Meta finds itself at a rare moment of public vulnerability as its next-generation AI push falters. What Mark Zuckerberg had pitched as a defining year for Meta’s move into generative AI has been punctured by a delayed model release, internal reorganisations and talk of sweeping job cuts.
The company has postponed the debut of its much‑hyped new family of models, internally codenamed “Avocado,” pushing an originally scheduled March roll‑out to May after internal evaluations judged its capabilities short of the industry’s leading systems. That judgement is sharp: Meta’s new model reportedly improves on its earlier Llama releases but trails Google’s latest Gemini and top models from OpenAI and Anthropic on reasoning, coding and writing benchmarks.
At the same time Reuters has reported that Meta is preparing a major round of layoffs that could affect roughly 20% of its workforce as management seeks to rein in the soaring costs of AI compute and infrastructure. For a company that already cut more than a quarter of its staff across 2022–23 in two waves, another reduction on this scale would mark its biggest cull since the structural shake‑ups that followed the ad‑market slump.
The operational strains extend beyond product timing and headcount. Alexandr Wang, who leads Meta’s newly formed Superintelligence Lab and its core TBD team charged with delivering Avocado, is under intensifying scrutiny. Internal shifts that create a parallel AI engineering team reporting to long‑time executive Andrew Bosworth have fuelled speculation about internal friction and whether Meta’s AI effort is coherently managed.
Meta’s positioning in the AI ecosystem complicates the reaction. The company built a reputation with open‑weight Llama models, attracting developers and enterprises, but later Llama releases drew criticism over benchmark claims and performance gaps. Those missteps helped shape an internal debate on whether to keep new models open or closed — a question that goes to strategy as much as safety and commercialisation.
Recent acquisitions illustrate Meta’s current approach: the purchases of Manus, an AI agent product, and Moltbook, a social layer around the OpenClaw ecosystem, read as tactical bets rather than decisive moves. By contrast, rivals have attempted to internalise crucial talent and capabilities — OpenAI hired the founder of OpenClaw directly — while ByteDance has quietly stitched AI models into vast consumer products, giving it a practical edge in user engagement and monetisation.
The financial stakes are already substantial. Meta has signalled plans to spend heavily on data‑centre and compute capacity; internal forecasts point to sharply higher AI‑related expenditure in the near term. That appetite for capital is colliding with more immediate questions about return on investment: will new models translate into better ad targeting, product stickiness or new revenue streams, or simply balloon costs?
For global observers, Meta’s setbacks matter because the company’s scale grants it influence over how AI features get distributed across billions of users. A failure to deliver industry‑leading models — or to integrate them effectively into Facebook, Instagram and Threads — hands competitive advantage to Google, OpenAI, Anthropic and emergent Chinese players whose integrated product funnels are already proving effective.
Meta remains a formidable platform: roughly 4 billion monthly users and a 2025 revenue run‑rate in the tens of billions. Yet the AI era rewards not only compute and capital but the coupling of models with compelling consumer and enterprise products. If Avocado cannot become a new platform anchor, Meta risks ceding the terms of competition to others who are turning models into daily utility faster.
The coming weeks will test whether the company can reconfigure its engineering organisation, demonstrate financial discipline and restore confidence that its AI investments will pay strategic dividends. Absent a clear breakthrough, the combination of delayed technology, leadership questions and cost pressures makes Meta’s next chapter markedly less assured than the one Zuckerberg had envisioned.