At The Wall Street Journal’s Tech Live event on Wednesday, Sarah Friar, OpenAI’s C.F.O., said that the company was looking to develop an “ecosystem of banks, private equity, maybe even governmental—the ways governments can come to bear. … Meaning, first of all, the backstop, the guarantee that allows the financing to happen. That can really drop the cost of the financing but also increase the amount of debt that you can take.”

Since that comment slipped out, Friar has moved quickly to walk it back, clarifying on LinkedIn that “OpenAI is not seeking a government backstop for our infrastructure commitments.” Still, the idea—which comes as OpenAI is looking for a ton of debt financing to meet what Sam Altman recently described as $1.4 trillion in spending commitments—was more than a little attention-grabbing. David Sacks, the White House’s A.I. czar, eventually weighed in, saying that “there will be no federal bailout for A.I. The U.S. has at least 5 major frontier model companies. If one fails, others will take its place.”

In an uncharacteristically long post, Altman stepped in to clarify that OpenAI is not seeking loan guarantees from the federal government. He noted that the company is approaching an annualized revenue run rate of $20 billion, and indicated that OpenAI might continue to raise more equity or debt in the future.

And now for the main event…

Caught in the throes of a global race without a clear finish line, perhaps it was inevitable that the A.I. industry would look to outer space as the next frontier for expansion. This extraterrestrial push has coalesced around two somewhat distinct approaches. One involves moving semiconductor fabrication into orbit. The other involves building data centers in space—where, presumably, the issues of electricity consumption, water usage, and carbon emissions won’t be such a problem.

Besxar, which emerged from stealth mode last week, is among a small handful of companies that believe they can produce semiconductors in space. Founded in 2023 by Ashley Pilipiszyn, a former OpenAI employee, Besxar aims to address a decidedly terrestrial problem: The newest generation of ultrasmall, ultrapowerful A.I. chips are wildly expensive to manufacture, largely because they’re highly sensitive to even the slightest contamination during the fabrication process. Interestingly, much of the production expense comes from all the air filters, vacuums, and purifiers required. So Besxar is looking to circumvent these costs by moving manufacturing to the vacuum of space—the ultimate clean room. “We effectively have broken the capacity for Earth to be the ideal manufacturing place,” Pilipiszyn told me.

Besxar is venture-backed, with support from Bee Partners and Union Labs, along with what Pilipiszyn referred to as “strategic angels” from SpaceX, OpenAI, and Anduril. The company also belongs to the Nvidia Inception program. Pilipiszyn didn’t disclose the amount of money Besxar had raised, its valuation, or its runway. But she did note that moving the semiconductor manufacturing process off-planet might also clear the runway for other, vital advancements in chip technology.

To wit: Silicon, the primary material used in semiconductors, doesn’t have a very high thermal conductivity. Sam Altman has said that OpenAI’s G.P.U.s literally “melt” when experiencing high demand. Right now, there’s plenty of excitement around so-called compound semiconductors, which combine multiple materials in one chip, but they’re even trickier to fabricate than silicon chips and require an ultrahigh vacuum (U.H.V.) production environment—not unlike what you’d find in low Earth orbit.

Of course, the trick will be to bring the chips back to Earth in one piece—a task Pilipiszyn refers to as the “ultimate egg drop challenge.” Besxar plans to use autonomous, reusable manufacturing pods called “fabships” for this delicate task, and last week, the company signed a deal with SpaceX to launch an initial round of 12 missions beginning this year. Once they’ve established that the chips can return intact, the plan is for SpaceX rockets to enter orbit and the fabship to start production; afterward, it’ll splash down with completed semiconductor wafers. Besxar is already working on a next-generation version of the fabship, optimized for these longer production flights. “For the most part, it really is designed to be set and forget,” Pilipiszyn said. “Humans are in the loop for oversight monitoring if something goes wrong.”

Pilipiszyn told me that Besxar has already secured a contract with the U.S. Navy, and that there’s “been a lot of interest” from the Department of Defense. Everyone, she said, is on the hunt for more and better wafers—and they don’t really care where from. Meanwhile, as the number of rocket launches continues to increase, largely thanks to SpaceX, the price of sending stuff to space has dropped to $3,500 per kilogram or less. “For what we’re trying to achieve, we’re looking at less than a billion dollars versus tens of billions of dollars that are doubling every four years,” Pilipiszyn said.

Of course, the fabship plan is not a sure bet, especially given the number of upfront, recurring costs and engineering challenges. Still, Dr. Hugh Barnaby, a professor of electrical engineering at Arizona State University, told me that the idea is not “completely crazy” and could have major benefits, including a smoother chemical process (which is central to microchip production) and the possibility of fabricating compound semiconductors. “Ultimately, you’ve got to bring things up and down, and then you’ve got to fix things,” Barnaby said. “So I think the short answer is that it could work, but they’re probably many years out from really making this viable for the real production companies. I’d give it a soft maybe.”

Then there’s Starcloud, another company from the Nvidia Inception program, which aims to build fully functioning data centers in space. The idea is to leverage the space vacuum, rather than water, as a means of cooling these systems, while using the sun for “nearly infinite” solar power. Soon, the startup will send a G.P.U.-equipped satellite into orbit—the first step toward achieving its goal of orbital data centers.

Philip Johnston, the company’s co-founder and C.E.O., has said that within a decade, all new data centers will be put in space. (I’ll believe it when I see it.) And yet, Johnston isn’t alone in making this bet: Elon Musk recently indicated that SpaceX could achieve something similar by upgrading its Starlink satellite constellation.

Starcloud isn’t the only business hoping to take advantage of solar proximity: Google’s Project Suncatcher aims to get a bunch of solar-powered satellites into orbit, laden with A.I. chips, and use the sun’s power to scale compute in space. “I can see the attraction of solar power and probably some advantages for cooling, but the reality of sending the components into space and maintaining them is truly daunting,” Dr. Nathaniel Cady, a researcher at the University of Albany, told me. Not to mention “the communication bandwidth challenges for beaming data back and forth to Earth,” he added. “Another major challenge would be radiation hardness, or lack thereof, for the electronics themselves. That is already a challenge for long-term space missions.”

Google acknowledged these challenges—thermal management in space, communications from the satellites to the ground, and general system reliability—in its qualification of this plan as a moon shot. In any case, by early 2027 (assuming the A.I. investment cycle is still booming then), Google plans to launch a couple of prototype satellites to test the theory.

That’s all for today. I’ll see you next week.
Ian

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