{{ 'now' | timezone: 'America/New_York' | date: '%b %d, %Y' }}
|
|
|
|
Welcome back to The Hidden Layer. I’m Ian Krietzberg, with an inside look at how
the war in Iran could delay, or even derail, the A.I. takeoff. (As my partner Leigh Ann Caldwell reports, Hill Republicans aren’t pressuring Trump to deescalate… yet.) People have been yapping about the A.I. bubble popping for years. Has the catalyst finally arrived?
Plus, news and notes on Elon Musk’s hard reboot,
Zuckerberg’s “Avocado” retreat, and the Pentagon’s problem with Claude’s “soul.”
Also discussed in this issue: Sam Winter-Levy, Nada Sanders, Phil Kornbluth, Gil Luria, Emil Michael, Mira Murati, Jony Ive, Manuel Kroiss, Ross Nordeen, Benjamin De Kraker, Jason Ginsberg,
Andrew Milich, Andy Stone, Dean Ball, Horace He, Peter Oppenheimer, Brian Kennedy, Jensen Huang, and more…
|
Four Things You
Should Know…
|
- What
the hell is happening at xAI?: As the evidence piles up that xAI has fallen light-years behind OpenAI, Anthropic, and Google in the A.I. race, C.E.O. Elon Musk is initiating a hard reboot. “xAI was not built right first time around, so is being rebuilt from the foundations up,” he tweeted last week. Musk also admitted at a recent public appearance that xAI’s “Grok is currently behind in coding.” At the same time, an exodus of talent has left the company with just three of its original dozen co-founders: Musk, Manuel Kroiss, and Ross Nordeen. “I strongly suspect that many of the foundational problems at xAI were kept hidden from Elon until very recently,” former xAI employee Benjamin De Kraker
tweeted in response. “Those of us who were further down saw the cracks but were told to shut up.”
Musk reportedly held an all-hands meeting last Wednesday, during which he predicted xAI could catch up to its rivals within the next few months. He also hired Jason Ginsberg and Andrew Milich, two former leaders from A.I. coding startup Cursor,
who will report directly to him. But xAI faces all sorts of deeply rooted problems: Musk is, of course, notoriously difficult to work for. And there’s likely less financial upside for new employees after xAI acquired X (formerly Twitter) and absorbed all of its debts, then was subsequently acquired by SpaceX in a deal that reportedly valued the company at $1.25
trillion. Musk has said an I.P.O. is imminent. - Is Avocado toast?: Meanwhile, over in Menlo Park, Meta hasn’t released a new model in months and is reportedly delaying the launch of its latest
L.L.M.—code-named “Avocado”—due to concerns that its performance wasn’t good enough compared to the competition. Reuters also reported that C.E.O. Mark Zuckerberg is considering slashing head count by 20 percent to offset the costs of Meta’s massively expensive infrastructure capex. The company currently employs just shy of 80,000 people, after cutting about 10,000 in 2022 and another 10,000 or so a year later. Meta spokesperson Andy Stone told me that “this is speculative
reporting about theoretical approaches.”
- Claude’s got a soul problem: Undersecretary of Defense Emil Michael went on CNBC last week to defend the department’s decision to label Anthropic a supply chain risk, pointing to the company’s (perhaps poorly named) “constitution”—an internal document intended to guide models’
behavior. “Their model has a soul, has a constitution that’s not the U.S. Constitution,” Michael said. “The other day their model was ‘anxious,’ and they believe it has a 20 percent chance of being sentient and [having] its own ability to make decisions.” Sounds pretty woke.
Michael went on to hand-wave about the risk of hallucination—entirely unrelated to Anthropic’s self-stated (and entirely groundless) theorizing about sentience—which is not unique to Claude. “Emil
Michael now appears to be making an argument that no generative A.I. should be used in the DoW supply chain,” former White House A.I. policy advisor Dean Ball pointed out in a post on X. - Checking in on those productivity promises: The good news for A.I. boosters is that workplace adoption massively accelerated in 2025,
according to a recent report by ActivTrak, an analytics firm that’s been collecting behavioral data from 1,111 companies and 163,000 employees since 2023. Adoption hit 80 percent last year, up from 53 percent two years ago. But while the promise of A.I. has long been that it’ll make everything easier for everyone, the data tells a different
story. “A.I. is adding to workloads rather than redistributing them,” according to the report. It’s an interesting, albeit early, data point amid the intensifying debate over whether A.I. will “amplify” work, as ActivTrak contends, or simply replace workers altogether.
|
|
|
|
Social workers are identifying at-risk veterans sooner with the help of AI-powered tools. Too many of our veterans reach a crisis point before anyone knows. Now, the VA uses AI to spot the warning signs early — and make sure someone calls before it’s too late. This is AI in action. Learn more at AI4America.com.
|
|
|
|
- Armadin,
a self-described “A.I.-native cybersecurity company,” announced the closure of a $189 million seed and Series A funding round led by Accel, with participation from Google Ventures and Menlo Ventures. The company is pitching its ability to build a fully autonomous swarm of
agents—the “ultimate attacker”—to defend against A.I. cyberattacks.
- Nvidia partnered with Thinking Machines Lab, the Mira Murati–led startup that charged out of the gate in June with a head-spinning $2 billion in funding, a $12 billion valuation, and no product in sight. The partnership features a “significant” investment on
Nvidia’s part and will see the lab deploy a minimum of 1 gigawatt of Nvidia’s latest chips. Confirming that the industry has taken the gospel of scale to heart, Thinking Machines Lab’s Horace He tweeted that “one gigawatt isn’t enough to get to where we want to go, but it’s a start!”
- OpenAI acquired Promptfoo, an A.I. security
company whose suite of tools is intended to red-team and evaluate L.L.M.s. It’s the latest in a growing list of acquisitions by Sam Altman, including the $6.5 billion deal that OpenAI struck with Apple designer Jony Ive, the $1.1 billion it shelled out for
Statsig, and a series of acqui-hires of smaller startups.
|
And now for the main event…
|
|
|
|
The Strait of Hormuz isn’t just a chokepoint for oil—it’s also a major artery for elements
essential to semiconductor manufacturing. And there’s growing fear that the market gloom could leech into A.I. valuations, too.
|
|
|
|
It’s not only gas prices that are spiking after Iran effectively closed the Strait of Hormuz, throttling
one-fifth of the world’s oil supply. As it turns out, the 21-mile-wide channel is also a chokepoint for roughly a third of the world’s helium and about half of the global seaborne trade in sulfur—both essential components in semiconductor manufacturing. “The overarching story is that the world’s most important chip producers depend on energy and chemical
supplies that move through some of the most geopolitically volatile waterways on Earth, and the Trump administration doesn’t seem to have a contingency plan for it,” said Sam Winter-Levy, a fellow at the Carnegie Endowment for International Peace.
The ripple effects could seriously set back the global A.I. race, which has been consuming trillions of dollars in energy and raw materials at an unprecedented clip. In a best-case scenario, helium production
could be disrupted for three to four months, according to industry expert Phil Kornbluth—and the impact could easily stretch longer. Even after the war is over and QatarEnergy restarts its shuttered plants, Kornbluth told me, it would take “around four to five weeks before helium production resumes [and then] several weeks to a month or more to reach markets in Asia and Europe.” Meanwhile, spot prices for sulfur, used to make the sulfuric acid necessary for cleaning chips, have
skyrocketed.
There are also the reverberations of the oil shock itself. “Overall, equities face rising correction risk; valuations are stretched, macro conditions are deteriorating at the margin, and cracks are appearing,” Goldman chief global equity strategist Peter Oppenheimer wrote in a research note this week. Oppenheimer thinks the “weakness should be temporary,” but Goldman economists still raised their 12-month recession odds to 25 percent. At a moment when the
entire A.I. industry is cruising on investor euphoria, a downturn would likely hit the highest-flying stocks first—and force even the biggest V.C. firms to rethink their appetite for risk.
|
Many analysts assume the war in Iran will be short-lived, especially as the White House belatedly comes to
terms with the entirely foreseeable economic impact of conflict in the Persian Gulf. There’s no better way to lose a midterm than to tell Americans not to worry about rising gas prices. But if the crisis continues to escalate, the A.I. industry may be forced to countenance its own black swan event.
|
|
|
|
Social workers are identifying at-risk veterans sooner with the help of AI-powered tools. Too many of our veterans reach a crisis point before anyone knows. Now, the VA uses AI to spot the warning signs early — and make sure someone calls before it’s too late. This is AI in action. Learn more at AI4America.com.
|
|
|
|
Taiwan and South Korea, home to chip fabs TSMC and SK Hynix, respectively, import the vast majority of their
crude oil from the Middle East. “The A.I. boom had already driven chip prices to historic highs and created shortages before the strait was closed,” Winter-Levy said. “A sustained disruption to Korean and Taiwanese energy supplies would ripple through the entire global A.I. ecosystem.” (A TSMC spokesperson told me that the company does “not anticipate any significant impact at this time, and we will continue monitoring the situation closely.”)
Meanwhile, in the U.S., data centers had
already been struggling to get access to enough electrical capacity. The closure of the strait, Winter-Levy theorized, will likely increase the global costs of data center operation, while the general economic uncertainty stemming from the war could slow down both data center and chip factory construction. Dr. Nada Sanders, a professor of supply chain management at Northeastern University, agreed. “We’re not going to see data centers going dark overnight, but I think we’re
looking at prices that are going to creep—a little bit more of a slow boil,” she said. “It’s an energy system stress story.” Furthermore, even if the conflict were to end tomorrow, she contended, it would take months for the A.I. supply chain to normalize.
Of course, supply shocks move through economies at variable speeds. Helium prices, for instance, are typically locked in with long-term contracts, which have yet to rise (though Kornbluth expects that to change soon). Belden, which
sells the fiber cables that enable data center network connectivity, has yet to see a material impact from the war, according to revenue management officer Brian Kennedy. A contact at Chatsworth, which builds storage for data centers, also said there’d been no impact on their business yet. And a source at an electricity company told me that while data center operators are likely to see higher prices for both diesel and natural gas, most hedge energy risks with fixed-price
contracts.
Still, every month matters in an A.I. race, where the competition is moving at exponential rates. Both data center and semiconductor factory construction require hyperscale investments and relatively lengthy time horizons; the construction of TSMC’s second fabrication facility in Arizona, for instance, was completed in 2025, but “volume production” won’t begin until
2028. “Sustained geopolitical volatility complicates those decisions and could slow the build-out of the chip supply that A.I. companies are counting on,” Winter-Levy told me. “In other words, the A.I. supply chain turns out to have some of the same geographic vulnerabilities as the energy supply chain.” The difference, he added, is that “oil has decades of conflict-exposure planning baked into the infrastructure. For A.I., that planning doesn’t exist yet.”
|
The interplay between broader macro conditions and the specific market dynamics of the hyperscaler hype cycle
is difficult to suss out. One way to think about it, said D.A. Davidson analyst Gil Luria, is that if the Iran conflict sparks a “very broad recession,” the A.I. industry will not be able to “escape” it—people would buy less from Amazon; companies would spend less on advertising with Google and Meta, depriving them of cashflow; etcetera. “That’s the mechanism by which this could impact [the industry],” Luria explained. “But it’s way too early to think that’s going to happen,
because if this is just a temporary spike in oil prices and the Iran situation gets resolved one way or another in the next few months, we may not get into a deep recession that would impact these companies.”
I asked Luria what it might take for the hyperscalers to scale back their A.I. infrastructure capex plans, estimated at roughly $700 billion for 2026. “These companies have
very good businesses,” he said. “We’re speculating way down the line. We’d have to be in a pretty bad state for these companies to change course.”
Of course, the entire A.I. economy—propelled by raucous investor excitement and a narrative of almost violent urgency—was unproven even before the war started. Now that the uncertainty has compounded, it’s not hard to imagine how dramatically things could unwind if a prolonged conflict and steadily rising prices erode investors’ risk tolerance.
And if the money faucets get turned down, the industry’s ultra-optimistic projections (Nvidia chief Jensen Huang just forecast $1 trillion in chip revenue through 2027) would almost certainly need to be revised. “If there’s less demand for data centers, there’s less demand for chips,” Luria conceded. “And that’ll ripple all the way through.”
While the full ramifications are impossible to predict, what’s clear is that the U.S.–Israel war on Iran has transformed an
already unstable market sector into something messier—a “hornet’s nest,” as Sanders put it. Worse, said Winter-Levy, it all screams of “strategic incoherence” and was, obviously, entirely avoidable. “Washington is spending tens of billions of dollars through the CHIPS Act to onshore semiconductor fabrication from Taiwan precisely because everyone acknowledges Taiwan sits in a geopolitical hotspot. But the Trump administration has decided that the Gulf states should be its closest partners in
the A.I. data center build-out.” Who could have predicted that launching a war next door might put all that at risk?
|
That’s all for today. I’ll see you on Thursday.
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
|
|
|
|
|