“SpaceX isn’t an AI company,” Hoffman said in a conversation with Rana el Kaliouby on her Pioneers of AI podcast. “XAI is, as Elon himself has described, it’s a complete train wreck for its kind of building of foundational models and other kinds of things.” He also noted that all of its founders have left, and it’s on its “third restart.”

The xAI cofounder exodus has been well documented. By May 2026, all 11 of xAI’s original cofounders had departed the company, a cascade that began in earnest in February when Tony Wu, described as one of the most operationally central cofounders, announced his resignation. Musk restructured xAI’s teams in response, but the departures continued. The company’s flagship Grok models have faced persistent criticism for lagging behind competitors from Anthropic and OpenAI in benchmark performance.

The timing of Hoffman’s remarks is pointed. SpaceX went public on June 12, with AI central to its IPO narrative. Within days, the company announced it was acquiring Cursor, the AI coding tool. Hoffman’s read: That’s not proof of AI capability, but evidence of its absence. “You could almost think of it as the IAC of AI,” he said, invoking the serial acquisitions roll-up strategy of Barry Diller’s internet-era conglomerate. “Use the market cap to buy AI companies and try to buy your way into relevance.”

His assessment of SpaceX’s core compute business was equally withering. The company has positioned revenue from leasing AI infrastructure—including to Anthropic—as validation of its AI credentials. “You’re a premium-priced CoreWeave,” Hoffman said. “I get it. Which is not an AI company.”

The Anthropic situation

If Hoffman’s SpaceX commentary was skeptical, his reaction to the U.S. government forcing Anthropic to pull its Fable and Mythos models from the market was something closer to alarmed.

The directive, issued by the U.S. government on June 11 as an export control order, suspended all foreign national access to the two models. The trigger, according to reporting from Fortune, was Amazon CEO Andy Jassy raising the alarm about a discovered jailbreak in the Fable 5 model—a vulnerability that Anthropic itself had been working to address. Cybersecurity experts widely criticized the government’s response as disproportionate and poorly scoped.

Hoffman lands in a similar place. “It doesn’t look like there’s anything that’s a particular principled, ‘Here’s the way that we’re navigating through things’—apply kind of a rule of law and predictability,” he said. “It’s more like, ‘Hey, we kind of had some contentious interactions with this company anyway, so we’re going to hit them with a stick.’” And that doesn’t come with any kind of principled explanation, he added.

He called the approach “autocratic willy-nilly” and “very sub-optimum,” while acknowledging there may be a legitimate cybersecurity basis. The asymmetry—Anthropic penalized while OpenAI was not—is what troubles him most. Notably, Anthropic itself had flagged security concerns about the models, a detail that el Kaliouby raised in the conversation.

For a company preparing for what is expected to be one of the largest IPOs in history, unpredictable regulatory intervention creates a new category of investor risk—one that the Fable/Mythos episode has now made concrete.

Room for both—but not for everyone

Hoffman, who is an investor in both Anthropic and OpenAI, pushed back firmly on the narrative that the two companies are in a zero-sum race. “We tend to want to tell these stories as cage matches,” he said, as in two companies enter and only one leaves, but “in fact,” he claimed, “there’s a lot of room for both of them to win incredibly.”

He sketched distinct competitive lanes: Anthropic strong in code and expanding into design and legal; OpenAI and ChatGPT functioning more like a consumer search front end, with its Codex coding product “insufficiently talked about” given its strength. The one pointed question he raised: whether Cursor, just acquired by SpaceX, had already peaked. “Cursor seems to have had its bright star some number of months ago and seems to be fading over the horizon,” he said. Cursor has faced mounting pressure since early 2026 as Claude Code and Codex have gained ground, with developers increasingly questioning whether a stand-alone coding IDE (integrated development environment) still commands a premium. But of course, Hoffman has a vested interest in arguing each of these positions.

On the subject of a bubble, he offered a framework for understanding speculation in the markets. It’s wrong to say all of the valuations are crazy, he said, but not some: “The trick is, which ones?” His anchor for the bullish case on OpenAI and Anthropic: If AI becomes as pervasive as electricity, these will be two of the primary utilities—and the revenue model doesn’t need to be fully visible today. Google’s early theory of monetization, after all, was enterprise servers, and then came AdWords—“the best business model invented in human history” thus far.

Tell Gen Z to stop booing AI

On the question of how young people should navigate an AI-saturated job market, Hoffman’s advice was blunt and cut against a prevailing narrative.

“I’ve been thinking about writing an essay on the kind of mistakes that are made by college graduates booing or otherwise dissing AI,” he said. As if speaking to all of Gen Z, he added: “You guys have the opportunity to be generation AI—where you come into the workforce saying, ‘I know this a lot better than all of you. You should be hiring me in order to help you become AI-native organizations and so forth.’ And it should be an opportunity, not a threat.”

Goldman Sachs publishes a semi-regular AI tracker, which found in April 2026 that AI was already erasing roughly 16,000 net U.S. jobs per month, and 11,000 as of earlier this month, with Gen Z bearing a disproportionate share of the impact as entry-level knowledge roles face the highest displacement risk. Separate research found graduate unemployment had risen from 3.6% in 2019 to 5.6% in 2026. By mid-2026, 35% of entry-level job postings required at least three years of experience, and 45% of companies were using automated rejection systems at early hiring stages.

Hoffman’s counter is that most of that pain is being misattributed. The entry-level slump, he argued, has more to do with other factors—with “AI washing” doing much of the narrative work in the interim. “It’s actually, in fact, because of global turbulence and businesses being unable to figure out how to invest and plan,” he said. “It’s because of basically over-hiring in the pandemic and the, ‘Hey, maybe remote work really does work.’ ‘Oh, right. Remote work’s pretty hard to make work.’”

His prescription, drawn from his book Superagency, is an agency mindset: Treat AI not as a threat to your career but as the instrument of it. “The AI is my tool, companion, car, et cetera, as I navigate things,” he said. “The AI can do a whole bunch of amazing things itself but is not complete—and humans can add in a lot of significant and important things.”

The Microsoft chapter closes

Hoffman’s departure from the Microsoft board—he chose not to stand for reelection but remains on through year-end—closes a chapter that included facilitating the LinkedIn acquisition (“one of the epic M&As of history”); steering the GitHub purchase; and helping broker trust between Microsoft and OpenAI in the early days of their partnership. He described the decision simply: He’d rather be a founder than a governance person. He and Microsoft CEO Satya Nadella, he noted, still talk strategy. “As a matter of fact, Satya and I were just on the phone today.”

What comes next is drug discovery. Manas AI, his company with cofounders Ujjwal Singh and Dr. Siddhartha Mukherjee, is generating small molecule proposals that its computational chemists are calling genuinely promising—the trigger, Hoffman says, for his decision to go all in. The pitch deck, he revealed, describes Manas as “an AI drug discovery factory for creating monopolies”—legally permissible, he notes, because pharmaceutical IP functions as a sanctioned monopoly by design. For the man who helped build the LinkedIn and Microsoft era of tech, it may be the longest-horizon bet he’s ever made.

For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.