But first…

And now, here’s Ian…

In 1998, John Chambers, the former president and C.E.O. of Cisco, declared that “with internet leaders and government working hand in hand, America can look to a bright horizon filled with hope.” At the time, Cisco’s stock was trading at an all-time high; a few years later, it collapsed in the crash that vanquished the dot-com bubble. Chambers’s comments embodied the utopian optimism that defined the early days of the internet, and the same mood permeates these yawning days of artificial intelligence. In fact, the language used by industry leaders to describe the promise of A.I. might be even more extreme. If we’re to believe OpenAI C.E.O. Sam Altman, “access to A.I. will be a fundamental driver of the economy, and maybe eventually something we consider a fundamental human right.”

Since co-founding OpenAI in November 2015, Altman has never shied away from grandiose claims, and seems to genuinely believe the technology will unlock the kind of magic usually reserved for science fiction. On Tuesday, he wrote that the growth of the industry “has been astonishing,” and posited that A.I. could cure cancer or provide customized tutoring to every student on Earth if it were powered by another 10 gigawatts of compute. It was an essentially baseless assertion, and yet it’s Altman’s line amid OpenAI’s flurry of commitments to spend unfathomable amounts of other peoples’ money on more data centers to power its technology.

The latest head-spinning announcement came Monday, in the form of a letter of intent between Nvidia and OpenAI to develop a minimum of 10 gigawatts of data center capacity—a partnership that comes alongside the chipmaker’s $100 billion investment in OpenAI. (The money will be delivered on a progressive basis as each data center is erected.) A few weeks earlier, cloud giant Oracle entered into an agreement with OpenAI to develop about 4.5 gigawatts of data center capacity. And on Tuesday, OpenAI announced plans for five new U.S. data center sites, bringing its expected capacity for the Trump-approved “Stargate” project to 7 gigawatts—an investment of more than $400 billion over the next three years. It all rounds up to a minimum of 17 gigawatts worth of planned data centers—basically 17 nuclear power plants—representing close to a trillion dollars in spend, according to CNBC. (OpenAI didn’t return a request for comment about how it intends to power all these new data centers.) Finally, this morning, OpenAI expanded its agreement with CoreWeave by $6.5 billion, bringing the total size of CoreWeave’s OpenAI contract to $22 billion.

It’s unclear how OpenAI plans to finance its infrastructure play, but it’s likely that debt will play a role given the mind-boggling costs involved. (Of course, debt also largely financed the fiber build-out during the rise of the internet, and its increasingly prominent role in the A.I. build-out has some investors nervous.) Meanwhile, the company itself, which was last valued at $500 billion, is expected to make about $13 billion in revenue in 2025—and expects to lose well over $100 billion through the end of the decade. But Altman isn’t worried. As he said yesterday: “This is what it takes to deliver A.I.”

Obviously, I’m a skeptic. So I called up Tejas Dessai, the director of thematic research at Global X, whose largest A.I. exchange-traded fund has about $5 billion in A.U.M., so he could explain why this situation won’t end in disaster. In short, Dessai believes the companies’ astronomical valuations make sense, the cash burn is reasonable, and the costs associated with the infrastructure build-out—which OpenAI, as a first mover in the space, will need to frontload—are both necessary and possible to recoup. “We’ve seen unreal and unprecedented adoption momentum from ChatGPT, as well as other services from OpenAI,” he told me.

Dessai continued: “This type of growth—we didn’t see this from Uber, or WeWork, or any of the previous tech cycles in the past. That’s number one. Number two, it is monetizable growth.” It’s worth noting, though, that, as of June, OpenAI had about 500 million weekly active users, about 3 million of whom were paying. Now, the company claims to have around 700 million weekly active users, although how it measures “active users” isn’t clear.

Anyway, I wondered how Dessai viewed Nvidia’s investments, and whether OpenAI’s revenue should be able to cover its infrastructure build-out. “That’s one way to look at it,” he said. But then he echoed OpenAI C.F.O. Sarah Friar’s view that Nvidia is attempting to unlock the next level of innovation for the A.I. industry, and that OpenAI is the best vehicle for achieving that goal. He contended that this isn’t a historical anomaly, and that Nvidia was merely “trying to solidify their position in that vertical value chain.”

Dessai didn’t seem bothered by the fact that Nvidia’s chips seem to either die or depreciate at a somewhat abbreviated rate due to their heavy, 24/7 workloads, or Nvidia’s frenzied pace of next-generation releases. But he argued that even if the infrastructure isn’t permanent, the product is. Aging chips don’t become worthless, he said, adding that developers could use older chips to run models rather than train them.

I asked Dessai whether his expectation that things will work out for OpenAI—and the rest of the A.I. players—was at least partially based on an assumption that artificial general intelligence is achievable and imminent, and will somehow be priceless. “I think the company has a reasonable shot at a sustainable business model even if they continue to do what they’re doing here,” Dessai said. He argued that preexisting subscription-based companies aren’t a good comparison, given that Netflix’s addressable market is only touching entertainment spend.

OpenAI, he told me, touches a little bit of everything. “When you have a total addressable market that is expanding at an exponential rate, it’s very hard to say that the business models have matured or that they’re flat,” he said. “It’s unlike any previous tech cycle in the sense that we’re not only talking about exchanging information anymore—it’s about acting on that information, and those actions produce more information. That’s an exponential loop. It’s very hard to come up with a negative case that says, This company is going to miss a step and fall back.”

Needless to say, both Dessai’s point of view and mine are speculative. It’s possible that OpenAI’s soaring, mind-blowing revenues will never materialize. They also might come in even higher than expected. OpenAI is ostensibly attempting to build artificial general intelligence, which it defines as “A.I. systems that are generally smarter than humans,” and which many researchers have dismissed as more fiction than science. But the mythology around that effort has proven to be quite powerful. Indeed, SoftBank cited it as the literal “rationale” behind its decision to invest up to $40 billion in OpenAI—a number that looks downright meager when compared to their accruing infrastructure costs. Back in April, SoftBank declared its “mission to realize Artificial Super Intelligence for the advancement of humanity. Recognizing OpenAI as the partner closest to achieving A.G.I., a key milestone on the path to A.S.I., SBG has positioned OpenAI as its most important partner.”

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St, New York, NY 10006