Blackstone bets $5 billion on Google's secret AI chip to rival Nvidia

Blackstone is backing Google’s artificial intelligence ambitions with a major investment in a new data center venture built around Google’s in-house AI chips.

The alternative-asset giant is committing an initial USD 5 billion (approx. $7.3 billion) in equity to a new US-based AI cloud venture with Google, with plans to bring 500 megawatts of data center capacity online by 2027.

Including leverage, the investment could eventually reach about USD 25 billion (approx. $36.4 billion), a Bloomberg report said.

The bet seems enormous as AI compute demand explodes, Google wants its Tensor Processing Units, or TPUs, to become a real alternative to Nvidia’s dominant GPUs.

Google’s Nvidia challenge

The new company is designed around Google’s TPUs, the custom chips it built for artificial intelligence workloads.

That makes the venture more focused than a conventional hyperscale data center project.

It is not simply about adding server capacity, but also about creating a third-party platform where customers can rent access to Google’s AI hardware, software and infrastructure as a service.

That puts the deal closer to the “neocloud” model made popular by companies such as CoreWeave, which built its business around providing AI developers with access to Nvidia-powered compute.

The difference lies in the chips: CoreWeave built its business around Nvidia’s ecosystem, while Google and Blackstone are now trying to create a similar marketplace centered on Google’s TPUs.

For Blackstone, the appeal is obvious as AI infrastructure has become one of the largest capital-deployment opportunities in the world.

Data centers require large amounts of land, power, cooling, fiber connectivity and financing, making infrastructure-focused investors such as Blackstone natural participants in the sector.

Blackstone President Jon Gray framed the deal as a “generational opportunity” to invest in AI infrastructure, saying the new company can help meet unprecedented demand for compute.

Google’s chip bet gets a new route to market

Google’s TPUs are not new, as the company has spent more than a decade developing and using them internally to train and run AI models.

They help power Google’s own products, including Gemini, and are built specifically for AI training and inference rather than general-purpose computing.

What is new is the distribution model.

Historically, Google’s TPUs have been closely tied to Google Cloud. This venture gives them a wider route to customers, backed by Blackstone’s capital and data center development capabilities.

In simple words, Google is trying to make its AI chips easier for outside companies to access without forcing every customer relationship to fit neatly inside the traditional Google Cloud box.

The leadership choice also sends a signal as the venture will be run by Benjamin Treynor Sloss, a longtime Google executive with deep experience in infrastructure and operations.

For Google Cloud, the prize is strategic as Nvidia has become the default supplier for much of the AI boom.

But cloud providers increasingly want more control over the hardware stack, both to reduce costs and to differentiate their platforms.