{{ 'now' | timezone: 'America/New_York' | date: '%b %d, %Y' }}
|
|
|
|
Welcome to The Hidden Layer. I’m Ian Krietzberg. Thanks to Mayer Brown and
everyone who came out for my fascinating conversation with journalist Joanna Stern to discuss her new book, I Am Not a Robot: My Year Using A.I. to Do (Almost) Everything, and what it was like attempting to “date” ChatGPT.
In today’s issue, an investigation into the mysterious partnership between xAI and Anthropic. Has Elon given up on the frontier model race? I’ve got thoughts. Plus, news and notes on Sam Altman’s congressional
summons, Waymo’s flood problems, and an exclusive bit of polling that interrogates the public mindset on A.I.
And in case you don’t regularly imbibe CBS Mornings, here’s a clip of my Wednesday appearance breaking down the nuances of the Musk v. Altman trial, closing arguments of which occurred today.
Also mentioned in this issue:
Dario Amodei, Gil Luria, Tejas Dessai, Greg Brockman, Solly Malatsi, Cathie Wood, David Bader, Jennifer Pruitt, and more…
|
Three Things You Should
Know…
|
- Anthropic’s
mini P.R. win: Fresh polling, conducted by Slingshot Strategies and shared exclusively with me, confirms that Americans remain thoroughly confused about A.I. Only 24 percent of those polled categorized themselves as strong proponents of the tech; the rest ranged from being concerned about misinformation (6 percent) to worrying about job loss (9 percent), being strongly opposed (15 percent),
and being concerned but having “mixed feelings” (33 percent). As one Republican man from Ohio put it: “A.I. will help the business industry world. However, I think it will hurt humankind overall.”
Asked about the role of A.I. in warfare, 27 percent replied that it was dangerous or wrong, 17 percent weren’t sure but had mixed feelings, and 9 percent said it needs human oversight. Only 15 percent said its use on the battlefield was both inevitable and necessary, and a further 8 percent
praised Anthropic for its well-publicized decision to uphold limits on how the military can use its models. This impression of Anthropic spanned parties, demographics, and geographies. “I was very impressed by their ethics and morality,” one Democratic woman said. “I think there may be a time when A.I. tools will benefit the military, but I’m very concerned. It’s like 1984 in reality. I’d certainly be more interested in using Claude because of their ethics.”
|
|
|
|
A MESSAGE FROM OUR SPONSOR
|
Retailers can now process millions of transactions in minutes—not hours. Toshiba Tec partnered with McKinsey using
Nvidia accelerated computing to enable real-time recommendations, faster promotion testing, and measurable lifts in sales, profit and long-term customer value. Read the case study.
|
|
|
|
- Summoning
Sam: The House Oversight Committee wants information from Sam Altman. Specifically, lawmakers have asked for a briefing from OpenAI’s chief legal officer about the company’s internal auditing for conflicts of interest, plus any and all relevant internal documents from 2015 until now. The request refers to the ongoing
litigation between Altman and Elon Musk, in which OpenAI co-founder Greg Brockman revealed that he holds stakes in two of Altman’s startups in addition to a percentage of Altman’s family fund. Apparently, the committee is worried about OpenAI potentially being used to “bolster the value of other companies and startups” in which Altman has a stake.
- Waymo recall: Waymo issued a voluntary recall of roughly 3,800 robotaxis
this week in response to a software bug causing some cars to “slow and then drive into standing water on higher speed roadways,” according to a letter posted on NHTSA’s website. The number, a Waymo spokesperson told me, represents the company’s entire fleet, inclusive of both testing and commercial vehicles; their commercial fleet numbers 3,000 robotaxis. On April 20, a Waymo
in San Antonio—thankfully not carrying passengers—drove into a flooded road and was carried away, precipitating the latest NHTSA probe. “We are working to implement additional software safeguards and have put mitigations in place, including refining our extreme weather operations during periods of intense rain [and] limiting access to areas where flash flooding might occur,” the spokesperson said. Waymo
told CNBC that its San Antonio service is still temporarily suspended following the incident, but will reopen soon.
|
Hallucination(s) of the
Week
|
South Africa was recently forced to withdraw its draft National Artificial Intelligence Policy because it
contained “various fictitious sources”—the result of A.I.-generated citations. So yeah, that’s still happening. It’s proof, Minister Solly Malatsi said in a statement, of why A.I. governance is important in the first place. Well
played, sir.
Runner-up: Amazon employees have been using an internal A.I. tool for pointless tasks to artificially juice the company’s A.I. use metrics. That’s one way to prove you’re actually using A.I., although perhaps not what Andy Jassy had in mind.
And now for the main event…
|
|
|
|
A surprising deal with Anthropic is raising questions about whether Musk’s xAI is abandoning
the frontier model arms race to focus on neocloud services instead—including launching G.P.U.s into space.
|
|
|
|
Last week, Elon Musk threw the tech industry for a loop when he revealed that SpaceX, which
recently acquired xAI, had cut a deal worth as much as $4 million in annual revenue to lease Anthropic the entire capacity of its Colossus 1 data center. Not three months earlier, Musk had called Anthropic, an A.I. archrival, “evil.” Now it appears he’s becoming their data landlord.
The deal makes plenty of sense for Anthropic: Access to Colossus will, in their
words, “substantially increase our compute capacity.” Recently, Anthropic had been struggling to keep pace with massive, unprecedented demand for Claude Code and its Opus models—as has become evident in the form of ever-tightening usage restrictions and higher prices. As soon as it announced the partnership, Anthropic immediately loosened the rate limits on its products.
What the deal means for SpaceX, however, is less clear. More revenue, certainly, as the company prepares for an
I.P.O. But it also suggests that xAI—which, until recently, was attempting to compete with Anthropic and OpenAI—may be waving the white flag in the extremely costly frontier model arms race. Musk’s chatbot, Grok, has lagged far behind its rivals in usage and popularity. But xAI does have a ton of compute, having built two Colossus data centers. So why not lease what he doesn’t need, essentially transforming SpaceX into a partial neocloud?
|
|
|
|
A MESSAGE FROM OUR SPONSOR
|
Retailers can now process millions of transactions in minutes—not hours. Toshiba Tec partnered with McKinsey using
Nvidia accelerated computing to enable real-time recommendations, faster promotion testing, and measurable lifts in sales, profit and long-term customer value. Read the case study.
|
|
|
|
In truth, it seems to be more of an A.I. pivot than a surrender. “I wouldn’t call it a stepping back,” said
Tejas Dessai, the director of thematic research at Global X. “I would call it doubling down on what is working for them.” In other words, the model race will be hard to win, he said, and “infrastructure is a good business for SpaceX.” Anthropic also expressed interest in collaborating with Musk to develop data centers in space as part of the partnership—a notion that, despite formidable cooling challenges, began to grow in popularity
last year. This last deal point, in particular, would have been music to Elon’s ears. When he folded xAI into SpaceX earlier this year, he specifically referenced the economic opportunity surrounding orbital data centers.
The financial upside could be sizable. Gil Luria, an analyst at D.A. Davidson, told me the deal would help SpaceX
materially “shore up” its financials ahead of its I.P.O. journey. Ark Invest founder Cathie Wood, the controversial investment manager and longtime Musk booster, predicted that the agreement would allow SpaceX to pivot “from massive losses at Colossus to significant profitability as a neocloud on an estimated $5-6+ billion in annual revenues.” They’re not alone in
that thesis: The Wall Street Journal reported on Tuesday that Google is in talks with SpaceX to secure a rocket-launch deal to send its own early-stage orbital data centers into space. And in March, Blue Origin
filed for F.C.C. permission to launch Project Sunrise, a satellite constellation designed to do advanced computation.
Of course, Elon’s pivot also raises questions about the future of Grok, and whether xAI is meaningfully diverting from its foundational mission. As the Journal
reported this week, downloads of Grok have fallen by more than 50 percent from their peak of 20 million in January—the month Grok gave users the ability to digitally undress people. Since then, the app has tumbled down various app store rankings, far behind ChatGPT and Claude. Indeed, a recent survey of 260,000 Americans found that only 0.174 percent
pay for Grok while 6 percent pay for ChatGPT—suggesting that all those “hey @grok” prompts on X are more of a gimmick than a real business.
|
When Musk launched xAI in 2023, his stated reasoning was fairly straightforward: The notion of “pressing
pause” on the A.I. race—a trendy conversation among safety advocates at the time—didn’t “seem realistic” to him, so he might as well be the one to win it. He stressed the need to “create a third option” to rival OpenAI and Google—a “maximally truth-seeking” alternative to the products coming out of the other frontier labs, which he deemed too “woke.” A few months later, Musk would sue OpenAI, accusing co-founders Sam Altman and Greg Brockman of
conspiring to steal the nonprofit entity that Musk had initially funded to the tune of millions. (Musk, as I wrote last week, is fighting for $134 billion in damages, the removal of Altman from the OpenAI board of directors, and the return of the company to its original nonprofit structure… none of which is likely to happen.)
While Musk was “late” to the chatbot
game, as he acknowledged, he quickly closed the delta between xAI and OpenAI, primarily by raising the kind of funds accessible to the world’s richest man: two $6 billion rounds, a Series B and C, in 2024, plus a $20 billion Series E in January, followed by a merger with SpaceX, another of his companies. Meanwhile, he set about constructing the world’s largest supercomputer in Memphis—a data center loaded with hundreds of thousands of G.P.U.s that he christened Colossus. Incredibly, Musk brought
the facility online in just 122 days (thanks in part to dozens of gas turbines, which haven’t been popular with the local community). More recently, xAI stood up a second data center, Colossus 2, bringing its combined G.P.U. count to well above a million.
Now, having fallen behind other frontier models, all that extra compute is
presumably looking too good to waste on Grok. “The race to model development might already be over,” Dessai said. “The real business that still has plenty of opportunity is the infrastructure layer of that story.” For Musk, he said, this might well be a step away from the kind of model business currently powering OpenAI and Anthropic, and a step toward other kinds of non-chatbot applications, such as geospatial imagery, data analytics, self-driving, and robotics. There’s “a broad range of
applications they could go into,” he said. “The deal is another reminder that the A.I. race is not just about models,” agreed Jennifer Pruitt, a portfolio manager at Liberty Street, which is invested in SpaceX. “It is also about compute, power, infrastructure, distribution, and the ability to execute at scale.”
Hanging over this constant jockeying for frontier dominance is the simple fact that the business models “are still insufficient to support the A.I. models we have
today,” David Bader, a distinguished professor of computing at New Jersey Institute of Technology, told me. As a result, he expects collaborations like the one between Anthropic and SpaceX to become “the norm”—sort of like how the streaming wars ended with companies tearing down their walled gardens and rebranding as suppliers or platforms. “There will be a consolidation, and these companies will need to support each other and really look at the efficiency of the cost for
operating data centers, both for training and inference,” he said. “And in doing so, they will have to work together.”
|
That’s all for today. I’ll see you next week.
Ian
|
|
|
|
Join Emmy Award-winning journalist Peter Hamby, along with the team of expert journalists at Puck, as they let you in on the
conversations insiders are having across the four corners of power in America: Wall Street, Washington, Silicon Valley, and Hollywood. Presented in partnership with Audacy, new episodes publish daily, Monday through Friday.
|
|
|
|
A professional-grade rundown on the business of sports from John Ourand, the industry’s preeminent journalist, covering the
leagues, players, agencies, media deals, and the egos fueling it all.
|
|
|
|
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
|
|
|
|
|