AI Compute Infrastructure
All research in AI Compute Infrastructure — 2 reports.
An Australian sovereign AI/GPU cloud infrastructure neocloud whose 2025 revenue was just 1.57 million dollars and free cash flow roughly -13.59 million dollars, with a valuation propped up by an order narrative of about 2.2 billion dollars in contracts and continual fundraising. Internal controls carry a material weakness, and there is essentially no margin of safety. Rating Watch: a hot AI infrastructure story that has yet to prove cash-flow quality and already carries high-expectation, high-dilution risk.
The first landing point for AI CapEx is the "high-value compute unit"—GPUs/ASICs, HBM, advanced packaging, and leading-edge wafers—before the spend diffuses outward to rack-scale servers, networking, and liquid cooling and power distribution. The profit pool concentrates in the high-barrier component layer (NVDA / TSMC / SK hynix / Broadcom / Micron), while ODMs capture revenue on thin margins. Rating Watch: a sector where the durable profit pool and the real bottlenecks sit upstream, not in white-box data center floor space.