We’ve been told it’s not a bubble. We’ve been told this is nothing like the Dot Com bust. To be fair, there are substantive differences. And yet there are certain indicators that there’s something off about this whole AI boom. Nvidia’s recent $500 billion AI funding ploy is one of those waving red flags.
Here’s what happened: Nvidia announced it was partnering with private equity firms Apollo, BlackRock, Blackston, Brookfield, Goldman Sachs and KKR to pool $500 billion in third-party capital for AI infrastructure buildouts.
There was initially some confusion about where the money would come from and who would be assuming the risk here (turns out it’s offering a 25% backstop – aka an emergency guarantee of funds). There was also a debate about whether this marked the creation of a new investment asset class or just vendor financing dressed up in a fancy suit.
Warning sign?
Eternal Wall Street optimists said it’s an emerging asset class akin to planes, trains and power. Others deemed it a “competitive moat” that Nvidia's rivals simply don’t have the balance sheet to touch.
The cynical view? It’s spiffed up vendor financing from a company that needs to keep a monster growth pace rolling or risk crashing an extremely fragile market.
In short: Nvidia is finding money for people to buy its chips as it becomes apparent that existing enterprise demand (or, at least, capex capacity) outside of hyperscale clients is not sufficient to sustain its growth trajectory.
And this isn’t Nvidia’s first foray into the financing world. As Futuriom’s Founder and Principal Analyst Scott Raynovich noted in a blog, Nvidia was also reportedly in talks to provide a $250 billion backstop for one of OpenAI’s data center projects and dreamt up a new financing model for its GPUs. Oh, and don’t forget it also has a multi-billion backstop deal with neocloud provider CoreWeave.
“There is a grim history when technology companies pivot their focus to finance. See: Lucent Technologies and Qwest Communications,” Raynovich wrote.
I’m not trying to just go hard at Nvidia here. Its financing moves are symptoms of a much broader problem in the AI market. And there’s also plenty of evidence to back up the conclusion that something about this whole situation isn’t sustainable. See also: all the explainers about the circular financing happening across the AI market. The whole thing, it seems, is propped up by a handful of companies buying from and funding each other.
Don’t take my word for it. Michael Burry, a hedge fund manager known for predicting the collapse of the housing market in the 2000s, is also getting the jitters about the AI market.
“I’m not sure why Nvidia has shifted the emphasis of their news announcements from technology to financing,” Raynovitch told Fierce. “This makes me nervous. If everything’s great, why are they searching for new sources of financing with elaborate financing schemes, SPVs and debt structures designed to outsource the risk from Nvidia?"
Middle ground
The conclusion you reach about Nvidia’s move depends heavily on whether you’re bullish or bearish about the AI market.
AvidThink Founder and Principal Analyst Roy Chua laid out two very distinct takes in an interview with Fierce. The bullish take is that Nvidia is creating a door for more parties to participate in the AI gold rush. As AI creates more value and wealth, investment vehicles like the one Nvidia is creating makes it possible for that wealth to be spread around to more people, he said. It is a smart way for Nvidia to spread risk in that it allows more parties to reap the rewards.
The flip side is a bit more disheartening. “It is a dangerous way of spreading risk potentially because now you can suck in pension funds and more sources, and you can increase the blast radius of potential failure,” he said, outlining the counterargument.
“It comes down to are you bullish or bearish,” Chua added. “The problem for me is I am both at the same time.” Chua said working in Silicon Valley, he has seen the benefits of AI and the potential it has to dramatically improve business operations. At the same time, he noted AI adoption still has a long, long way to go.
Where I’ve landed on all of this after talking to analysts and many players in the industry is that I’m bearish on the short-term market and bullish AI’s long-term transformative potential. Here’s why.
Bubble déjà vu
Bubble disbelievers make it feel like calling out a house of cards is akin to denying the game-changing potential of AI. But bubbles and revolutionary technologies aren’t mutually exclusive. Just look at the Dot Com Crash. Values were wildly inflated and after the market came back to reality (and mildly tanked the economy), the internet proved to be a long-term game changer.
The AI market feels a lot like it’s going to follow the same path – and not just because history has a tendency to repeat itself.
There are three things that bring me to this conclusion.
First, demand is real but perception of it is wildly inflated by deals being driven by a handful of companies. In fact, it’s mostly just leading AI companies buying chips and compute from one another.
For example, Microsoft said in January that nearly half (45%) of its order backlog was attributable to OpenAI. In its most recent quarter, its backlog grew 84% year over year in total but just 25% when excluding OpenAI. Amazon and Google Cloud didn’t specify their backlog mix, but both also have massive compute deals with OpenAI and Anthropic. Absent these deals, the pie would look significantly smaller.
And there’s evidence that the demand signals outside of these companies have been warped by recent trends. Tokenmaxxing over the past year drove a temporary spike in enterprise demand that is now being moderated by so-called “valuemaxxing,” as companies try to rein in token use and spend less on AI.
While some – like Together AI – are capitalizing on the pivot to lower cost AI, it may prove harder for proprietary model providers to do so. Or perhaps the trend toward open models may leave them with a smaller slice of the pie than they have been banking on as enterprises look to avoid vendor lock-in.
Second, leaving money and demand aside, there is another hard constraint on continued short-term growth: power.
Fierce has covered this issue extensively, but the long and the short of it is simple. The current pipeline of data center projects across the globe demands WAY more power supply than what’s available. Nvidia and others can throw all the money they want at AI, but projects can’t move faster than power.
It is not a trivial matter to spin up new power generation, even if you’re looking at shorter-term bridge solutions like natural gas. The lead times for natural gas turbines have hit seven years in some places. And on the nuclear front, though progress is being made to restart plants and deploy small modular reactors, the best-case scenarios have these coming online in 2030 and beyond. A market used to monstrous growth every quarter simply won’t tolerate such a delay.
It's a hard truth that a portion of data centers in the current pipeline simply won't be built, either due to power constraints or community backlash (that's a whole other can of worms).
Finally, the AI diffusion cycle is still in its infancy. While bleeding edge companies are using and finding value in AI, the enterprise masses simply haven’t caught up yet. The infusion of AI into Industry 4.0 efforts is similarly lagging. It will take time for them to catch up. And again, that doesn’t square with where growth expectations are today.
The takeaway
In many ways, you can’t blame Nvidia for its latest move. It is, as one person told me, doing exactly what it is expected to do as a smart, aggressive company leading the market. But the move is evidence that expectations around the current AI infrastructure boom have jumped the shark.
The AI market can be both real and overhyped: Nvidia’s financing move underscores how much of today’s boom depends on financial engineering, circular demand and growth expectations that may be impossible to sustain near term.
Expectations may be outrunning real enterprise demand, available power and sustainable growth — even if AI itself remains transformative long term.
“Across all of history, these things pop. It’s just how bad the pop is and who gets hurt,” Chua remarked. “But then after that, on the other side if you can hold on, it actually is great.”
Opinions from industry experts, analysts or our editorial staff do not represent the opinions of Fierce Network.