Nvidia will invest up to $2.1 billion in IREN as part of a broader 5-gigawatt infrastructure deal between the chip giant and the listed neocloud, the two companies said on Thursday. The investment takes the form of warrants to purchase up to 30 million IREN shares at a $70 strike price, against IREN's regular-session close of $56.85.
IREN began life as a bitcoin miner. The Australia-listed company traded as Iris Energy before rebranding and pivoting to AI infrastructure, repurposing power-rich sites and grid-connected campuses originally built for crypto mining.
IREN shares jumped roughly 9% in extended trading on the announcement. The five-year warrant package is structured so that Nvidia is effectively paid in IREN equity upside if the neocloud's stock rises by more than 23% from current levels, with the chip company taking a meaningful position in a customer that buys substantial volumes of its hardware. Future deployments under the deal are expected to focus on IREN's 2-gigawatt Sweetwater campus in Texas.
The two companies said the partnership is intended to accelerate the deployment of large-scale AI factories by combining Nvidia's reference architecture with IREN's data centre operations. IREN signed a $9.7 billion cloud capacity agreement with Microsoft last year. Thursday's deal layers a direct Nvidia equity exposure on top of that contract base.
What a "neocloud" is, and why Nvidia is investing in one
IREN is part of a class of operators that have come to be called neoclouds: companies that build and lease GPU compute capacity, almost always running on Nvidia accelerators, to customers that prefer not to or cannot wait to build their own data centres. CoreWeave, Lambda Labs, Crusoe, and IREN all sit in roughly the same category, with different mixes of legacy crypto-mining heritage, hyperscaler contracts, and direct frontier-lab capacity deals.
The neocloud category exists because hyperscaler queues for AI capacity have lengthened, frontier model developers want to diversify supply, and the underlying chips are still allocated by Nvidia rather than purchased on a market. A direct Nvidia equity stake in a neocloud is partly a customer-allocation signal, partly a financial bet on operating leverage, and partly an additional anchor for the chip company's broader build-out narrative. Nvidia has done variants of the same trade with several neocloud and AI-lab counterparties over the past two years.
The pattern has drawn analyst attention because the cash flows are at least partially circular. Nvidia commits capital to a customer, the customer uses that financial capacity to commit to Nvidia GPU purchases, and the GPU purchases show up in Nvidia revenue. The accounting is unambiguous in each individual instance, but the aggregate effect, across enough such deals, is to make a meaningful share of frontier AI capex look more like vendor-financed pre-commitments than independent demand.
The cloud Nvidia is quietly assembling
The deal also fits a structural pattern that has gone less remarked on than it deserves. Nvidia has spent the past two years taking equity positions in or near most of the major neocloud operators (CoreWeave, IREN now, and a handful of smaller capacity providers), alongside its own DGX Cloud product, which the company launched in 2023 as a managed AI offering running on hyperscaler infrastructure.
Read together, the moves look less like opportunistic financial bets and more like the ground floor of an Nvidia-orchestrated GPU cloud. The eventual product surface would be a single interface customers submit AI workloads to, with Nvidia routing the jobs across a federation of capacity sources: its own DGX Cloud installations, IREN's Sweetwater build, CoreWeave's footprint, and smaller partners. Customers would see Nvidia as the cloud. Behind the scenes, Nvidia would be the orchestrator of a network of operators it has already part-funded.
That product, if it materialises, would compete directly with AWS Bedrock, Azure's OpenAI and Foundry stack, and Google Cloud's Vertex AI, the same three hyperscalers that today are Nvidia's largest customers. The channel conflict is real, and Nvidia would have to position any such offering carefully relative to its biggest revenue lines. The structural advantage in the trade, though, is that Nvidia controls chip allocation. As long as GPU supply remains constrained, it can route the marginal incremental accelerator to whichever capacity provider it prefers, including its own.
The IREN warrant fits this thesis cleanly. Nvidia gets equity upside if a customer it already supplies grows, gets a long-dated foothold in a 2-gigawatt Texas campus that can plausibly serve as direct Nvidia-cloud capacity, and writes the warrant at a price that incentivises both sides to actually build out the planned footprint. From the hyperscalers' point of view, the deal is a third-party data centre being underwritten by their biggest chip supplier and run on hardware that all three will eventually compete to procure for their own clouds.
Texas grid, $700bn capex, and a 5GW commitment
The 5-gigawatt commitment headline is the part the chart-watchers will fix on. To put it in proportion, 5 GW is roughly the steady-state output of three to five large nuclear reactors, and IREN's planned Sweetwater campus alone is 2 GW. The bottleneck on AI build-outs at this point is no longer chips alone. It is the combination of power, packaging, and permitted sites, with grid interconnection in places like West Texas often the binding constraint.
The deal also lands the same week that all four US tech megacaps reported earnings and signalled that AI capital expenditure will not slow into the back half of 2026. The combined hyperscaler capex run-rate is now on track to surpass $700 billion this year. Nvidia is, in a sense, financing the marginal GW of that build-out from inside its own customer base.
For Nvidia shareholders, the IREN warrant is a low-cost call option on a customer it already supplies. For IREN shareholders, the deal is both validation and dilution: the warrants, if exercised, would represent a meaningful chunk of additional shares outstanding, but the alternative is competing for hyperscaler-grade capacity without an explicit Nvidia anchor. The 9% extended-trading gain suggests the market sees the dilution as more than offset by the strategic positioning.
What to watch next
Two questions follow from the deal. First, how much of Nvidia's 2026 revenue book is now structurally tied to financing instruments it has supplied or sponsored, directly or indirectly. The IREN warrant is small in isolation; aggregated across the equivalent CoreWeave, OpenAI, and other related transactions, the answer is materially larger. Second, whether the Sweetwater build can close on time given the Texas grid's existing AI-driven backlog. The same week's Cloudflare layoffs show how quickly the AI-spending narrative is propagating into headcount decisions; the IREN deal is the upstream side of that same flow.
