The Hidden Risks in AI Funding
Bankers and investors, especially when they are sitting in front of an audience, are not shy about offering their opinions. Yet when I asked a panel of money types at our Financing the AI Revolution conference on Monday about risks in the current market, I got near silence.
We had been discussing the billions of investment dollars pouring into AI. The question was: “Anytime there’s a boom, there are risks built up in the system and it may not be a systemic risk, but it may be specific risks. And so, what should we all be watching? What are things that you look at and you have doubts?”
After an uncomfortable pause, Temasek’s Martin Fichtner raised the question of whether booming demand would continue to accelerate. Fichtner, who invests in the technology and consumer sectors from San Francisco for the Singaporean investment fund, said he is watching the “second derivative” of growth in demand. In other words, he’s watching whether the growth in demand will continue to accelerate or whether it will slow.
“Second derivative” is a term out of calculus best described as the rate of change of the rate of change. It is a favorite metric for growth investors, especially when expectations are high. When the second derivative declines, the growth can still be accelerating, but the rate of that acceleration is slowing. That’s enough to get people nervous. “[Acceleration] doesn’t need to decrease—just if it flattens out a little bit,” he said.
Fichtner, who is broadly positive on the AI boom, wasn’t the only speaker at our conference who was willing to at least address the risks. Fellow panelist Jim Prusko, a senior portfolio manager at investment firm Magnetar, cited the risk of regulations on AI and political pressure against data centers, seeing both as a threat to the U.S. AI build-out. “I think there’s a risk that there’s some sort of backlash against data centers and we lose the ability to deploy the compute we need in America,” Prusko said. Magnetar is a big backer of data center developer CoreWeave.
There have long been doubts about the spending on AI. But surging demand for models such as Anthropic’s Claude, especially from businesses, has wiped most of those doubts away. Lenders have lined up to fund data center developers, while investors have poured tens of billions of dollars into AI model makers such as Anthropic and OpenAI. The market is buzzing about likely initial public offerings for these companies and for SpaceX, which is almost on the launchpad.
This all makes sense: The models have improved dramatically, and the potential scale of AI is hard to grasp. Still, investors have a way of getting ahead of reality. That doesn’t mean AI is a bubble, just that there will be ups and downs as cash pours in and the technology rolls out.
CoreWeave, for example, can’t raise cash fast enough to keep up with demand, said Nick Robbins, the company’s vice president for corporate development. “So the way we think about it is there’s always gonna be a timing mismatch,” he said.
Right now the mismatch is that demand exceeds supply. At some point for some company, supply will exceed demand, and that company will scramble to find customers. If it is borrowing as CoreWeave and its competitors have been, then the mismatch could be dangerous for the company and its lenders.
How can that happen? We’re seeing one scenario play out right now. Facing high demand and soaring costs for computing power, Anthropic raised prices to such an extent that customers’ costs could double or triple, according to one estimate. Higher costs are hitting many companies before they see measurable gains from AI, potentially reducing their appetite for big spending plans.
Here’s another current scenario—this one at OpenAI. The Information has been reporting about the company’s missed targets, C-suite dysfunction and lagging growth for some time. A story this week that added even more issues knocked down shares of Oracle and CoreWeave, which have placed big bets on OpenAI’s success. Sam Altman, in a race for computing power, has committed the company to paying hundreds of billions of dollars for computing power in the coming years.
OpenAI is not the only cash-burning company to commit to paying billions of dollars for computing power. True, it just raised $122 billion in its last funding round, its models are getting better and demand is still strong. But all of that is baked into expectations already.
What else can go wrong? Panelists at our event raised two other risk factors. The coming SpaceX IPO will be the first to provide detailed financials for a major private AI developer. XAI’s numbers won’t be pretty, and its deal for Cursor can be read both as a smart catch-up play and as an admission that things are going badly at xAI.
The second is more nebulous. Katherine Kaminsky, U.S. chief commercial officer for PwC, advises companies on adopting AI. She said compared to other recent strategy shifts, such as moving computing power to the cloud or outsourcing back-office functions, AI is far more complex and requires a rethinking of nearly everything a company does. That won’t happen overnight: After an early surge in AI adoption, companies may move at a more measured pace.
Investing is all about expectations, and they are staggeringly high right now.
New From Our Reporters
