The NVIDIA partnership with IREN looks like a procurement deal on the surface. It isn't. It's a defensive move by NVIDIA.

NVIDIA is squeezed from two sides right now.

Supply: of the ~18 GW of US data center capacity planned to be commissioned over the past 6 quarters, only ~11 GW actually got built (Goldman). The rate of new completions has actually declined recently, mainly because power has become a major bottleneck. No new builds means no new GPU sales.

Demand: the hyperscalers' custom silicon (TPU, Trainium, Maia, MTIA) is no longer just for internal use. It's now being leased to large AI labs, with Anthropic alone signing deals worth billions for Google TPU and AWS Trainium capacity. AMD and Cerebras are also chipping away at market share. NVIDIA's dominance is being contested on multiple fronts.

This is exactly where IREN comes in. They've got the largest secured power portfolio of any neocloud at 5.8 GW and growing fast, they develop 100% of their data centers themselves, and they're not building competing silicon. That makes them the most reliable demand outlet NVIDIA can partner with at scale.

The Sweetwater partnership, positioning the 2 GW campus as a flagship DSX deployment, isn't NVIDIA doing IREN a favor. It's NVIDIA solving its two biggest problems at once.

If you want the longer version of this take, check out my last note: