Australian AI infrastructure firm wins $10B financing to accelerate data‑center buildout

Firmus Technologies, an Australian developer focusing on infrastructure for artificial intelligence workloads, has secured a $10 billion private‑credit facility arranged by a consortium that includes Blackstone‑led funds and Coatue. The financing is earmarked for construction of multi‑campus sites and the procurement and installation of high‑performance accelerators from a leading chip supplier, with aggregate IT power capability targeted at roughly 1.6 gigawatts by 2028. The capital is structured as credit rather than equity, creating fixed repayment obligations that will shape deployment pacing, commercial terms offered to customers and contingency planning for cost overruns. The deal reflects a broader shift in which private credit and institutional lenders are underwriting purpose‑built AI capacity—alongside other structures such as leases of accelerator fleets and equity infusions seen in parallel market moves. Practically, the build program will intensify near‑term demand for GPUs, power‑distribution systems and advanced cooling, tying Firmus’ timetable to global supply chains and chip allocation cycles. Execution risks are concentrated in three areas: grid readiness and transmission upgrades, local permitting and community acceptance, and timing of hardware deliveries; each can materially affect when revenue can begin to service the facility’s debt. Lenders and sponsors will also need to contend with customer‑concentration and utilization risk: while hyperscalers and large AI developers are the most likely anchor tenants, market-wide trends show demand often bursts around training cycles and can lag front‑loaded construction. The financing therefore joins a series of market responses—ranging from accelerator leasing vehicles to strategic equity placements by chipmakers—that seek to match capital to the highly specific economics of GPU‑dense facilities. If Firmus can align grid upgrades, secure multi‑year commitments from tenants and avoid material supply bottlenecks, the assets could establish locally accessible, low‑latency capacity for large‑model training and inference and attract cloud and enterprise customers. Conversely, permitting delays, strained interconnection timelines or prolonged underutilization would pressure the sponsor’s leverage profile and test lender appetite for similar private‑credit structures. The transaction is thus both a signal of investor conviction in regional AI compute demand and a case study in how debt markets are adapting to power‑intensive, accelerator‑centric infrastructure.