Meta has a problem it cannot open-source its way out of. The company's next-generation AI model, code-named Avocado, was supposed to launch in March 2026. It did not. Reuters reports the release has been pushed to May or June, and internal benchmarks show it falling short of frontier competitors from Google, OpenAI, and Anthropic.

But here is the part that should make every Llama believer nervous: according to TestingCatalog, Meta leadership has reportedly discussed temporarily licensing Google's Gemini to fill the gap. Let that sink in. The company that built its entire AI identity on open-source independence is considering renting a competitor's brain.

Avocado represents a fundamental shift in Meta's strategy. Unlike the Llama family, which was released as open-source, Avocado will be proprietary. No weights, no downloads, no community tinkering. Mark Zuckerberg spent two years preaching that open-source AI was the future. Now his most important model is closed, and it is still not ready.

The reason for the pivot is not hard to find. DeepSeek proved that open-source models can be used against you. The Chinese lab built a highly competitive system using Llama and Qwen components, essentially turning Meta's own generosity into a weapon. Once that happened, the calculus changed overnight.

Meta is also burning cash at an extraordinary rate. The company raised its 2026 capital expenditure guidance to $115 to $135 billion, and investors are getting restless. A $14.3 billion investment in Scale AI for data labeling. Massive GPU clusters. And the model that is supposed to justify all of it? Delayed.

This is happening while Google just dropped Gemma 4 under Apache 2.0, directly threatening Llama's open-source dominance. While Alibaba shipped Qwen 3.6-Plus with omnimodal capabilities. While Microsoft declared it is pursuing superintelligence independently. Meta is not just behind on Avocado. It is getting flanked on every side.

The biggest question nobody at Meta wants to answer: if Avocado launches in June and it is merely competitive, was $135 billion in capex worth it? And if they end up licensing Gemini even temporarily, what does that say about the strategy Zuckerberg has been selling to investors for the last three years?

First reported by Reuters. Additional details from The Next Web and TestingCatalog.