Chamath Palihapitiya warns AI data‑center slowdown could deal a disproportionate blow to Anthropic and OpenAI

During a recent All‑In Podcast, Chamath Palihapitiya sounded an alarm that the AI boom might soon run into a very human snag – local push‑back against massive data‑center projects. He said the very thing that fuels today’s race for ever‑larger language models – cheap, abundant compute – could start to dry up if towns and state officials become stricter about where these power‑hungry facilities go.

Palihapitiya pointed to a heated public hearing in Cochise County, Arizona, where engineers from Space Exploration Technologies (the company better known as SpaceX) laid out a plan for a 6‑gigawatt data‑center. The pitch was glossy: a promised 40 % cut in electricity costs, zero water use and a 20 % reduction in property taxes, plus enough revenue to retire a county bond. Yet the audience reacted with more than a few raised eyebrows and outright hostility.

"They were getting run over," he remarked, suggesting that the backlash isn’t just about noise or traffic – it’s tied to broader resentment over tech‑giants and perceived economic inequality. Palihapitiya wondered whether the biggest players really need generous tax breaks to pour $30‑$50 billion into infrastructure that could later generate $300‑$500 billion in revenue.

What’s at stake, according to Palihapitiya, is the balance of power in the AI world. Companies like Google, Microsoft and Amazon already own massive stretches of compute and are adding more. In contrast, newer “neocloud” firms – CoreWeave, Iren, Nebius – and AI‑only outfits such as Anthropic and OpenAI largely rent the chips they need from those hyperscalers.

If new data‑center construction slows, those renters will feel the pinch. “If you constrain the assets that can actually generate compute, the leading companies will have to pay more,” Palihapitiya explained. That could push hyperscalers to keep more of their fresh capacity for internal projects – think Google directing extra silicon to its Gemini model instead of leasing it out.

The ripple effect could be dramatic for Anthropic and OpenAI, which reportedly operate on 60‑ to 90‑day model‑improvement cycles. A tighter supply of compute would lengthen those cycles, giving rivals a chance to close the gap.

Palihapitiya also warned that incoming governors are likely to demand concrete economic benefits before green‑lighting such projects, further throttling the build‑out speed. In his view, the infrastructure crunch could become a competitive advantage for firms that already own their own racks, while renting‑heavy AI developers might see their ambitions curbed.

Market reaction was modestly positive on the day of the interview – the SPDR S&P 500 ETF (+0.17 %), Invesco QQQ (+0.21 %) and the Dow Jones ETF (+0.06 %) all edged higher – but sentiment on StockTwits for the S&P 500 leaned bearish. Meanwhile, AI‑focused ETFs like Global X AIQ (+1.32 %) and iShares U.S. Tech IYW (+1.31 %) posted solid gains, underscoring that investors remain hungry for exposure despite the looming compute bottleneck.