m concerned about.

Double-click on that. Will these companies have access to sufficient electrons to run the data centers? What is the scarce resource in the chain?

In the United States, I think that’s a direct function of whether the permitting regulations that the current administration is working on end up getting executed. There was a big plan that was introduced—the AI Action Plan—about 2 months ago, which I think was a fantastic start. If you go line by line through that, it really is a very precise, methodically laid-out document that says, “Here’s what we need to do to unblock progress.”

And I think if we can operationalize it and execute it, then we should be good. But rarely has that ever happened at scale without a ton of bureaucracy. A ton of bureaucracy. And this is my second concern, which is whether it can happen without a ton of civil blowback.

Because the reality is that putting these massive data centers down, cabling, and reallocating parts of our power grid from other things results in tough trade-offs we've got to make as a society. I want to respond to the previous point a little bit, where it is true that we are seeing enormous wealth creation amongst this generation.

Anthropic has gone from a company that was a couple hundred million in valuation just 4 years ago to $183 billion in 48 months. But I don't think we should be celebrating that as much as we kind of are right now, because at the end of the day, the public is not participating in that wealth creation.

The vast majority of wealth being created by frontier AI is locked up inside private capital, like our funds. It's locked up inside a small group of talent that is super mission-oriented, but I don't think we've really figured out what happens when the rest of the public goes, "Where's my piece of the future?"

And I don't think we're ready. I don't think we're talking about it enough, and I don't think governments are doing enough to realize how dire it's about to get when 30% of your IT services GDP sector gets vaporized by tokens.

If you're in India, for example, where double-digit percentages of your GDP are literally IT services, what do you do when Claude and GPT-5 tokenize vast portions of that flow? We love to talk about productivity growth, and we don't talk about how to manage the short-term transition pains. That's going to be ugly.

So you're adding that to our risk profile: civil unrest.

Absolutely. A good example of that, too, is that just a few months ago, when Sam Altman said, "Hey, I'm going to give everybody in the company a $1 million retention bonus—everybody," the intention was for that to be cool. The reaction worldwide was, "That's not cool."

And so now you're seeing the AI leaders—the AI leaders, it comes up on the Moonshots podcast a lot—really downplaying the rate of progress, because the people who are picketing outside the door at OpenAI headquarters are lined up deep now. They're like, "Look, all this wealth—you guys are all billionaires—what about everybody else out here on the street?" They don't need that.

Just to put a finer point on that, I know a number of the technology leaders and investors in Silicon Valley who have been getting death threats. Then they lock down their companies and lock down their homes. This is before we're seeing CPI of electricity going up, and before we're seeing the real layoffs that will occur.

I think this is important. I think AI is going to get blamed for a lot of layoffs that have nothing to do with AI. A lot of the layoffs we're seeing today from big tech companies are really just people correcting for over-hiring during the serf era of 2010 to 2020.

Well, also the print-money era of COVID.

So the easy money's gone, and a number of big tech companies that just thought they could keep chasing returns by over-hiring—which was a fairly rational thing to do then—are now facing a different reality.

In fact, the government was paying you to go higher.

Exactly. But the incentives have changed. There will be a lot of boogeymaning around AI that has nothing to do with AI.

Agreed. But once we're through that era, what happens is people are going to start asking, "Why isn't my pension fund, my sovereign fund, or my retirement plan participating in the AI wealth-creation opportunity?"

That's why, to the point of this panel—how do we fund the future of AI?—we should be asking, "How do you connect frontier AI growth to public wealth creation?"

There are a bunch of institutions whose job it is to steward our wealth: sovereign funds, pension funds, and state funds. Why aren't they investing on the cap tables? Why is it family offices and high-net-worth individuals?

When we went out to raise the seed round for Anthropic, I made 22 introductions to them up and down Sand Hill Road. They got 21 no's. So we had to scrape together $100 million, which sounds like a lot of money, and which was a lot more money back then. Now, actually, to Dave's point, it may not be that much.

But really, that funding round had to be pieced together from angels and high-net-worth individuals. I'm still shocked at how often today traditional venture funds, sovereign funds, and traditional pension funds are not being aggressive enough in taking their job as stewards of public capital seriously and exposing it to frontier AI wealth creation. It's just not happening fast enough.

Dave, do you want to add on the risk side?

Yeah. I completely agree with what Anjney said, and I'll give you another parallel risk. The core AI companies that do things like customer support and white-collar automation are just killing it—adding immense amounts of value.

The investment community coming in has started to extend that to, "Tech is a good place. Let me put $200 million into fusion energy." They're like, "Well, that's not AI." "But it's going to create the electricity 4, 5, or 6 years from now to fund it, so it's related to AI." I'm like, "Well, okay, but that's very capital-intensive, and you're not sure it's going to work."

I think it will work, and I think it's a good area to invest in, but if it doesn't work, that's where you're going to have what happened to the internet in 2000. The internet was very real, and if you waited long enough, it came roaring back, but everybody lost confidence in 2001. Why? Because of some really bad peripheral investments.

We're seeing that now, and I don't want to throw too many things under the bus, but things like robotics and fusion energy are very capital-intensive. They're not the obvious win of AI; they're peripheral investments. Some of those will be good. Some of them are going to consume a ton of money and turn into losses, and that may scare off the entire investment community.

That would be tragic, because if you look at AI voices doing sales and customer support, that's $500 billion of payroll worldwide today. AI does it better than anyone on the phone already. It exists. We just need to deploy that $500 billion. If you invest in that, you cannot go wrong.

But if you get sold an investment in something that's kind of like, "Well, quantum computing also might work," maybe. Maybe it will, maybe it won't. It's much more speculative and very capital-intensive.

I do want to chime in there. I'm listening to all this, and one part of me is saying, "I want to democratize these investment opportunities. Let more people partake in the party." But on the other hand, given how the current ecosystem has built out, the valuation has already been hovering somewhere up here.

Opening the door for investors, especially retail investors, to partake in the public markets also causes me concern. For all I know, they could be the last ones in at the party before the whole thing collapses.

I would call that a risk. How do we actually find that new equilibrium where these opportunities are not just monopolized by a very small group? How do we make more sense out of the valuations we're seeing, which are being established by a very small and rather opaque price-discovery mechanism?

To your original question about risk, I do see the energy piece as one that is very difficult to solve. Even at my company, we're exploring what we can do with AI. We've come up with a few use cases that we're experimenting with, and the next thing you know, the electricity bill arrives and you start scratching your head.

I thought AI was going to help me with productivity and make things faster, easier, and more accessible. Yes, but there's always a cost there.

We have a minute left for closing thoughts from each of you. Anjney?

I think the answer lies in institutions that represent the public: sovereign funds and wealth funds. You're right, opening up the markets to retail investors who may not understand what's going on may not be the answer. But institutions that represent the public are the answer.

It's our job to educate them and make them more aggressively take a position in the wealth-creation opportunity that's happening. Otherwise, the public will get left behind.

Bonnie, closing thoughts on who's going to fund this?

I agree with that. I really think that everyone in this ecosystem needs to work together to find that new equilibrium. It shouldn't be wealth creation for a tiny fraction of the world's population, and we need to find the right way to get it done.

All right, Dave.

I think the most important thing that I heard on this stage today was what Anjney was saying: the story of how Anthropic got funded. So many people are not getting in the game, and Silicon Valley investors who are just walking down the street and investing in each other are killing it and running away with all the gains, because it's just not that hard.

You just need to get into the loops, get into the places that are making these investments, and get in the game. The pro rata rights on that deal alone would have allowed you to invest a follow-on of probably, what, $4 billion, $5 billion, $10 billion. But you just had to be there in the game at the outset. Every week my team and I study the top 10 technology metatrends that will transform industries over the decade ahead. I cover trends ranging from humanoid robotics, AGI and quantum computing to transport, energy, longevity and more. There's no fluff. Only the most important stuff that matters that impacts our lives, our companies and our careers. If you want me to share these metatrends with you, I write a newsletter twice a week sending it out as a short two-minute read via email. And if you want to discover the most important metatrends 10 years before anyone else, this report's for you. Readers include founders and CEOs from the world's most disruptive companies and entrepreneurs building the world's most disruptive tech. It's not for you if you don't want to be informed about what's coming, why it matters and how you can benefit from it. To subscribe for free, go to demandis.com/metatrends to gain access to the trends 10 years before anyone else. All right, now back to this episode.