Nvidia plans to invest up to $2.1 billion in IREN, the data center operator once known as Iris Energy, through a five-year warrant to buy as many as 30 million shares at $70 each, tying one of the world's most important chip companies directly to a bitcoin-mining turned AI cloud operator with massive power access in Texas and Canada.
The structure matters because this is not just a passive cheque. IREN said Nvidia will receive the right to buy up to 30 million shares over five years, which would amount to the full $2.1 billion only if the warrant is exercised in full. At the same time, the two companies are working on a much larger infrastructure plan that could expand to 5 gigawatts of AI compute deployments, with the first major focus on IREN's 2-gigawatt Sweetwater campus in Texas. The announcement also came with a separate $3.4 billion AI cloud computing contract for Nvidia Blackwell processors, which tells you this is a financing partnership and a supply agreement at the same time.
IREN is not starting from zero. The company has spent the last year aggressively repurposing its bitcoin-mining footprint into AI infrastructure. It had already built out 660 megawatts of data center capacity and 52 exahash per second of mining capacity, while pausing further mining expansion to focus on AI cloud services and AI data center businesses. By September 2025, IREN said it had 23,000 GPUs, including H100s, B200s, GB300s, and AMD MI350X units, and was targeting a $500 million annualized run rate from its AI cloud segment. That is a real operating base, not a concept deck. The company has also signed a $9.7 billion cloud deal with Microsoft, which already put it in the category of infrastructure provider rather than pure miner.
For SF readers, this is one of the clearest examples yet of crypto infrastructure bleeding into AI infrastructure. Bitcoin miners have always had an unusual asset base, cheap power, large land footprints, and experience running dense electrical loads. Those attributes were mostly treated as a way to mine coins at a low marginal cost. Now they are being repriced as a strategic advantage for AI compute. If you already own energized sites, grid relationships, cooling systems, and construction permits, you are closer to becoming a GPU cloud operator than a greenfield startup starting from scratch. The Nvidia deal simply puts a prestigious balance-sheet backer on top of that transition.
The timing also says something about Nvidia's broader strategy. The company has become increasingly willing to finance, support, or structurally back the infrastructure layer around its own chips. That does not mean it is buying customers outright. It means Nvidia understands that GPU supply is not the only bottleneck anymore. Power, land, data center readiness, and financing are all constraining the pace of AI deployment. By aligning with IREN, Nvidia is effectively helping create another large-scale destination for its Blackwell hardware while also reducing the risk that demand is stranded by a lack of usable infrastructure.
That has competitive consequences. GPU cloud startups depend on scarce access to chips and capital. If Nvidia itself is helping capitalise a former miner with strong power access and a growing AI cloud business, the barrier to entry rises for smaller operators without energized land or an existing balance sheet. It is one thing to buy GPUs. It is another to build the facilities and finance the power contracts needed to house them. IREN's story suggests miners with the right sites may become the overlooked winners of the AI buildout, especially in markets where electricity and permitting are the limiting factors rather than hardware procurement alone.
There is still execution risk. IREN's mining business remains significant, and the transition from bitcoin infrastructure to AI infrastructure is capital intensive, timing sensitive, and exposed to GPU supply, utilization, and customer concentration. But the Nvidia investment is a strong signal that the market is willing to underwrite that pivot at a much larger scale than before. For the AI industry, it reinforces a familiar truth. The next winners are not always the companies with the flashiest model demos. Sometimes they are the ones that already own the power, the dirt, and the wiring.
Also read: The Nvidia chip smuggling network shows Washington's AI containment strategy has a serious leakage problem • TSMC's record quarter is proof that AI infrastructure spending is real, not just projected • Minnesota's first-in-the-nation AI nudification law shows regulation is moving from rhetoric to criminal liability
