Meta Is Ditching Nvidia for Google's AI Chips

Image on Unsplash

Meta is in talks to use Google's custom-designed AI chips, known as tensor processing units or TPUs, in its data centers by 2027. They may also rent TPUs from Google's cloud division as early as next year.

Till now, Nvidia has dominated the AI chip market. Almost every major tech company building AI infrastructure has relied on Nvidia's GPUs. This year, Meta alone is spending between 70 and 72 billion dollars on AI infrastructure.

What changed is that companies are actively seeking alternatives to reduce their dependence on a single supplier. Google's TPUs are customized specifically for AI tasks, which gives them an efficiency advantage over Nvidia's more general-purpose GPUs.

Once the tech companies understood this, the market changed immediately. Nvidia's stock fell 3.6% in premarket trading on Tuesday. Google's parent company Alphabet rose by 3%. Broadcom, which partners with Google to design the TPUs, also gained over 2%.

This does not mean Nvidia is losing its dominance overnight. The company still controls the vast majority of the AI chip market. But it signals that the era of total reliance on a single chip supplier is ending.

Companies building AI infrastructure now want optionality. They want multiple vendors competing, multiple designs available, and multiple supply chains as well.

For Meta specifically, this also means the company can negotiate harder with Nvidia by having a credible alternative.