Chamath Palihapitiya questions who is really profiting from the AI boom

Chamath Palihapitiya is asking if the money being poured into artificial intelligence is paying off for anyone beyond the small group of companies already collecting it.

The Social Capital founder, in two posts made on X on July 17 and 18, called out AI labs over how they trained their own models versus how they treat others that copy their work. Chamath then pointed out what he noticed about buzzwords that dominate SEC filings and how they tend to deflate over time, a reference to the current buzz of AI and agentic technologies.

He wrote, “Right now everyone is glomming onto AI like it’s a life raft. But these same folks have not yet shown repeatable, audited, verifiable ROI even as their CapEx and OpEx are increasing with token costs on all things AI.”

Uber, Microsoft, and Meta are already reining in AI budgets in response to findings from a McKinsey survey that reported that most companies see no earnings impact from generative AI. Other critics just don’t see a future where the whole market is not dragged down by one big stumble.

Did the AI Labs play by different rules?

Palihapitiya threw a jab at Anthropic and, by extension, other players in the frontier model space, like OpenAI, on Friday when he posted an assessment from Anthropic’s Fable model about distillation, the practice of using one model’s outputs to train a cheaper rival.

Palihapitiya stated distillation as a moral problem is genuinely contested. He pointed out that the labs themselves built their systems on the open internet, including copyrighted books, articles, and code, among others.

Now the same firms that relied on these resources scraped across the world to train their frontier models are now against others doing the same thing to them.

Are big tech budgets finally being reined in?

The froth is what large tech firms are now trying to drain. Uber’s AI budget for 2026 was exhausted in roughly four months. The company had to cap coding tools at $1,500 per employee per tool, and it is reportedly tracked on an internal dashboard.

Microsoft is phasing out Claude Code licenses in its Experiences and Devices division. It is now encouraging its engineers to use GitHub Copilot CLI, a move an internal memo referred to as deliberate benchmarking.

In an April memo, Meta’s CTO Andrew Bosworth said, “All motion is not progress and token usage alone is not a measure of impact of any kind.”

MIT’s NANDA initiative looked at enterprise AI pilots across industries and found that 95% of them produced no measurable financial return at all.

Is the AI industry too concentrated to fail safely?

Technology critic Ed Zitron warned this week that OpenAI has become “one of the largest liabilities in recent economic history,” stating that its failure would be the AI era’s Lehman Brothers moment.

By his accounting, OpenAI plans to spend more than $50 billion on compute this year, has taken on around $748 billion in obligations to Microsoft, Amazon, and Oracle, and posted a $38.5 billion net loss in 2025 on $13.07 billion in revenue.

Oracle, which has committed over $340 billion to build capacity for OpenAI, saw its credit rating cut to the lowest investment-grade rung by S&P Global, with OpenAI named as a key risk.

That is the setup Palihapitiya is prodding. A few firms sit at the center of trillions in committed spending; however, the returns remain unproven for most buyers.

Palihapitiya acknowledged that AI is real, adding that “it is the defining change of our lifetime.” However, what he wants to change is that returns should start coming in for more companies than a few, writing, “We are in the early phase where a few companies are making all the money from our largesse. This needs to be reset for everyone to win.”