Breaking

Chamath Says Meta 'Completely Fumbled' Its AI Strategy

June 25, 2026 at 08:07 EDT

Foundation Models
Open Source

Investor and former Facebook executive Chamath Palihapitiya said Meta "completely fumbled" its open-weight AI strategy. In an interview published by Axios on The Axios Show on June 25, 2026, the remark is drawing attention as an insider's view from a former Meta executive.

June 25, 2026 · The Axios Show

"Completely Fumbled": A Facebook Insider's Verdict on Meta's AI

Chamath Palihapitiya — the former VP of Growth who helped scale Facebook to a billion users — argues that Mark Zuckerberg's once-clear lead in AI has slipped away, despite spending in the tens of billions.

50M→1B

Facebook's user growth during the era he helped build

$10s of B

Meta's reported AI infrastructure spending plans, as of 2025

405B

Llama 3.1 — among the largest open-source models in 2024

The Llama trajectory — reception by release

Llama 2 / 3

Well received

Llama 3.1 405B

Scale milestone

Llama 4

Muted reception

Each block reflects relative community reception. After early wins, Llama 4 drew concern over weak performance, benchmark questions, and ethical controversy.

What went wrong inside

Synthetic data stumbled

Use in Llama 4 training reportedly faltered, forcing a shift toward manual work.

Teams reshuffled

Internal frustration in Applied AI; Zuckerberg acknowledged "mistakes" in a memo.

Developers leaving

More builders say they are switching to DeepSeek and rival models.

The case for Zuckerberg

The open-source path won early developer goodwill, and the advertising business remains strong — a durable foundation under the AI noise.

The critique

Huge spending hasn't translated into results. The AI stall is seen as bearing directly on long-term competitiveness and Zuckerberg's reputation.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.