Google continues to be one of the best names to own in AI. And it has to be on top of every portfolio.

Right now Google’s Gemini 3 is getting a lot of attention right now. And a lot of people think that it puts Nvidia at risk. But that’s not entirely true.

Gemini 3 is an incredible model. I’ve tested it thoroughly over the last few days. And it shows that scaling laws still hold. But that actually supports Nvidia’s position. Bigger models lead to higher compute needs. Higher compute needs increase demand for Nvidia hardware, not the opposite.

Google’s TPUs are impressive. It can optimize TPUs so well because they only need to serve one customer, which is Google itself. That gives them tight control and strong performance. But it also limits the value for anyone outside Google. Enterprises, startups, and labs need hardware that runs any model, any framework, and any workflow with minimal friction.

There is no CUDA support, no broad developer community, no open supply chain, and no global ecosystem around them. They are not built to ship at scale to data centers worldwide. Maybe that changes later, but not today.

And timing matters. Data centers are planned 2 to 3 years ahead. They need hardware partners that can deliver at scale to order from today. Nvidia owns this market. Data centers are designed with Nvidia chips in mind.

Building custom chips is a long and complex process. It takes years of internal capability building, vendor relationships, system design, and optimization. Google has been working on TPUs for more than a decade and is now on generation 7 (Ironwood). Most companies are far behind.

Below are the key suppliers behind Google’s TPU. Many of them have been highlighted here often. And they all in AI infrastructure basket.

Google actually helps Nvidia by keeping TPUs private. Since Google does not sell TPUs, Nvidia faces almost no price pressure. GPU prices stay high. The global price of inference stays high. And Google uses this to its advantage. They run TPU inference inside GCP and capture that same premium. Nvidia sets the price anchor for compute. Google collects cloud margins on top. They never even have to compete head on.

Both have a bright future. Google can build TPUs for Google. Everyone else depends on Nvidia.