← Back to feed

Blackstone, the world’s largest private owner of data centers, is committing $5 billion in equity capital to a new AI infrastructure joint venture with Google. The new US-based company will run on Google’s tensor processing units — not NVIDIA GPUs.

The first 500 megawatts of compute capacity is scheduled online by 2027, with plans to “scale significantly over time.”

What This Is Really About

Google has been building its own silicon stack for years. This deal turns that internal advantage into a market-facing product: institutional capital backing compute that routes around NVIDIA entirely.

The customer profile matters. Blackstone is not a startup or a hyperscaler. It manages $1 trillion in assets and owns more data center real estate than any other private entity. A five-billion-dollar equity commitment from that player is not a pilot. It is a conviction bet on the viability of non-NVIDIA frontier compute.

The TPU Strategy

Google recently split its TPU product line into two dedicated chips: the TPU 8t for training and the TPU 8i for inference. Broadcom sealed Google’s TPU roadmap through 2031 via an 8-K filing, making the hardware commitment multi-year and contractual. The joint venture arrives at the moment Google’s silicon strategy has the most structural depth it has ever had.

Blackstone brings what Google does not have at scale: site permitting expertise, construction relationships, and a balance sheet built for large-scale real estate deployment. The combination lets Google place TPU compute into institutional-grade facilities without Google needing to own the land.

Numbers

• Equity commitment: $5 billion (Blackstone)
• First capacity tranche: 500MW online by 2027
• Hardware: Google TPUs exclusively
• Structure: New US-based joint venture company
• Google supplies chips and compute design; Blackstone supplies capital and site development

What It Tells You About 2026

NVIDIA’s position as the default AI compute substrate has never been structurally challenged at scale. Individual enterprises have deployed alternative silicon in isolated pockets. This is different: a purpose-built JV between the world’s dominant private data center operator and the only hyperscaler with a credible multi-year custom chip roadmap.

For the first time, a large private capital buyer is making a commercial-scale bet that TPU infrastructure is an investable asset class independent of NVIDIA hardware cycles. If the bet works, it validates Google’s silicon path as a funding target, not just a cost center.

Nine cloud giants are forecast to spend $830 billion on AI infrastructure in 2026. The question going into 2027 is what fraction of that capital routes to non-NVIDIA silicon. The Blackstone deal puts $5 billion and 500 megawatts in the answer column.