With a declared investment of roughly $30 billion over three to five years, Blackstone is poised to become a cornerstone of the infrastructure on which next-decade artificial intelligence models will run. The plan, announced in an interview with Japanese business daily Nikkei by president and COO Jonathan Gray, involves building data centers in Japan with an aggregated capacity exceeding one gigawatt. The world’s largest alternative asset manager seems unfazed by a chorus of analysts comparing the AI race to a speculative bubble: it is betting on the compute demand expected in the coming years, and it does so in a market long sensitive to latency and data sovereignty.

Why a gigawatt is no longer science fiction

One gigawatt of capacity for AI-oriented data centers equates, for scale, to the consumption of a large industrial city. It is the power needed to feed tens of thousands of GPUs running simultaneous training and inference pipelines for ever-larger Large Language Models. Blackstone has not disclosed hardware specifics, but the investment size suggests infrastructure designed for high-density clusters, likely with NVLink systems and ultra-low-latency networks linking thousands of accelerators. For those planning on-premise LLM deployments, knowing that a single player is placing this much capacity on the table means rethinking total cost of ownership (TCO) estimates: as cloud supply expands so rapidly, the cost gap with self-hosted infrastructure could narrow for certain workloads, yet the arguments around control and data residency remain strong.

Japan as a computational sovereignty lab

The choice of Japan is no coincidence. The country enforces stringent personal data protection rules, and many local enterprises prefer to keep sensitive loads within national borders. For AI-focused data center infrastructure, this means the colocation or private cloud contracts offered by Blackstone could attract both Japanese corporations and Western hyperscalers seeking to serve Japanese clients without violating sovereignty constraints. From an AI-RADAR perspective, where the focus is on those evaluating on-premise or controlled-environment LLM stacks, the rise of local hubs with gigawatt-scale power represents a hybrid alternative: it allows avoiding global public cloud while sidestepping the full capital outlay of proprietary infrastructure.

Bubble fears and the lesson of past investments

Jonathan Gray’s interview betrays a certain impatience with alarmist narratives: Blackstone, he says, sees solid fundamentals in AI compute demand, comparable in intensity to the cloud computing explosion of the previous decade. Critics, however, point to the same excesses that led to the dot-com bust: too much capital pouring into a still-maturing technology, uncertain monetization for many end users, and the risk of a capacity glut. For teams making architecture decisions, the matter is starkly concrete: building an on-premise environment sized for today’s AI workloads today must account for the possibility that in three years the market could be flooded with cut-rate cloud capacity, eroding the initial investment’s cost advantage. On the other hand, waiting means depending on third-party availability, with potential bottlenecks if demand keeps climbing.

The energy equation and its impact on European ecosystems

Finally, developing one-gigawatt data centers raises enormous questions about electricity supply. Blackstone has not specified energy sources, but the issue is particularly acute in Japan after the nuclear downsizing. In Europe, and in Italy, any discussion about on-premise AI infrastructure must contend with energy availability and environmental constraints. For IT leaders, the equation includes not only chip and software costs but also access to a stable and, ideally, carbon-free electrical grid. Blackstone’s Japanese gambit shows that professional investors consider this hurdle surmountable – a signal that could accelerate similar projects on the European continent, refocusing attention on deployment models that balance performance, cost, and sustainability.

Those weighing whether to bring LLMs in-house or entrust them to an external operator may not yet have conclusive figures, but moves like Blackstone’s offer a window into the direction the entire sector will take over the next five years.