Welcome to Eye on AI, with AI reporter Sharon Goldman. The professional-Iran meme machine trolling Trump with AI Lego cartoons…Amazon’s Andy Jassy defends Amazon’s $200 billion spending spree...OpenAI pauses Stargate U.Okay. knowledge middle, citing power prices.
It’s been one other a kind of wild weeks in AI, with Anthropic electing not to launch its new Claude Mythos mannequin due to issues in regards to the cybersecurity dangers it poses (and forming a coalition to make use of a preview model of the mannequin to bolster cybersecurity defenses); Meta releasing its first AI mannequin since hiring Alexandr Wang; and mounting expectations about OpenAI’s upcoming new “Spud” mannequin.
Most of those AI fashions run on Nvidia GPUs, the subtle and costly AI chips (at over $30,000 a pop) that energy their coaching and output. However throughout the business, entry to these chips stays a bottleneck. OpenAI president Greg Brockman, for instance, has stated allocating GPUs at OpenAI is “pain and suffering.”
This week, on the HumanX convention in San Francisco, I found that even inside Nvidia, GPUs are scarce.
I sat down with Bryan Catanzaro, who leads utilized deep studying analysis at Nvidia, overseeing groups engaged on AI-driven graphics, speech recognition, and simulation. Catanzaro was additionally among the many first, again within the early-to-mid 2010s, to note researchers snapping up Nvidia GPUs to coach AI fashions—a sign that helped push CEO Jensen Huang to double down on AI, setting the stage for the corporate’s now-historic run.
Right this moment, although, even Catanzaro’s groups are struggling to entry sufficient GPUs. “My team uses AI very deeply in our work, and their primary complaint is they want higher limits,” Catanzaro instructed me. “They want more GPUs.”
“Efficiency is also intelligence”
In reality, he stated certainly one of his predominant jobs now could be merely attempting to safe extra compute for his groups. “We’re all supply constrained,” he stated. “Jensen will say, ‘I’m sorry, Bryan, but those are sold.’ We operate within those constraints.”
Certainly one of Catanzaro’s initiatives has been main the staff constructing Nvidia’s Nemotron, a household of fashions which can be open supply—which means customers can freely obtain them to make use of, examine, or modify. To be clear, Nvidia isn’t attempting to compete within the model-building race with the likes of OpenAI and Anthropic. As an alternative, it’s constructing them to strengthen a developer ecosystem that continues to be tied to Nvidia {hardware} and software program.
The Nemotron fashions are recognized for being notably GPU-efficient. And Catanzaro stated it’s the very constraints on GPU entry at Nvidia itself that’s driving the push to make Nemotron fashions extra environment friendly. “In a supply-constrained world, efficiency is also intelligence,” he stated.
Now not a science mission
However surprisingly, effectivity isn’t unhealthy for enterprise. Catanzaro stated it was Jevons Paradox at work: When one thing turns into extra environment friendly, demand usually surges. “People find all sorts of new ways to use a thing when it gets more efficient,” he stated.
Nonetheless, he acknowledged that Nemotron’s rising visibility inside Nvidia has additionally helped unlock extra sources. “We’ve been working on [Nemotron] for a long time, but it’s really only in the past six months that it’s gotten more attention. As people inside Nvidia better understand the importance of this work, you get better storytelling, better collaboration, and more support across the company.”
Nvidia has realized, he added, that it might now not take a hands-off method to the AI ecosystem. Up to now, Nvidia might depend on others to construct the fashions and functions that drove demand for its chips. Now, as AI turns into extra aggressive and chip-constrained, the corporate sees a extra energetic function for itself in shaping how that ecosystem develops.
“In the past, some people felt like we could just let the ecosystem take care of itself,” he stated. “Now it’s much more obvious that Nvidia has a bigger role to play—a real responsibility and opportunity with Nemotron.”
That framing additionally helps elevate the Nemotron work inside Nvidia, the place groups are competing for scarce GPU sources. “This isn’t a science project,” Catanzaro stated. “It’s not just me asking for resources for my team. This is about Nvidia’s future.”
FORTUNE ON AI
Meta unveils Muse Spark, its first AI mannequin since hiring Alexandr Wang and a bellwether for CEO Mark Zuckerberg’s multi-billion greenback AI push–by Jeremy Kahn
Supermicro launches inside probe after cofounder’s arrest on costs of $2.5 billion in chip smuggling–by Amanda Gerut
A Meta worker created a dashboard so coworkers can compete to be the corporate’s No. 1 AI token person—and Zuckerberg doesn’t even rank within the prime 250–by Jacqueline Munis
AI IN THE NEWS
The professional-Iran meme machine trolling Trump with AI Lego cartoons. A brand new report from Wired describes how a gaggle of younger pro-Iranian creators known as Explosive Media is utilizing AI-generated, Lego-style movies to unfold subtle, viral propaganda through the present battle, reaching hundreds of thousands throughout TikTok, X, and Instagram. In contrast to conventional state messaging, the movies mix humor, internet-savvy cultural references, and simplified storytelling to resonate with American audiences, even incorporating memes and English-language rap. Researchers say the technique is efficient as a result of it distills complicated geopolitical occasions into extremely shareable content material whereas tapping into current disaffection within the U.S., illustrating how AI instruments are enabling a brand new form of “slopaganda” struggle—the place affect campaigns are quicker, extra focused, and much more culturally fluent than previously.
Amazon’s Andy Jassy defends Amazon’s $200B spending spree. GeekWire reported on Amazon CEO Andy Jassy’s newest shareholder letter, which revealed that AWS’s AI enterprise has already reached a $15 billion annual income run charge, which Jassy argued means demand is robust sufficient to justify roughly $200 billion in deliberate capex. Jassy framed AI as a “once-in-a-lifetime” alternative and positioned Amazon squarely in the course of the present AI “land rush,” pointing to surging demand for its customized chips like Trainium—a few of that are already largely offered out years prematurely—in addition to curiosity from prospects wanting to safe future capability. The letter makes clear that Amazon is betting aggressively on proudly owning extra of the AI stack, from infrastructure to chips to doubtlessly promoting these capabilities externally.
OpenAI pauses Stargate UK knowledge middle, citing power prices. Based on Bloomberg, OpenAI is pausing its deliberate Stargate knowledge middle mission within the UK, highlighting how even essentially the most aggressive AI infrastructure buildouts are operating up towards real-world constraints like power prices and regulation. The transfer comes as the corporate reins in spending forward of a possible IPO and narrows focus to its core ChatGPT enterprise amid intensifying competitors from Anthropic and Google. Whereas OpenAI says it nonetheless sees long-term potential within the UK, the choice underscores a broader actuality: Huge AI infrastructure bets—from Texas to Norway to the UAE—are more and more formed not simply by ambition, however by economics, geopolitics, and entry to inexpensive energy.
EYE ON AI NUMBERS
75%
That is what number of executives say their AI technique is extra about optics than any precise inside steering, based on Author’s new 2026 Enterprise AI Adoption Report, which surveyed 2,400 data employees together with 1,200 C-suite executives and 1,200 workers. As well as, 39% haven’t any plan for the way AI truly drives income. But, 69% are planning layoffs this 12 months.
In a LinkedIn submit, Author CEO Might Habib known as this development “‘AI theater’ at its worst,” including “this strategy vacuum up top is literally tearing companies apart.”
AI CALENDAR
June 8-10: Fortune Brainstorm Tech, Aspen, Colo. Apply to attend right here.
July 6-11: Worldwide Convention on Machine Studying (ICML), Seoul, South Korea.
July 7-10: AI for Good Summit, Geneva, Switzerland.
Aug. 4-6: Ai4, Las Vegas.
