Meta's Adam Mosseri Floats Per-Engineer AI Token Caps as Chamath Palihapitiya Warns of Earnings Hits

Since Meta pulled the plug on its internal AI token spend leaderboard earlier this year, the company's AI cost problem has only gotten more attention. Instagram head Adam Mosseri now says the next step could be formal caps on how many tokens individual engineers are allowed to burn.

Speaking on Lenny's Podcast, Mosseri said he can imagine a near future, maybe a year or two out, when "the burn rate of a strong engineer might be the same as their salary, or their cost of employment." His point: once AI usage costs rival payroll, companies will have to manage it the same way they manage headcount or GPU capacity. Meta doesn't have per-employee token caps right now, according to TechCrunch, but Mosseri thinks that changes soon.

His analogy is straightforward. Meta already rations GPUs, storage, RAM, labeling budgets and payroll across teams. Token spend, he argues, is just the newest line item that needs the same discipline, with caps set based on how much a company trusts a given employee or team to spend tokens in a way that actually returns value.

The leaderboard Meta killed

Mosseri referenced Meta's decision to shut down an internal tool that tracked how many tokens employees were burning, a leaderboard-style system that, according to TechCrunch, was pushing the company toward billions of dollars in AI costs in 2026. Mosseri didn't mince words about what that leaderboard produced: "It's not that hard to build a token incinerator, and that doesn't create a lot of value," he said.

Meta isn't the only shop that's hit this wall. Uber blew through its entire 2026 AI coding budget by April, according to TechCrunch. Microsoft canceled Claude Code licenses for its engineers and pushed them onto its own Copilot CLI tool instead. That move reads less like innovation strategy and more like a company trying to stop the bleeding on a vendor bill.

Palihapitiya says the bill is coming for earnings

While Mosseri is talking about internal engineering budgets, investor Chamath Palihapitiya is warning the problem is bigger and less visible than most executives realize. Palihapitiya, founder of Social Capital and CEO of the AI company 8090, told CNBC that "CEOs and the CFOs, in my opinion, probably have no idea how much tokenmaxxing is going on inside of their organizations."

His prediction is blunt: a company is going to miss earnings by a few pennies per share, and the CEO is going to turn to the CFO asking what happened, only to find out AI token spend quietly ate the margin. Palihapitiya said his own company's AI spending is trending past $10 million a year, which he called "very scary" for the founder of a small startup. In a post on X earlier this year, he said he suspects "many other companies are also feeding this revenue ramp without getting any meaningful ROI from it."

Palihapitiya's warning lines up with comments from Palantir CEO Alex Karp, who told CNBC's Squawk Box that OpenAI and Anthropic's token-based pricing models have gone sideways for enterprise customers. "Something has gone completely wrong," Karp said, arguing that companies are effectively wasting time and money "chillaxing" with tokens instead of getting real productivity gains.

The case for skepticism on the doom framing

Token costs are, according to multiple AI model makers, expected to keep falling as competition between OpenAI, Anthropic, Google and others intensifies. Mosseri himself said he expects a pricing war to eventually bring costs down. Companies experimenting hard with AI right now, even wastefully, are also the ones building institutional knowledge about what actually works, which has value that doesn't show up cleanly in a quarterly token invoice.

Palihapitiya's specific complaint, that finance leaders may not have visibility into what's being spent, is a legitimate governance gap, not speculation. It's the same problem companies had with cloud computing sprawl a decade ago: distributed spending decisions made by individual engineers or teams, invisible to the CFO until the bill lands.

Palihapitiya himself isn't a neutral messenger here. He built his public reputation partly by promoting SPACs during the pandemic that later collapsed, wiping out value for retail investors who bought in on his pitch. He acknowledged as much on CNBC, calling it a "huge mistake" to have hyped those deals on social media and on CNBC itself, and saying the people who lost money were "speculators" whose incentives were misaligned with his own. That history doesn't make his current AI warning wrong, but it's worth knowing who's delivering it.

Neither Meta nor Uber nor Microsoft has disclosed exact dollar figures tying token overspend directly to a specific earnings miss. Whether Palihapitiya's prediction of a token-driven EPS surprise actually materializes at a major public company is the open question analysts and CFOs will be watching for in upcoming quarterly reports.