Meta's big gamble on AI has been a disaster; the mother of all losses. While investing in GPU computing power at least shows some results, the acquisition of the team cost $16.3 billion, which is about one-third of Meta's annual net profit. So far, no particularly impressive成果 has emerged, and Meta still lags far behind competitors in AI capabilities.

In June 2025, Meta spent $14.3 billion to acquire a 49% stake in Scale AI. Based on the latest 2026 valuation of around $12 billion, this is nearly a 60% decline from when Meta first invested.
With self-evolving agents like openclaw, the reliance on manual annotation for large models is decreasing. Meta originally bought AI data mining rights, only to find out that it was being mined as a gold mine itself.
By the end of 2025, Meta acquired Manus AI for over $2 billion. The strategic direction Meta took in the AI Agent track was promising before openclaw appeared, but now, perhaps not so much. With openclaw's emergence, other companies have quickly caught up with Manus, achieving similar results at lower costs.
A $2 billion price tag is extremely high for a team whose business model is not yet fully closed-loop. In the fast-changing world of AI, whether the team and product are worth that much is something everyone has their own judgment on.
Zuckerberg's logic: competitiveness = compute power (1 million H100s) + algorithms (Llama open-source ecosystem) + data/agents (Scale AI & Manus). It sounds impressive, but upon closer inspection, each seems to have its weaknesses and no particularly strong advantage.
Meta's previous big push into the metaverse ended in a hasty retreat. Now, rushing into AI again, is anyone still optimistic about Meta's future development?