Firmus Technologies secures $10B Blackstone loan for AI data hubs

A $10 billion debt package for an unlisted Australian AI infra startup is a glaring sign of how capital is flowing into the compute layer of the AGI race. Firmus isn’t building models; it’s building 1.6 GW of immersion‑cooled, Nvidia‑backed data centers that will power whoever can afford to rent the racks. That scale puts it in the same league as the better‑known neoclouds, and Blackstone’s involvement confirms that private credit is now a primary conduit for AI capex.([akm.ru](https://www.akm.ru/eng/news/blackstone-has-provided-10-billion-for-the-australian-ai-startup-firmus-technologies/?utm_source=openai))

Strategically, this shifts some bargaining power away from hyperscalers. If mid‑tier or sovereign buyers can source compute from specialist operators like Firmus instead of being locked into AWS, Azure or Google Cloud, they gain optionality over both price and geopolitics. At the same time, Nvidia’s fingerprints on yet another mega‑cluster underline how tightly model progress is coupled to one vendor’s silicon roadmap. The effect is to deepen the moat for large, well‑financed labs that can lock in multi‑year capacity deals while leaving startups and academic labs scrambling for leftovers.

The bigger picture: AGI timelines are increasingly constrained not just by algorithmic insights but by who controls multi‑gigawatt data center pipelines. Deals of this size harden a world where a handful of countries and capital pools own the majority of compute, and everyone else has to negotiate access on their terms.