Blackstone takes the majority position in Google’s new TPU cloud

At a glance:

• Blackstone and Google have formed a $25 billion joint venture to build a U.S.-based AI compute-as-a-service business powered by Google's tensor processing units (TPUs), with 500 MW of data center capacity targeted for 2027.
• The venture positions Google as a direct competitor to NVIDIA-aligned CoreWeave in the AI cloud market, leveraging its decade-refined TPU architecture and Blackstone's infrastructure financing.
• Potential customers include other AI labs, enterprises seeking TPU alternatives for inference workloads, and sovereign entities affected by NVIDIA export controls.

The Strategic Partnership

Blackstone and Google announced a joint venture on Monday to construct a U.S.-based AI compute-as-a-service platform built on Google's tensor processing units. Under the deal, Blackstone will contribute $5 billion in initial equity and assume majority ownership, with the total enterprise value reaching approximately $25 billion, including leverage. The first phase aims to deploy 500 MW of data center capacity by 2027, marking a significant expansion in AI infrastructure.

This partnership combines Google's advanced TPU technology with Blackstone's expertise in large-scale infrastructure financing. The joint venture will operate as a separate entity, allowing Google to scale its external TPU capacity while Blackstone manages the capital-intensive data center build-out. The structural focus on TPUs—rather than general-purpose GPUs—highlights Google's bet on its custom silicon as a differentiated offering in the competitive AI cloud landscape.

Competitive Dynamics with CoreWeave

The venture is explicitly framed as Google's response to CoreWeave, the NVIDIA-aligned neocloud provider that went public in 2024. CoreWeave has become a central player in the AI-compute-as-a-service trade, relying on NVIDIA's GPUs to serve enterprise and model lab customers. By contrast, Google's TPU cloud leverages its own chip architecture, which has been refined over a decade, offering an alternative to NVIDIA-dominated ecosystems.

The competitive read suggests that a Google-Blackstone TPU cloud at 500 MW capacity represents the most credible challenge yet to CoreWeave's model. Not only does it provide differentiated silicon, but Blackstone's financing capacity—effectively unlimited within infrastructure mandates—enables rapid scaling without the balance-sheet constraints faced by pure-play tech companies. This could shift the supply-demand dynamics in the AI cloud market, which has been characterized by undersupply and long waitlists.

Financial Engineering and Off-Balance-Sheet Structure

The deal follows a recognizable Blackstone infrastructure-fund template. The $5 billion equity contribution sits within a $25 billion total project value, implying $20 billion in debt financing secured against the underlying data-center and equipment assets. This structure allows Google to move the infrastructure-financing burden off its balance sheet into a Blackstone-controlled vehicle while retaining the TPU supply and architecture relationship that generates underlying margins.

Blackstone's existing infrastructure portfolio includes the QTS Realty Trust data-center platform acquired in 2021, demonstrating its capability to execute such large-scale projects. For Google, this off-balance-sheet approach mitigates capital expenditure risks while accelerating capacity deployment. It also aligns with broader trends in Big Tech, where companies are increasingly partnering with infrastructure investors to fund AI build-outs without straining their own financials.

Google's External Commitments and Internal Challenges

Google has been rapidly expanding its external TPU capacity, highlighted by a $40 billion-plus deal with Anthropic announced in October, which includes five gigawatts of TPU capacity over five years and access to up to one million seventh-generation Ironwood chips. Meta also signed a TPU deal earlier this year. However, this external push has created a well-documented internal compute problem, where Google's own AI researchers—including teams at DeepMind—now queue for the same fabric sold to external partners.

The Blackstone JV serves as the third leg of Google's external-distribution strategy, complementing its direct sales and large-scale contracts. By creating a dedicated infrastructure vehicle, Google can better manage the tension between serving internal needs and capitalizing on external demand. This separation may help streamline resource allocation and prevent internal teams from being bottlenecked by commercial commitments.

Market Context and Capital Cycles

The scale of the deal is underscored by the broader AI infrastructure spending boom. Big Tech AI-infrastructure capex is projected to clear $700 billion this year, with Google itself running between $175 billion and $185 billion. Against this backdrop, a $25 billion joint venture is relatively small for Google but strategically valuable for its structural benefits. It allows Google to leverage Blackstone's financing to move quickly without diluting its own balance sheet.

This capital-cycle context explains why such partnerships are becoming more common. As AI infrastructure demands soar, tech companies are seeking innovative financing models to avoid over-leveraging their own assets. For Blackstone, the JV represents a long-tenor cash-flow vehicle that fits its growing focus on the AI infrastructure category, which has scaled dramatically in recent years.

Implications for the Neocloud Ecosystem

For CoreWeave and the wider NVIDIA Neocloud category, the Google-Blackstone partnership poses a sharp competitive threat. CoreWeave's stock has traded on assumptions of structural undersupply and a multi-year head start due to its NVIDIA-allocation relationships. However, a 500 MW TPU cloud from Google—with differentiated silicon and unlimited financing—could erode CoreWeave's market position, especially if it attracts enterprise customers seeking alternatives to NVIDIA GPUs.

The customer mix will be crucial in determining the JV's success. The most likely buyers fall into three buckets: foundation-model labs other than Anthropic or Meta wanting TPU capacity under long-term contracts; enterprises currently buying GPU compute through CoreWeave or Lambda who desire a TPU option for inference-heavy workloads; and sovereign-AI buyers in regions where NVIDIA export controls have created procurement gaps. Which of these segments the JV optimizes for will dictate whether it stops at 500 MW or expands further.

Operational Road Ahead

Operational specifics, including data-center sites and TPU generations, are still under development. The next visible proof point will be the JV's first named anchor customer, which could signal its strategic focus. Industry watchers will also monitor how quickly the 500 MW target is achieved and whether additional phases are planned. The venture's success hinges on execution, customer acquisition, and the continued performance advantages of Google's TPU architecture against NVIDIA's dominant GPUs.

In the evolving AI cloud landscape, the Google-Blackstone partnership marks a significant shift toward diversified silicon options and infrastructure financing models. As competition intensifies, customers may benefit from increased choice and potentially lower costs, while the industry grapples with the geopolitical and supply-chain implications of reduced reliance on NVIDIA.