Nvidia’s recent “trouble” is not about weak demand for artificial intelligence it’s about who controls the hardware powering AI’s future. The company lost nearly $250 billion in market value in a single day after reports revealed that its biggest customers, including Meta and Google, are actively working to reduce their dependence on Nvidia’s GPUs. This marks a critical turning point in the AI hardware landscape.
The Meta–Google Alliance: Breaking Nvidia’s Software Moat
For years, Nvidia’s dominance in AI chips has rested on CUDA, its proprietary software platform that developers rely on to run AI workloads. CUDA created a powerful lock-in effect, making Nvidia GPUs the default choice for AI training.
That moat is now under direct attack. Meta and Google are collaborating on TorchTPU, a project aimed at making Google’s TPUs fully compatible with PyTorch, the world’s most widely used AI framework originally created by Meta. If successful, developers can run AI models on Google’s chips as easily as on Nvidia’s hardware.
This is a game-changer. It removes the software bottleneck that forced companies to buy Nvidia’s ultra-expensive H100 and Blackwell GPUs, opening the door to cheaper alternatives at scale.
Meta’s MTIA Chip: Cutting Nvidia Out of Everyday AI
Meta is also doubling down on its MTIA (Meta Training and Inference Accelerator), an in-house AI chip designed to handle inference tasks the constant, day-to-day AI workloads behind Facebook and Instagram recommendations.
By shifting inference to MTIA, Meta can reserve Nvidia GPUs only for cutting-edge AI research. The result? Fewer Nvidia purchases over time, even as Meta’s AI usage grows. Reports also suggest Meta may acquire AI chip startups to further accelerate this strategy.
Capex Is Rising but Nvidia’s Share Isn’t Guaranteed
Meta’s capital expenditure is projected at $70–$72 billion in 2025, potentially reaching $85 billion in 2026. While Nvidia still benefits today, investors are concerned about diversification. Google has TPUs, Amazon has Trainium, and Meta has MTIA.
As big tech becomes its own supplier, Nvidia’s explosive growth rate could slow even if overall AI spending continues to surge.
The Circular Deal Problem
Nvidia has also faced scrutiny over so-called circular deals, where it invests in AI startups that then use the funds to buy Nvidia hardware. Regulators and investors are watching closely. Meta, lacking a cloud business to offset costs, is pursuing the most direct and affordable hardware path custom silicon and Google TPUs.
The Bigger Picture
Nvidia still controls roughly 90% of the AI training market, but the threat is structural, not cyclical. Zuckerberg’s goal is clear: Meta will no longer depend on a single company for the “oxygen” of the AI era. The AI hardware race is entering its most competitive phase yet.
