Stop everything. Meta Platforms is talking to Google. It’s a deal for chips. We’re talking billions of dollars worth of Google’s chips—their Tensor Processing Units, or TPUs—for use in Meta’s own data centers starting in 2027.
This is a massive move. It instantly casts Google as a deadly serious rival to Nvidia.
The talks aren’t just for 2027, either. They also involve Meta renting chips from Google Cloud as early as next year. This is part of Google’s new push. Historically, Google only used TPUs in their own data centers. Now? They want customers to adopt those AI-specific TPUs in their own facilities.
The Battle for the AI Chip Market
This is a complete departure from Google’s old strategy. And then the market expansion follows—it puts Google right into the middle of the hundreds of billions of dollars being spent on AI data-center processors.
Some Google Cloud executives are openly suggesting this could let them capture as much as 10% of Nvidia’s annual revenue. That slice? It’s worth billions. And it’s not hard to see why companies are looking for alternatives. Nvidia’s GPUs are pricey, and the supply is often constrained. Custom chips like Google’s TPUs are suddenly in high demand. Anthropic already expanded its Google deal to use up to one million of those AI chips, that too.
Stock Shockwaves
The market reaction was immediate and messy.
Alphabet (Google’s parent): Shares jumped over 3% in premarket trading. They were suddenly within striking distance of a $4 trillion valuation.
Broadcom: The company that helps Google build these TPUs? They gained 2%.
Nvidia: Shares fell 3.2%. A single report taking a chunk out of the market leader. That’s how high the stakes are.
Clinching a deal with Meta—who plans to spend up to $72 billion this year—would be a major coup for Google. They are already winning the generative AI boom through cloud services demand.
But here’s the complication. Here’s the obstacle. Taking on Nvidia’s dominance requires overcoming nearly two decades of proprietary code. Nvidia has that lock-in with its CUDA software platform. Over 4 million developers worldwide rely on it. Dislodging that entire software ecosystem? That’s the real war. It’s an ongoing challenge.
End….
