Artificial Investment?

It’s no secret: Investors and companies are pouring billions into AI. 

 

Venture capitalists have plowed $12 billion into generative AI startups so far this year. Elon Musk claims his company xAI will train new models on a staggering 100,000 of Nvidia’s latest H100 GPUs; Meta is reportedly buying some 350,000 more. (In perspective, the world’s fastest supercomputer, Frontier, boasts a mere 38,000 GPUs.) Alphabet, Microsoft, and Amazon are also building out enormous AI server farms. And this spending shows no signs of slowing.

 

Goldman Sachs says outlays on AI infrastructure could total $1 trillion over the next few years—a number on par with annual US military spending or the GDP of the Netherlands. But there are strings attached: Big investment is tied to expectations of big revenue and profits. Eighteen months into the boom, investors are beginning to review the latter.

 

Some companies are clearly knocking it out of the park. Nvidia, which commands 80 percent of the AI chip market, is enjoying a historic run. The chipmaker has consistently posted revenue and profits beyond the market’s expectations, and in late June, it was (briefly) the most valuable company in the world, beating out Microsoft and Apple. (The stock has fallen off since then, but it’s still up some 140 percent year-to-date.)

 

But not everyone is raking it in like Nvidia. If Nvidia is selling “picks and shovels” for an AI gold rush, those using Nvidia’s equipment are still digging for gold. Exactly when they’ll strike it rich and how much they’ll unearth is a matter of speculation.

 

Though OpenAI is reportedly on pace to make $3.4 billion annually, an impressive number compared to last year, its costs are even greater. The Information recently reported the company is set to lose $5 billion this year and may need to seek new investment within the next 12 months. The publication also wrote that Anthropic will burn $2.7 billion in 2024.

 

The biggest players, like Alphabet and Microsoft, have deeper pockets. But driven by AI, their capital expenditures are growing fast. Alphabet’s capex spending is on pace to hit $49 billion this year, an increase of 84 percent compared to its average over the last half decade. Investors are increasingly scrutinizing the impact of these investments too. The companies’ cloud offerings—the huge server farms they rent to those building and running AI models and products—are the most obvious revenue-makers. But the sustainability of those revenues is dependent on the success of AI products.

 

And revenue from those products is lagging investment. This gap, which is widening, is what Sequoia Capital’s David Cahn calls the “$600 billion question.” That’s the difference between infrastructure investment and revenue in the industry. Barclays analysts put it like this: The industry is building enough infrastructure to support 12,000 ChatGPTs. While the technology is impressive, customer demand is unlikely to support that kind of supply anytime soon.

 

But the spending is also a bet on future performance improvements. This line of thinking suggests that if today’s capabilities aren’t enough to hook the world on AI, tomorrow’s will be. And to get to the next level—which will only be realized by bigger models trained on more (and better) data—companies need more computing power.

 

Anthropic CEO and cofounder, Dario Amodei, has said models being trained today already cost upwards of $1 billion. By next year, that number could be $10 billion, and in the years following, it could reach $100 billion—in line with the amount Microsoft is said to be spending on its 2028 AI supercomputer, Stargate. It's an eye-watering price tag. But Amodei thinks that if algorithms and chips continue to improve as they have in recent years, then there’s a chance those future AI models could be “better than most humans at most things.”

 

FOMO is another big driver. The CEOs of Alphabet and Microsoft have made that much clear. In a recent earnings call, Sundar Pichai said, “the risk of underinvesting is dramatically greater than the risk of overinvesting for us here.” And in a New York Times profile, Satya Nadella is described as being motivated by the desire to avoid missing the boat, as the company did with the internet in the early 2000s and smartphones several years later.

 

It seems clear the technology will have an impact—whether it’s through productivity boosting AI assistants or scientific discovery—and the impact’s size will scale with the capabilities of the models driving it. But exactly how big a splash it will make, how soon, and how this will compare to current investment isn’t clear, even to those at the helm.

 

“Of course, it’s possible…that all of this will be a bust. The models don’t turn out to be that powerful,” Amodei said recently. “That’s not my bet. That’s not what I think is gonna happen. What I think is gonna happen is that these models will produce a great deal of revenue. …That’s the bullish scenario that I’m betting on by leading this company.”

 

“But I’m not sure. It could go the other way. I don’t think anyone knows.”