Toll Free:
877.456.EFSI (3374)
Everyone is talking about the blockbuster $500bn financing package put together by Apollo Global Management, Blackstone, BlackRock, Brookfield Asset Management, Goldman Sachs and KKR to finance compute for NVIDIA’s customers.
So, I’m going to, as well!
Six of the biggest names in alternatives have just decided that compute is a new asset class. Maybe it is. The GPUs, servers and associated infrastructure that generate AI processing capacity provide (billions of) people with everything from recipes for meals to entire software systems need to be paid for by somebody, and until now that somebody has mostly been the hyperscalers themselves, straight out of their own capex budgets.
But now, it’s being pushed out to private capital and the credit markets instead. Which, if you squint, makes compute a real asset. Kinda.
Historically, infrastructure investors have been comfortable owning assets such as airports, power stations, telecom towers and rail networks. These assets generate long-term revenue streams because people need to use them. The argument behind this deal is that AI compute is becoming just as essential. As AI adoption grows, companies will increasingly need access to large amounts of computing power in the same way they need electricity or internet connectivity.
The easiest analogy that I found was aircraft leasing.
Most airlines don’t buy every aircraft they operate, of course. Specialist investors and leasing companies often purchase the planes and lease them to airlines. The investors earn a return from the lease payments, while the airlines gain access to expensive equipment without having to fund the entire cost upfront (helpful stuff in a small margin business like airlines).
This deal appears to set up something similar for AI infrastructure. Instead of an AI company funding $billions of GPU capacity itself, an investor could finance the infrastructure and earn returns from the revenue generated when customers pay to use that compute.
The opportunities for investors are pretty obvious: Demand for AI infrastructure is growing rapidly (at the moment) so compute could become a genuinely important new asset class. If AI adoption continues to accelerate, investors could gain exposure to the growth of AI without needing to pick winners among individual software companies. Instead, they would be investing in the picks and shovels of the AI ‘gold rush’. Although you don’t escape concentration risk that way, as I’ll come back to.
And these assets could potentially generate long-term cash flows. If compute capacity remains scarce and valuable, the owners of that infrastructure could benefit from years of demand across many different industries and customers. That’s the pitch, anyway: compute as critical infrastructure, with the investment characteristics to match.
This wouldn’t be alternative finance without risks though, right?
The most obvious risk is future demand. Today’s investment assumptions rely heavily on the belief that AI adoption will continue to grow at the same rate (or more) than it has been doing in recent years. Michael Burry himself is a doubter, indeed.
So, if businesses struggle to generate meaningful returns from AI, demand for compute could disappoint. Investors could find themselves owning expensive infrastructure that is underutilised.
Another major risk is technological obsolescence. A school, a highway or an airport can sit there for forty years and still do the job it was built to do. AI hardware can’t. A cutting-edge GPU cluster today may look much less attractive in a few years. And this is where that concentration risk comes back. Investors are implicitly betting not only on AI demand, but on the ongoing value of NVIDIA’s technology ecosystem specifically. You haven’t avoided picking a winner. You’ve picked one chip designer instead of a software company, and you’re hoping it stays ahead.
There is also the risk of oversupply. If enough capital floods into the sector, compute may become abundant rather than scarce (yes, not everyone has $500bn to throw around, but some sovereign wealth funds do) so prices could fall, margins could shrink and returns might end up looking far less attractive than investors expect today.
Ultimately, the biggest question is whether compute really deserves to be viewed as infrastructure. Roads, airports and utilities have decades of performance data behind them. The same can’t be said for AI compute. Investors are being asked to believe that GPU-powered computing capacity will become just as fundamental to the economy as electricity or telecommunications.
It’s a fascinating story. The opportunity is enormous if AI continues to transform industries and generate ever-growing demand for compute. But it also feels like an experiment – we’re still in the early innings here, and the current players, quite clearly, are those that have the big check books.
This structure asks investors to place a long-term bet that AI compute will become a permanent, indispensable layer of the global economy. That may prove to be one of the smartest infrastructure investments of the next decade, or it may prove that not every fast-growing technology deserves to be treated like a toll road.
Exciting times!
**********
Gregory Poapst is a Managing Partner at Fundviews Capital. Connect with him on LinkedIn here.
Fundviews Capital is a full-service end-to-end Fund Management Platform. Our platform provides a complete end-to-end solution for asset managers or wealth managers to structure, launch, operate and grow their professional investment funds. You can launch a fund in a matter of weeks, not months, and with minimal capital outlay – not only reducing the risk of launching a fund but also maximizing your chance of success. Once launched, you will find that a dedicated team of professionals is just a phone call or email away at all times, handling all aspects of the back and middle office for your fund.