Google and Blackstone Launch AI Cloud Firm With Custom Chips

Google and Blackstone are launching a new AI cloud company to provide computing capacity powered by Google’s proprietary Tensor Processing Units (TPUs) [1]. The venture, which aims to bring 500 megawatts of capacity online by 2027, will be led by Benjamin Treynor Sloss, a 22-year Google veteran who previously oversaw the company’s global data centers and server operations [1, 2].

Blackstone is providing an initial $5 billion in equity capital to the project and will serve as the majority owner [2]. When including leverage, the total investment is valued at $25 billion [2]. This partnership marks a significant expansion of Blackstone’s infrastructure portfolio, which already includes the data center operator QTS and the computing provider AirTrunk [2].

The new firm is designed to compete directly with "neocloud" providers like CoreWeave and Nebius Group, which currently rely heavily on Nvidia’s graphics processing units (GPUs) [2]. By offering Google’s TPUs as a "Compute-as-a-Service" model, the venture provides an alternative for organizations seeking to train and run AI models without relying on the industry-standard hardware dominated by Nvidia [1, 2]. Google intends to supply the software and services necessary to support the infrastructure, effectively extending its hardware reach beyond its own Google Cloud Platform [1].

This move follows a broader trend of hyperscalers attempting to reduce their reliance on external chip suppliers [1]. While Google continues to partner with chip design firms like Marvell and Broadcom to develop its custom silicon, it is simultaneously scaling its own hardware to meet the surging demand for AI compute [2, 3]. Google’s TPUs already power its internal products, including Gemini and Google Workspace, and are used by companies such as Apple, Meta, and Anthropic [1, 3].

The collaboration is the second major AI-focused business Blackstone has initiated recently; earlier this month, the firm announced a separate venture with Anthropic to help midsize companies integrate the Claude AI model into their operations [1].

The success of the new venture hinges on whether Google’s custom chips can gain significant market share against Nvidia’s entrenched GPU ecosystem. With Sloss at the helm, the company now faces the challenge of scaling its infrastructure rapidly enough to satisfy the unprecedented demand for AI-ready data centers.