Blackstone Plans Public Company for AI Data Center Spree
Private equity giant Blackstone is planning to launch a public company for an "aggressive buying spree" of AI data centers. The move signals that major capital markets see a sustained, multi-year demand for "AI factories" and are financing large-scale infrastructure beyond traditional hyperscaler budgets.
This new public company is being formed to give retail investors a way to invest in the AI infrastructure boom. Blackstone is reportedly seeking initial capital from sovereign wealth funds and other institutions, with an eventual goal of raising tens of billions of dollars. The move is part of Blackstone's broader strategy to become a dominant investor in AI infrastructure. Blackstone's real estate and infrastructure funds have already committed to a massive $25 billion investment in digital and energy infrastructure in Pennsylvania alone, with plans to catalyze an additional $60 billion. This investment will fund the development of new data center campuses by QTS, a Blackstone-owned company and one of the world's largest independent data center operators. As of late 2025, Blackstone's data center portfolio already included $70 billion in existing assets with another $100 billion in the pipeline. The demand for AI-specific data centers is surging, with the market expected to grow from around $49 billion in 2026 to over $152 billion by 2031. This growth is driven by the need for high-density infrastructure capable of supporting AI workloads, which require significantly more power. An AI-focused hyperscale data center can consume as much electricity as 100,000 homes or more. This insatiable power demand is a direct consequence of next-generation AI hardware. NVIDIA's upcoming Blackwell GPUs, for instance, are expected to consume up to 1 kilowatt per chip, a 40% increase from the previous Hopper architecture. This jump in power density is forcing a shift to liquid cooling, as traditional air cooling becomes ineffective for high-density deployments. A single rack of Blackwell GPUs could require 100 kW or more, a significant increase from the 30-40 kW per cabinet in most current servers. In response to the high cost and power demands of merchant silicon, hyperscalers like Google, Amazon, and Meta are increasingly developing their own custom AI accelerators. This "silicon sovereignty" allows them to optimize performance and cost for their specific workloads, bypassing the so-called "NVIDIA tax." Google's TPU v7, for example, is engineered to handle massive models with greater energy efficiency, while Amazon's Trainium 3 aims to offer a better price-performance ratio for both training and inference. This trend of vertical integration is fundamentally reshaping the competitive landscape for AI compute.
