According to Financial Times News, in a turbulent 2025, Mark Zuckerberg made an all-consuming bet to turn Meta from an AI laggard into a leader, poaching top researchers with stratospheric salaries and even homemade soup. He’s pumping billions into infrastructure, with capital expenditures expected to hit at least $70bn in 2025 and potentially top $100bn next year, a plan that wiped over $208bn from Meta’s valuation after a vague October earnings call. To compete, Meta’s new elite lab aims to release a brand-new AI model codenamed “Avocado” in Q1 2026, targeting performance on par with Google’s acclaimed Gemini models. This follows the underperformance of its Llama 4 model earlier in the year, a blow to Zuckerberg’s ambitions. Internally, the push has caused corporate whiplash, with 600 AI team layoffs, executive reshuffles, and tensions with new hires like 28-year-old billionaire Alexandr Wang, who now heads the renamed Meta Superintelligence Lab. Zuckerberg’s overarching goal is now “personal superintelligence”—AI that surpasses human intelligence and fits in your pocket.
The Avocado Hail Mary
So, “Avocado.” It’s not just another Llama iteration. They’re building it from scratch, which is a massive, expensive admission that their current architecture might be a dead end. The target is to match Google’s Gemini 3 by next summer. That’s an insanely aggressive timeline, considering Gemini 3 is currently seen as leapfrogging OpenAI’s ChatGPT. Basically, Meta isn’t just trying to catch up; they’re trying to skip a generation in one go. The pressure is immense. If Avocado flops like Llama 4 did, the “Great Talent Heist” Zuckerberg orchestrated—with those huge $100mn pay packages—will reverse itself. Talent in AI is incredibly mobile, and they’ll flee a sinking ship for the next hot lab. This is a bet-the-company move on a single model release.
Culture Clash and the Eye of Sauron
Here’s the thing about Meta: it’s a “tenure-driven” place, built on “Friends of Zuck” (FoZ). Now, Zuckerberg is trying to force a cultural revolution by importing expensive outsiders like Alexandr Wang and Nat Friedman. It’s creating chaos. Wang reportedly finds Zuckerberg’s infamous micromanagement—the “Eye of Sauron”—suffocating. And internally, some question if a data-labeling startup founder is the right person to lead frontier AI research at a giant corporation. Meanwhile, old guards are leaving. You’ve got this classic Silicon Valley clash: the founder’s intense, personal vision versus the operational expertise needed to execute at scale. The layoffs of 600 AI staff, framed as increasing “agility,” just add to the instability. Can a company known for moving slow and breaking things actually move fast and build something coherent?
The Vision: Pockets and Friends
Zuckerberg’s pivot to “personal superintelligence” for “pockets” is fascinating. After the metaverse flop, he’s anchoring AI in a very human, almost intimate problem: loneliness. He cites research that Americans want more friends, and his answer is an AI companion. It’s a clever reframe from dry productivity tools to relationships and creativity. The long-term play is to embed this AI into AR smart glasses, aiming to replace the iPhone as your primary computer. But think about that for a second. The hardware challenges alone are staggering. You need immense computing power in a glasses form factor, which is a brutal engineering problem. It’s a vision that requires dominating in AI *and* hardware simultaneously. Apple and Google should be worried, but only if Meta can actually execute.
A High-Stakes Poker Game
One former exec nailed it: Zuckerberg likes high-stakes poker with his company. He’s buying momentum by deploying massive resources—cash, talent, political capital (hence courting Trump to avoid regulatory friction). The question isn’t whether he’ll be a player; with $100bn, he will be. The question is whether he can “win.” Can he force a sprawling, politically fractured organization to align behind a single, wildly ambitious technical goal? Or will it remain a collection of fragmented teams “rooting for their own products”? The 2026 Avocado release isn’t just a product launch; it’s a referendum on whether Zuckerberg’s hands-on, chaotic, spend-whatever-it-takes approach can work in the AI era. I’m wary of betting against his resources, but betting on his execution right now feels like a gamble.
