NVIDIA takes $2.1bn IREN stake in 5GW AI data-centre deal

NVIDIA has committed up to $2.1 billion to IREN through a strategic partnership that links the chipmaker’s reference architecture with the data-center operator’s massive 5-gigawatt infrastructure pipeline. The deal was unveiled Thursday, marking a significant expansion of NVIDIA’s investment approach in the AI cloud space.

The financial structure breaks from convention. Instead of a direct equity raise, IREN granted NVIDIA a five-year warrant for up to 30 million shares at a strike price of $70. The stock ended regular trading at $56.85 before surging roughly 9% in after-hours activity following the announcement. If NVIDIA exercises the full warrant, the cash outlay would reach $2.1 billion at the strike price, with the timing of that payment left entirely to the chipmaker’s discretion.

On the operational front, the Sweetwater campus in Texas takes priority. That site already has 2GW of planned capacity. IREN signed a $9.7 billion cloud agreement with Microsoft last year for capacity at the same location. This NVIDIA tie-up extends that blueprint, making the chipmaker both a partner and a shareholder.

What NVIDIA is essentially buying is a slot in a portfolio that has taken shape over the past 18 months. The company has taken equity positions in CoreWeave, Nebius, and several other neoclouds , firms that buy NVIDIA GPUs in bulk and lease them back to hyperscalers and frontier-model developers. Nebius’s acquisition of Eigen AI to maximize tokens per GPU was a closely watched move in the same vein. The logic is simple: AI demand outstrips hyperscaler capacity faster than those giants can build, and NVIDIA profits twice when neoclouds running its silicon fill the gap.

The numbers backing that thesis keep growing. The four largest U. S. tech companies have collectively guided to more than $700 billion in 2026 capital expenditures, much of it allocated to AI infrastructure. Even at that pace, model providers and emerging enterprise users still find themselves on waitlists for capacity. IREN’s 5GW pipeline represents a meaningful chunk of the solution, and the larger Texas grid build-out reframes that as a multi-year commitment.

IREN began as Iris Energy, a Bitcoin miner with grid-connected infrastructure spread across Texas and British Columbia. It pivoted to AI compute in 2024, dropping the Iris Energy name for the IREN ticker and a cloud-focused brand. The Microsoft deal in 2025 confirmed that pivot at scale, and this NVIDIA announcement locks in the supply side.

The shift from crypto to AI cloud is becoming common. Hut 8, Applied Digital, and Core Scientific have all redirected at least part of their capacity. What sets IREN apart at this stage is its portfolio of grid interconnects in regions with cheap, abundant power , a constraint that has become more binding than capital for new AI builds. Sweetwater is the proof point. Texas regulators have been faster to approve large interconnects than other states, and the campus sits next to renewable generation that IREN can secure through long-term contracts. A 2GW campus there rivals the scale of the largest hyperscaler builds in the region.

The neocloud structure underlying IREN’s model is straightforward: build dedicated capacity, sign long-term contracts with one or two anchor customers such as Microsoft, and finance GPU acquisition through a mix of vendor financing, equipment leasing, and equity. Meta’s $21 billion add-on with CoreWeave is the largest example of an anchor-tenant deal in this shape, and CoreWeave’s multi-year Anthropic agreement is another. The pattern is now established enough that public investors price these companies on contracted capacity rather than traditional cloud revenue multiples.

NVIDIA’s role in these structures has grown more direct over time. Instead of just selling GPUs, the company increasingly takes warrants or equity in buyers. This lets it share in the operating upside and ensures that capacity dedicated to its silicon stays priced and serviced as NVIDIA prefers. The IREN deal extends that approach into a public-company partnership rather than private financing.

The arithmetic that has to work revolves around 5GW, but the binding constraints are timing rather than capital. Substation interconnects, transformer lead times, GPU shipment schedules, and renewable power purchase agreements all operate on multi-year cycles that cannot be compressed. The Sweetwater campus, being closest to operational readiness, will absorb most of the early NVIDIA-IREN co-investment. Beyond that, the partnership functions as a framework rather than a fixed build plan, with future deployments conditional on demand and the regulatory and supply environment. This echoes Nscale’s Portugal expansion alongside Microsoft, where the build-out timeline trails the headline capacity number by years.

The risk for IREN is execution: turning grid rights and announcements into operational gigawatts at price points that anchor customers will accept. The risk for NVIDIA is concentration. The chipmaker is now equity-exposed to multiple customers in the same arc. If AI demand normalizes or shifts toward in-house silicon at hyperscalers, those positions could revalue downward. For now, the demand signal points the other way, and NVIDIA is placing its bet.