|
Over 300 organizations rely on The Information for exclusive insights into public and private companies, including in-depth analysis of the ways in which tech moves markets. Click here to contact our corporate and enterprise team to learn more. Welcome back! Billions and even tens of billions don't generate much excitement at the annual Milken conference, taking place now in Los Angeles. But trillions, and especially $6.7 trillion, get some attention. That's the much-talked about number from a McKinsey report about the money needed to build out data centers to power artificial intelligence by 2030. Silicon Valley inventions are creating more places for pension funds to park capital. And more avenues for Wall Street to capture fees. The synergies manifested physically at some of the fancy private dinners around town. Hedge fund luminary Steve Cohen, I’m told, played dinner host to guests Sunday night including not only Michael Milken himself, but two high-profile guests whose worlds’ are coming together rapidly: OpenAI investor Peter Thiel and Blackstone’s real estate and data center honcho Jon Gray. The conference is, in some ways, humbling for venture capitalists because the event is much more about people like Gray than Thiel—more about the asset-heavy world of AI infrastructure than the applications that use it. Milken, after all, was the junk-bond king and the tech industry typically doesn’t have much use for debt. Blackstone, the largest owner and developer of data centers in the U.S., has more than six times as much capital to invest in data centers alone—$80 billion—as Founders Fund has in assets under management to invest in startups. The Milken conference has many lofty goals–it is subtitled, “Toward a Flourishing Future.” But it has always been about where the money is flowing. “So, AI. Is this the next industrial revolution?” Milken, the man, asked Nvidia CEO Jensen Huang on stage Tuesday afternoon. “Yes,” Huang responded, before comparing AI data centers to factories. Huang spoke the audience's language. “How large can these factories be? We’re building ones that can be a gigawatt, and each gigawatt is about $50-60 billion. Over the next 10 years, or so I wouldn’t be surprised to see tens of gigawatts of AI factories around the world.” Even attendees from outside the infrastructure investing world talked up the opportunity in data centers. A tech investment banker, downtrodden about the sorry state of IPOs this year, found a silver lining in the possibility that private capital giants like KKR and Blackstone would need to take their data center companies public in the coming years. A bank executive rattled off stats about how the number of construction loan deals they were doing the last few years had skyrocketed, thanks to data centers and AI. (I could still find some skeptics, of course. “When everyone’s building, won’t there be overcapacity?” one large fund manager asked.) Gray himself went on CNBC Monday at the conference to talk down the “negative” headlines on data centers—including lower-cost models like DeepSeek and export restrictions on high-end chips. “You’ll need a lot of GPUs, a lot of data centers,” he said, referring to Nvidia’s graphics processing units. “I feel we’re in the early days of the infrastructure buildout.” The Thiels and Grays are still trying to learn about each other as their fates are now deeply intertwined. One bridge between the two worlds I met at the conference was Simeon Bochev, a former senior leader in Apple’s applied AI group who now runs Compute Exchange, which is trying to make Nvidia chips more affordable through an open exchange. He sat in a closed-door conference session Monday that included Wall Street types like private credit investors, as well as data center operators. He said he told investors that Nvidia’s AI chips are advancing much more quickly than the processors that powered data centers in the last major infrastructure buildout. In other words, chips may fall in value more quickly than lenders realize, and that could make loans more risky, he said. “I do think Wall Street is getting more educated, but it is not educated enough,” he said. Now over to Natasha... We're hearing a lot of buzz about a new startup that wants to use AI to find new materials to improve chips and achieve other breakthroughs in physical science. Periodic Labs, a startup founded by Liam Fedus, a former vice president of post-training research at OpenAI, has told potential investors it wants to raise hundreds of millions of dollars at a valuation of at least $1 billion, according to two people who spoke to company leaders. That’s a steep valuation for a startup only founded two months ago. Likely emboldening the lofty fundraising target is that Fedus was one of the top contributors to ChatGPT. He started Periodic Labs with Ekin Dogus Cubuk, who previously headed the chemistry and material science team at Google’s Deepmind. Periodic Labs is starting by building a “ChatGPT for material science” that uses AI to discover and analyze novel compounds and manufacture them, according to materials reviewed by The Information. The startup is looking at magnetic and ceramic materials that could work for semiconductors or space technology, say by aiming to increase the temperature of materials to transmit energy more efficiently. The company plans to use the new capital on chips to train models for discovery, it told one of the people. It’s planning to start a San Francisco-area research lab, according to the memo. OpenAI has previously said that it was planning to invest in the startup and we’re hearing talk about potential other investors. But there’s a reason many VC funds have avoided physical sciences. Zachary Bogue, co-founder and managing partner of DCVC, said that startups focused on the broad discovery of new materials risk coming up with nothing. For that reason, founders have traditionally turned to research grants. Still, “with all these research grants going away right now, we need to fund basic research,” he said. —Natasha Mascarenhas
| 
