NVIDIA + IREN / $2.1B warrant + 5 GW DSX

Why it matters

Three product-relevant updates. (1) The $2.1B warrant + $3.4B services contract structure is the first observed case of a chip vendor taking equity-like exposure in an AI hyperscale build partner at this scale, which is structurally different from the deal templates Cliff's competitive-landscape map has been tracking. The Microsoft → Nscale Mason County WV LOI (March 16, 1.35 GW) is a *purchase* commitment by the hyperscaler. The Meta → Entergy Hyperion 7.46 GW Louisiana deal (March, expanded last week) is a *PPA* commitment by the hyperscaler with the utility building generation. The Stargate JV (now restructured to bilateral, covered last week) was a *joint venture* across OpenAI/Oracle/SoftBank/MGX. The NVIDIA → IREN deal is a *vertical integration* template — the chip vendor takes equity exposure to the build partner's success, with cloud-services revenue providing the operational glue. Cliff's competitive-landscape page on hyperscaler-DC financing should add a 'vertical-integration / chip-vendor-equity' tier. The implication for site readiness: a chip-vendor-equity-backed build is much more time-constrained than a pure-PPA build because the chip vendor is in the value chain end-to-end and can't easily pivot to alternate sites. The NVIDIA-IREN deal locks the Sweetwater site as a binding deployment, not a soft hold. (2) Sweetwater is in ERCOT — meaning the May 19-20 ERCOT TAC vote on NPRR1325 / PGRR145 (Batch Zero) is now load-bearing for the most-watched hyperscale buildout in the country. If the TAC vote falls apart or remands either item back to subcommittee, Sweetwater 2's pathway through Batch Zero gets delayed. Cliff's submission validator should treat the Sweetwater Hub as a high-priority Layer-5 reference site for the PCLR module — IREN's PCLR commitments will be a leading indicator of how the new ERCOT large-load process actually works in practice. (3) The IREN pivot from Bitcoin-mining to AI infrastructure is a template for the next 18-24 months of capacity supply. Bitcoin-mining infrastructure is purpose-built for high-density compute on cheap power with minimal interconnection ceremony — which is a structural advantage for AI-buildout when the binding constraint is energization time. Watch other miners (Marathon, Riot, Cipher Mining, Bitfarms) for similar strategic pivots; if even 30-40% of the BTC-mining capacity pivots to AI infrastructure over the next 24 months, that's tens of GW of pre-energized, pre-permitted capacity flowing into the AI supply pipeline at a much faster cadence than greenfield buildouts. Cliff's regulatory-knowledge graph needs a 'BTC-pivot site' tier with different expected timelines than greenfield hyperscale.