As investors prepare for Alphabet Inc. (GOOGL) earnings, Wall Street is once again focused on AI spending, cloud growth and capital expenditures. But BlackRock Inc. (BLK) CEO Larry Fink believes the industry's biggest constraint isn't chips or models anymore—it's electricity.
Speaking this week, Fink argued that the global AI race will ultimately be decided by whoever can build enough power to support it. He pointed to China's rapid expansion of nuclear and solar generation, saying the country is positioning itself to meet the enormous electricity demands of artificial intelligence while warning that the U.S. should focus on adding power capacity rather than imposing restrictions on data center development.
Google's reported next-generation AI chip, however, suggests there may be another way to attack the problem.
Google's AI Bet Isn't Just About Faster Chips
According to a CNBC report citing The Information, Google is developing an AI chip known internally as Frozen v2, designed to permanently embed parts of its Gemini AI model directly into the silicon.
Unlike conventional AI accelerators that rely primarily on software to run increasingly large models, Frozen v2 aims to integrate portions of the model into the hardware itself, improving inference efficiency while reducing the computing resources required to perform AI tasks.
The objective isn't simply to make AI faster. It's to make AI more efficient.
That distinction matters as hyperscalers race to build ever-larger AI infrastructure.
The AI Race May Become A Power Race
China is currently building roughly 100 gigawatts of nuclear capacity and nearly 100 gigawatts of solar generation, investments Fink says are laying the foundation for the country's AI ambitions. If electricity becomes the industry's primary bottleneck, simply deploying more GPUs may no longer be enough.
That's where Google's reported chip strategy becomes particularly interesting.
Instead of solving the problem by generating more power, Google appears to be exploring how to accomplish more AI work with each watt of electricity consumed. If Frozen v2 delivers meaningful improvements in performance per watt, it could complement—not replace—the industry's massive investments in data centers and power infrastructure.
Alphabet's earnings will almost certainly focus on AI spending and cloud demand. But investors may want to listen for something else: whether the company is talking as much about AI efficiency as it is about AI scale.
If Fink is correct, the next contest in AI won't simply be over who builds the biggest models—it will be over who can power them most efficiently.
