Meta’s Billion-Dollar AI Strategy Revamp Sparks Cultural Clash

Meta’s chief executive Mark Zuckerberg was bullish last year about the company’s Llama family of AI models, predicting they would become “the most advanced in the industry” and “bring the benefits of AI to everyone.”

During Meta’s earnings call in January, the Llama brand was highlighted repeatedly. By the October call, however, the name was mentioned only once. The shift reflects a broader strategic pivot: Meta is now betting on a multibillion‑dollar hiring spree to acquire top AI talent capable of challenging OpenAI, Google, and Anthropic.

Insiders say the company’s AI roadmap has become fragmented, reinforcing the perception that Meta is lagging behind rivals whose models are rapidly gaining traction in both consumer and enterprise markets.

Meta is developing a next‑generation model, internally codenamed “Avocado,” intended to supersede the Llama line. Sources familiar with the project expected a 2025 launch, but the timeline has slipped to the first quarter of 2026 while the model undergoes extensive performance testing. A Meta spokesperson confirmed that training efforts remain on schedule.

Wall Street is watching closely. After a $14.3 billion acquisition of Scale AI’s founder Alexandr Wang and his engineering team, Meta raised its 2025 capital‑expenditure outlook to $70‑$72 billion, up from the prior $66‑$72 billion range. Analysts at KeyBanc Capital Markets note that Meta, once viewed as an AI winner compared with Alphabet, now faces heightened scrutiny over investment levels and return on capital.

Meta’s core revenue driver—digital advertising—remains robust, delivering more than $160 billion in annual sales with a year‑over‑year growth rate north of 20%. AI‑powered targeting and Instagram’s popularity have kept the ad business strong, and investors have praised the company for using AI to improve efficiency.

Nevertheless, Zuckerberg’s ambition extends far beyond ads. The new AI leadership, recruited from outside Meta, lacks a background in the company’s advertising engine, underscoring a strategic shift toward a broader AI vision.

Historically, Meta distinguished itself by open‑sourcing its Llama models, allowing researchers worldwide to build on the technology. In July 2024, Zuckerberg acknowledged that while most competitors were developing closed models, open source was “quickly closing the gap.” By the summer, after a lukewarm reception to Llama 4, he signaled a possible retreat from the open‑source posture, emphasizing risk mitigation and selective sharing.

Industry insiders suggest that Avocado will likely be a proprietary model, meaning its weights and code will not be freely downloadable. This marks a stark departure from Meta’s earlier philosophy and aligns with the company’s recent internal reorganization.

The Llama 4 rollout also triggered a leadership shuffle. Chris Cox, Meta’s chief product officer, stepped away from overseeing the GenAI unit following the model’s disappointing launch. In parallel, Meta hired Alexandr Wang as chief AI officer and placed him at the helm of the newly formed TBD Lab, where Avocado is being built.

Wang now shoulders the responsibility of delivering a flagship model that can close the gap with OpenAI’s GPT‑5, Google’s Gemini 3, and Anthropic’s Claude Opus 4.5. Each of those rivals has announced upgrades this year, intensifying the race for performance, safety, and developer adoption.

Analysts stress that no single model dominates the market; strengths vary across conversational fluency, coding assistance, and specialized workloads. Sustained leadership, however, requires massive and continuous capital outlays—funds that flow directly to GPU suppliers.

Nvidia, the premier provider of AI accelerators, reported a 62 % year‑over‑year revenue surge, citing major model developers—including OpenAI, Anthropic, and Google—as key customers. While Nvidia did not name Meta’s upcoming model, it highlighted Meta’s “Gem” foundation model, which powers ad‑recommendation systems and contributed to higher conversion rates in Q2.

Meta’s internal AI teams face intense pressure. Friedman, who leads the product and applied‑research arm of MSL, was tasked with delivering a breakthrough AI product. The September launch of “Vibes,” an AI‑generated short‑video feed, fell short of expectations when compared with OpenAI’s Sora 2, lacking advanced features such as realistic lip‑synced audio.

Employee morale has been strained by demanding work schedules and a series of layoffs. In October, Meta cut 600 positions within MSL, affecting the Fundamental Artificial Intelligence Research unit and contributing to chief AI scientist Yann LeCun’s decision to leave for a startup.

The influx of outside talent represents a cultural shift for a company that has traditionally promoted long‑tenured engineers to senior roles. Wang and Friedman bring a “foundry‑style” mindset focused on infrastructure, hardware efficiency, and rapid prototyping, contrasting with Meta’s historically consensus‑driven product development process.

Inside TBD Lab, communications are deliberately siloed. Team members rarely use the internal “Workplace” platform, operating more like an independent startup. Nonetheless, oversight remains: engineering VP Aparna Ramani manages computing‑resource allocation across MSL, while Vishal Shah, formerly head of Reality Labs, now serves as vice‑president of AI Products, bridging legacy social‑app expertise with the new AI agenda.

Meta is also reshaping its infrastructure strategy, increasingly relying on third‑party cloud providers such as CoreWeave and Oracle for AI training and testing, while simultaneously investing in its own hyperscale data centers. In October, Meta announced a joint‑venture with Blue Owl Capital to fund the $27 billion Hyperion data‑center complex in Louisiana, a facility designed to deliver the “speed and flexibility” required for next‑generation AI workloads.

These moves underscore a broader industry trend: leading tech firms are constructing private AI clouds to retain control over data, reduce latency, and avoid dependence on external providers. For Meta, this infrastructure investment is critical to supporting both its advertising stack and the anticipated rollout of Avocado.

In an earnings call in October, Zuckerberg reaffirmed his commitment to AI, stating that the company’s AI lab now has “the highest talent density in the industry.” He promised that Meta’s next‑generation models and products would be unveiled in the coming months, signaling an aggressive timeline despite recent setbacks.

Investors will be watching three key metrics closely: the performance of Avocado against rival models, the effectiveness of Meta’s new data‑center investments in reducing training costs, and the ability of the revamped AI organization to translate breakthroughs into revenue‑generating products—particularly within the advertising ecosystem that remains Meta’s financial engine.