SerpApi.Org
The new Google AI chip is reportedly codenamed Frozen 2.
The chips powering the AI revolution are at the center of a lot of conversations these days, from their fantastic cost to their complex power demands and environmental concerns. They’ve even transformed Nvidia from a computer graphics brand into a trillion-dollar AI juggernaut. But there are more companies than Nvidia in the AI chip game, and Google is reportedly making a big chip move to benefit its Gemini AI system. If the exercise succeeds, Google could become a more serious rival to Nvidia.
The new chip is codenamed Frozen V2, according to tech news site The Information. It’s said to be between six and 10 times more efficient than the chips Google currently uses. This efficiency is measured in tokens per unit of electrical power burned (tokens are the way AI firms measure chunks of data going into and out of their model). More complex client prompts use more tokens to achieve an answer, and many AI makers charge for their services based on how many tokens a customer burns through.
Google will have two options if it deploys Frozen V2 chips at scale. First, it could offer the same services to clients it does now but boost profits via lower running costs. Second, it could pass on the cost savings to clients and take part in a pricing war with rival AI suppliers.
But the new Frozen chips are interesting for reasons that extend beyond their raw power. The design reportedly includes hard-coding parts of Google’s Gemini AI system directly into the chips’ circuitry. This should reduce the number of computations the chips need to perform. Traditionally, there would be a lot more digital chatter between a generic chip and an AI model running in software. Frozen could also speed up processing, because the data has less distance to move across the surface of the chip—nanosecond savings can quickly add up to faster performance. Add these two features together and you get meaningful benefits that Gemini users may be able to sense when they interact with the AIs.
Google’s efforts versus its rivals
Apple has been using this approach for years, baking AI into custom chip designs for iPhones, iPads, and Macs. Though Apple lagged behind in the AI game, now that the new Siri AI has arrived, the tech giant is poised for more success, and it already has powerful, integrated AI hardware in place.
OpenAI has also taken this approach, via a recent partnership with chip maker Broadcom, as Quartz recently noted. And Anthropic is in discussions with Samsung about a similar move, Bloomberg reported.
There are downsides to this technique. For one, if you bake the design into your chips, you’re fixing some of the ways the AI works in stone. Radical model innovations could require new chips, which may shift how future AIs are designed and delivered.
Why does all this matter to you?
Little concrete information is available right now about Google’s new Frozen chip, and it may even turn out that they’re more of an experiment than part of a large-scale rollout. In a statement, a Google Cloud spokesperson said: “Our teams are constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers. While not every project moves into production, this rigorous exploration is central to our full stack approach. By co-designing our hardware and software from the ground up, we ensure our systems are integrated and highly optimized for real-world workloads.”
So, will Google beat Nvidia at its own AI chip game? It could. It has the spending power, and it’s already built Gemini into many of its business-centric products. Meanwhile, boosting Gemini’s power while lowering its cost could drive a bigger Google-shaped wedge into the AI supplier market.
But the very fact that Google, one of the biggest tech companies in the world, is considering this kind of move should be a reminder that the AI industry is changing at an unbelievable pace. Relying on select providers to serve your long-term AI needs and fixing long-term plans in place concerning paying for AI may prove to be mistakes. If you plan to rely forever on Google’s services, that could be naive: Anthropic’s integrated AI chips could eventually prove more powerful, or they may be overtaken by OpenAI or even Apple.
Savvy leaders should take this as a cue to constantly review their AI spending plans and to continually survey the AI market in case a different supplier can serve them better.
This post originally appeared at inc.com.
“Click here to subscribe to the Inc. newsletter: inc.com/newsletters"
