The Compute Economy: Is Capital AI’s Next Bottleneck?

David Moberley (0:00)

That type of number would put AI Compute as an asset class, the second largest asset-backed debt market in the world behind the US mortgage market, which is roughly about 13 trillion.

SPEAKER_2 (0:12)

In this episode of What Matters, we are joined by David Moberley, a Managing Director at MA Financial, with deep experience investing in high-growth technology businesses. David argues that capital is emerging as the next major bottleneck in the AI build-up, and explores just how much financing will be needed to fund what is emerging as one of the biggest capital investment waves in history. He explains how the finance model is evolving, including the role Nvidia is proposing to play in supporting the rapidly growing NeoCloud sector.

David also tackles one of the key questions facing the AI investment boom. With extraordinary amounts of capital already being deployed, are we starting to see the returns?

Tom Brown (0:58)

Welcome to What Matters, our latest episode. Today we're joined by David Moberley, who recently joined MA Financial as a Managing Director working in Asset Management. He brings deep experience investing in and identifying high growth technology businesses. And as a pretty impressive track record in doing so, David, for the first time on What Matters, welcome.

David Moberley (1:19)

Thanks so much, Tom. Great to be here. Really looking forward to the chat.

Tom Brown (1:22)

Now, just for context, David recently wrote a investor paper, Compute Economy, a six-part series looking at issues like renting the future, how outsourced compute came full circle, the autonomy of an AI factory, it's touched on the video, the token economy, where tokens are going, and where AI gets a body, talking about things like robotics and the next compute frontier. And I read it and thought, David, it's absolutely essential that we get you on What Matters, because they're all my pet topics that I like to read on as well. So perhaps selfishly, I got you involved in What Matters. I guess the most salient point to start on is just take us through some of the capital flow we're seeing into AI and technology at the moment and sort of where that fits in history.

David Moberley (2:05)

Yeah, absolutely. It's a great point. So it's a great place to start. So I think firstly, it's really important to highlight that this is not new. So one of the biggest themes that we've seen over the last decade was this on-prem to cloud transition.

You might recall companies, they used to host a lot of their own systems and servers themselves. So you think of a company like MA, go back 10 years ago, instead of having the number of floors we do have in this building, there'd be a whole another floor full of servers and racks and all these types of equipment to host the systems that we needed. Now, roll forward to now, some of the big hyperscalers realized that they were investing in a lot of their own compute, and they could rent out some of the additional capacity they had, and because of their scale, they were able to make really strong returns on that, and it was also a great selling point for the customer. So the customer got to shift a whole bunch of their CapEx into OpEx, and the cost was lower for them. And if you think about it, that makes sense, right? Because they were having to scale their CapEx for the one day a year that was the highest need for their compute, whereas someone who's investing across a whole bunch of customers gets to split that across, right? So that's the first and most important point, that this trend was already happening. And then you think about the chat GBT moment, all of a sudden they added 100 million customers in a matter of months, and it kicked this whole trend into overdrive. So training these models and inference for these models are hugely compute intensive.

And that's how we got here.

Tom Brown (3:43)

And you've explained in the paper there's only been a few capital investment trends like this in history.

David Moberley (3:51)

Yeah. So this is one of the biggest. So across those hyperscalers this year, it sounds like we're going to be spending roughly a trillion dollars. And sometimes you can throw out numbers and that just gets silly, right? So what is a trillion dollars? So if you think about that, that's roughly about 2% of US GDP.

And there's only a few other capex cycles which really can compare. And the one a lot of people like to talk to is the railroad build out. And that got up to about 6% of GDP at its peak in the 1880s. And the other one was the Telco or the Fiber build out, which peaked at about 1% in the year 2000