At GTC 2026, AI infrastructure CEOs including Perplexity's Aravind Srinivas explicitly dismissed responsibility for job displacement caused by AI, with Srinivas arguing workers should "look forward to" displacement and that "most people don't enjoy their jobs." The companies…
At the All-In podcast recorded at GTC this week, four AI infrastructure CEOs sat together and arrived at a remarkable consensus: the human cost of what they are building is not their problem. Perplexity cofounder and CEO Aravind Srinivas put it plainly — displaced workers should look forward to what is coming. "Even if there is temporary job displacement to deal with, that sort of glorious future is what we should look forward to," he said, according to a Fortune report. His addendum, that "the reality is most people dont enjoy their jobs," appeared designed to make the first part land easier. It did not.
That line would be unremarkable from a founder at a pitch competition. It is a different signal from a CEO whose company is valued at $21.21 billion after a Series E-6 funding round closed in February 2026, carries a $750 million three-year commitment to Microsoft Azure, and projects annual recurring revenue of approximately $200 million — up from $80 million in late 2024. Perplexity is not yet profitable. It is also not short on confidence about what it deserves.
CoreWeave CEO Michael Intrator was present. His company posted $5.13 billion in revenue in 2025 and lost $1.2 billion net. The operating loss was a comparatively modest negative $46 million — the gap between revenue and net income driven largely by interest and depreciation on the GPU fleet that makes the company run. CoreWeave plans to spend between $30 billion and $35 billion in 2026, according to an earnings call covered by CNBC, against a current market capitalization of $43.6 billion as of mid-March. The company has navigated cost of capital down 300 basis points over twelve months, saving approximately $700 million annually — meaningful relief, but not yet structural solvency. Kerrisdale Capital has been short the stock since 2025, calling CoreWeave a debt-fueled GPU rental business with no durable moat. The short position has not closed.
IREN co-CEO Daniel Roberts spoke from a different position on the certainty spectrum. His company has secured $9.3 billion in funding over the past eight months, across customer prepayments, convertible notes, GPU leasing, and GPU financing. In November 2025, IREN signed a $9.7 billion contract to offer its data centers to Microsoft. The company announced in March 2026 that it has entered purchase agreements for over 50,000 NVIDIA B300 GPUs, expanding its total fleet to 150,000 GPUs — a scale intended to support an annualized run-rate revenue target of over $3.7 billion by the end of 2026. Roberts frames this as investing in an AI cycle where supply and demand are "so tilted" that outsized returns are likely, a position defended on Stocktwits. The word likely does real work in that sentence.
Mistral CEO Arthur Mensch offered the most technically grounded comments of the four, arguing that open models give customers control over the technology they deploy, the ability to customize it for specific needs, and resilience against vendor dependency. Mistral Small 4, released in March, uses 128 experts with 4 active per token, totaling 119 billion parameters with 6 billion engaged at inference and a 256,000-token context window. Mensch also announced Mistral Forge, an enterprise training platform that proposes training models directly on internal documentation, codebases, compliance frameworks, and operational records — from pre-training through reinforcement learning, according to UC Strategies. That product, if it ships as described, is a meaningful bet that enterprise customers want to own their models rather than rent inference from someone elses API.
Nvidia CEO Jensen Huang, who hosted the GTC event where all four CEOs appeared, projected that AI infrastructure will generate at least $1 trillion in revenue from 2025 through 2027, a figure he arrived at by noting that computing demand has increased approximately one million times over the last several years. Whether that revenue accrues to the companies currently burning capital to build the infrastructure — or to their customers, their creditors, or the workers they are displacing — is a question none of the four CEOs addressed on the podcast.
Block CEO Jack Dorsey, who was not on the All-In panel but whose name came up in coverage of it, reduced his company headcount by 40% earlier this year and cited AI intelligence tools as the reason. Dorsey's bluntness is an outlier in a week where the dominant frame from AI infrastructure leaders was that displacement is temporary, that the work most people do is not worth mourning, and that the returns will justify the disruption. Those are separate claims. The first may be true. The second is a value judgment made by people who are not losing the job. The third has not been demonstrated.
The money moving into AI infrastructure right now is real. The $9.3 billion IREN raised in eight months, the $30 to $35 billion CoreWeave plans to spend this year, the $9.7 billion Microsoft contract with IREN signed in November — these are not speculative figures. They represent real capital being deployed against a real bet that AI inference and training demand will grow enough to justify the buildout. What the numbers do not capture is who pays while the bet resolves. The CEOs on the All-In stage this week did not pretend otherwise. They simply did not think it was their section of the conversation.