After billions in AI spending, Meta may miss deadline it set for itself

Meta’s highly anticipated AI model, codenamed Avocado, faces a two-month delay, missing its mid-March launch target. Falling short of rivals on internal tests, the model is now expected in May. This setback for Meta’s ambitious AI spending raises questions about its ability to compete with industry leaders. Internal discussions even explored licensing Google’s Gemini as a temporary solution.

Meta’s next big AI model—codenamed Avocado—has been delayed by at least two months after falling short of rivals on internal benchmarks, the New York Times reported on Thursday. The model, which was originally targeted for a mid-March release, is now unlikely to ship before May, three people familiar with the matter told the publication. It’s a significant stumble for a company that’s poured billions into catching up in the foundational AI race—and one that raises questions about whether sheer spending can close the gap with Google, OpenAI, and Anthropic.Avocado reportedly outperformed Meta’s previous Llama 4 model and edged past Google’s older Gemini 2.5, but couldn’t keep up with Gemini 3.0 from November.

That places it squarely behind the frontier, despite Meta committing up to $135 billion in capital spending this year—nearly double the $72 billion it spent in 2024.

Meta’s AI leaders have reportedly discussed licensing Google’s Gemini as a temporary fix

Perhaps the most striking detail from the NYT report: leaders within Meta’s AI division have floated the idea of temporarily licensing Gemini to power the company’s products while Avocado catches up. No decision has been made, but the fact that it’s even on the table underscores how urgently Meta needs a competitive model in the market.Avocado is being built by TBD Lab, an elite ~100-person unit assembled under Alexandr Wang, the former Scale AI CEO whom Meta brought on as chief AI officer last June as part of a $14.3 billion investment. TBD Lab finished Avocado’s pre-training late last year and began post-training in January, when the mid-March target was set. But some researchers have already departed, and Wang has reportedly clashed with Meta veterans Chris Cox and Andrew Bosworth over how the new models should improve the company’s advertising business.

Zuckerberg had tempered expectations for Avocado as early as January

On a January earnings call, Zuckerberg acknowledged the model wouldn’t immediately leap to the frontier, saying he expected it to be “good” but more importantly show “the rapid trajectory we’re on.” A Meta spokesperson reiterated that line to Reuters, adding the company is “excited for people to see what we’ve been cooking very soon.”Meta’s next model after Avocado is reportedly already in the works—codenamed Watermelon.