NVIDIA acquires a $2.1 billion warrant in IREN as part of a 5GW AI data center agreement.

      NVIDIA plans to invest up to $2.1 billion in IREN as part of a partnership that combines the chipmaker’s reference architecture with the data-centre operator’s 5-gigawatt infrastructure pipeline. This collaboration was announced on Thursday.

      The arrangement is unconventional; instead of a typical equity investment, IREN has granted Nvidia a five-year warrant for up to 30 million shares at an exercise price of $70. Nvidia's stock closed at $56.85 during regular trading and then increased by approximately 9% in after-hours trading following the announcement.

      If Nvidia fully exercises the warrant, the total investment would reach $2.1 billion at the exercise price, with cash outflows occurring at the chipmaker's discretion rather than at the closing of the deal.

      The initial focus will be on the Sweetwater campus in Texas, which currently has a planned capacity of 2GW. IREN previously entered into a $9.7 billion cloud agreement with Microsoft last year for capacity at that site; this partnership with Nvidia extends the same strategy by involving the chipmaker as both a partner and a shareholder.

      Nvidia is placing IREN into a category that has become familiar within its portfolio over the last eighteen months. The chipmaker has invested in CoreWeave, Nebius, and other so-called “neoclouds” that purchase Nvidia GPUs in large quantities and rent them to hyperscalers and developers of advanced models.

      Nebius's acquisition of Eigen AI to optimize tokens per GPU was one of the most closely observed strategies in this context. The reasoning is straightforward: the demand for AI outpaces hyperscaler capacity more quickly than they can expand, allowing Nvidia to benefit both from the demand and the operations of the neoclouds utilizing its technology.

      The statistics backing this reasoning have consistently increased each quarter. The four largest tech companies in the U.S. have projected over $700 billion in capital expenditures by 2026, with much of this planned for AI infrastructure.

      Despite this rapid growth, model providers and new enterprise users continue to find themselves on waiting lists for capacity. IREN’s 5GW represents a significant part of the solution; the larger build-out in Texas recontextualizes this as a multi-year commitment.

      IREN originally operated as Iris Energy, a Bitcoin mining company with grid-connected infrastructure across Texas and British Columbia. In 2024, it shifted focus to AI computing, rebranding itself as IREN. The agreement with Microsoft in 2025 confirmed this transition on a larger scale, and the current Nvidia announcement bolsters the supply side.

      The shift from cryptocurrency to AI cloud is becoming increasingly common, with companies like Hut 8, Applied Digital, and Core Scientific all reallocating parts of their capacity toward AI.

      What distinguishes IREN at this moment is its network of grid interconnections in regions with abundant, low-cost power, a critical factor that has emerged as more restrictive than capital for new AI developments.

      The Sweetwater campus serves as the prototype. Texas regulators have been quicker to approve large interconnections than those in other states, and the campus benefits from proximity to renewable energy resources that IREN can secure through long-term contracts. A 2GW campus matches the scale of the largest hyperscaler projects in the region.

      The designation of IREN as a neocloud reflects a specific business model: constructing dedicated capacity, forming long-term agreements with one or two primary customers (in this case, Microsoft), and financing GPU acquisitions through various methods, including vendor financing, equipment leases, and equity.

      Meta's $21 billion partnership with CoreWeave exemplifies the largest type of anchor-tenant deal, while CoreWeave’s multi-year agreement with Anthropic represents another example. This pattern has become sufficiently recognized that public investors evaluate these companies based on contracted capacity rather than traditional cloud revenue multiples.

      NVIDIA's involvement in these structures has evolved to become more direct over time. Rather than solely selling GPUs, the company now often takes warrants or equity in its buyers to participate in operating gains and ensure that capacity committed to its products is priced and serviced according to its preferences.

      The IREN partnership represents an extension of this approach into a public-company collaboration rather than a private financing arrangement.

      While 5GW is a significant figure, the crucial constraints are related to timing rather than financial resources. Interconnection at substations, lead times for transformers, GPU shipment schedules, and renewable power purchase agreements are all subject to multi-year cycles that cannot be expedited.

      The Sweetwater campus, being the closest to operational readiness, is where most of the initial co-investment between Nvidia and IREN will be allocated.

      In the future, the partnership will serve as a framework rather than a definitive build-out plan, with projected deployments contingent upon demand and the regulatory and supply context.

      This pattern resembles Nscale’s expansion in Portugal with Microsoft, where the construction timeline lags significantly behind the announced capacity.