Meta's AI Ambitions Face Setbacks Amid Massive Infrastructure Push

Meta's flagship AI model 'Avocado' is delayed and trails rivals, even as the company signs a massive $27B compute deal and faces reports of major workforce reductions.

This week has proven to be one of the most eventful periods in months for Meta's stock, as the company navigates a delayed flagship AI model, a colossal $27 billion infrastructure agreement, and reports of significant workforce reductions. Despite committing hundreds of billions to artificial intelligence, Meta has internally conceded that its upcoming core model lags behind key competitors.

A Strategic Pivot and Internal Challenges

At the heart of the current situation is the postponed launch of Meta's new AI model, internally codenamed "Avocado." Its release has been pushed back from March to at least May. Internal evaluations reportedly placed the model behind systems from Google, OpenAI, and Anthropic, specifically in areas like reasoning, coding, and text generation. While Avocado is said to outperform Google's Gemini 2.5 from March, it trails the capabilities of Gemini 3.0, which debuted in November.

In a revealing strategic consideration, sources indicate Meta's AI division has discussed temporarily licensing Google's Gemini models to power its own AI products—a move that would create an unusual dependency on a primary rival. No final decision has been made. Concurrently, development continues on other projects, including "Watermelon," slated as the next model generation, and "Mango," focused on image and video generation.

Billions for External Compute Capacity

Undeterred by the model delay, Meta is aggressively accelerating its infrastructure expansion. On March 16, Nebius Group announced a five-year contract with the tech giant. The deal includes $12 billion for dedicated computing capacity based on NVIDIA's Vera Rubin platform, with deliveries commencing in early 2027. Optional capacity additions worth a further $15 billion could bring the total contract value to approximately $27 billion.

This massive external commitment signals that Meta's own data centers are reaching their capacity limits. The company's planned capital expenditures for 2026 are between $115 and $135 billion—roughly double the prior year's level.

Cost Pressures and Operational Efficiency

The enormous spending is creating pressure to streamline costs. Reuters reported that Meta is planning cuts exceeding 20% of its roughly 79,000-strong workforce. The company refuted this characterization, labeling the report as speculative. However, Rosenblatt Securities analyst Barton Crockett estimated that such a reduction could yield savings around $6 billion, potentially boosting the adjusted operating profit by approximately five percent.

Meta's core business provides a solid financial foundation for these investments. In 2025, the company generated revenue of $201 billion, a 22% year-over-year increase, with an operating margin of 41%.

The Road Ahead for Zuckerberg's AI Vision

CEO Mark Zuckerberg declared 2025 the year of "major AI acceleration" during the Q4 2025 earnings call. The company's ambition is to integrate large language models into its social media algorithms, aiming to develop a form of "personal superintelligence." Since the start of the year, Meta's shares have declined by about 7%, trading notably below their 52-week high of €677.80.

Whether Avocado—and the vast infrastructure being built to support it—can fulfill these ambitions will become clearer with the model's anticipated launch in May.