On May 7, NVIDIA and IREN disclosed two linked moves in a single announcement: IREN could build up to 5 gigawatts of AI infrastructure for NVIDIA over five years, while NVIDIA is also committing about $3.4 billion to rent IREN's GPU cloud capacity.
The pairing is subtle but important. Taken together, it shows Jensen Huang doing something NVIDIA has rarely done outside the hyperscaler circle: putting equity weight behind an independent compute operator.
A $2.1 billion option is really a claim on future capacity
The first headline number is NVIDIA's five-year call option. It can buy up to 30 million IREN shares at $70 per share, for a maximum notional value of $2.1 billion.
NVIDIA is not writing that full check immediately. The option is a right it can exercise over the next five years. Two details matter:
- The strike is set at $70: IREN closed at $61.06 on the day of the announcement, so the option was not yet in the money.
- The stake would be meaningful: IREN is worth about $20 billion, and 30 million shares would be large enough to matter for governance and market signaling.
NVIDIA has usually reserved equity support of this size for companies such as CoreWeave and Lambda, operators that sit close to the new hyperscaler category. IREN is smaller. Giving it this level of backing suggests NVIDIA sees it as a long-term compute base, not just a temporary supplier.
The $3.4 billion GPU cloud deal says NVIDIA also needs capacity
NVIDIA did not emphasize the second number in its own announcement, but financial media surfaced the more revealing detail:
IREN agreed to provide NVIDIA with roughly $3.4 billion of hosted GPU cloud services over five years for NVIDIA's internal AI and research workloads.
At first glance, it feels backwards: the company that makes the GPUs is renting GPUs from someone else. In practice, it exposes the industry's bottleneck. Owning chips and operating a usable AI cloud are different businesses.
Power, cooling, network topology, operations and compliance are the hard parts NVIDIA does not necessarily want to run itself. A five-year rental agreement buys time and certainty.
The deeper point is that NVIDIA is not short of chips as much as it is short of ready data-center capacity. Binding itself to a company with land and power allocations is the fastest way around that constraint.
What 5 gigawatts means
Sweetwater is the flagship project in the partnership: a 2-gigawatt campus in Texas that is expected to be an initial deployment site for NVIDIA's DSX architecture.
Put 5 gigawatts next to other AI infrastructure projects:
| Company or project | AI compute scale (GW) |
|---|---|
| Stargate phase one (OpenAI + Oracle) | 1.2 |
| Anthropic + AWS Project Rainier | 1.0+ |
| Microsoft global AI data centers | ~3 |
| IREN ↔ NVIDIA planned ceiling | 5 |
In other words, the upper end of the IREN-NVIDIA plan would exceed Microsoft's current AI data-center power footprint. That is not exaggeration; the scale of AI compute is moving by the month.
Jensen Huang put it this way in the announcement:
AI factories are becoming foundational infrastructure for the global economy. Deploying these systems at scale requires deep integration across the full stack.
Translated into strategy: the era of simply selling boards is giving way to bundled products that combine electricity, buildings, networking, chips and software.
IREN is a former bitcoin miner
Many readers may not know IREN. Its path is a familiar one in this cycle:
- It listed in Australia in 2021 with a business built around renewable-powered bitcoin mining.
- In 2024 it began converting mining sites into AI data centers.
- From 2025 it positioned "AI Cloud" as its core business.
- Today it describes itself as a vertically integrated AI cloud company, spanning land, power and GPU clusters.
That origin is an advantage. The most valuable asset in bitcoin mining is industrial land with grid connections and stable, low-cost power. Those are exactly the scarce inputs for AI data centers. IREN is not a company hastily assembled to chase AI demand; it already had much of the physical base.
IREN shares rose 7.41% on the news to close at $61.06, with volume at 187% of the three-month average. The 52-week high is $76.87, and the stock is still below the $70 exercise price. If NVIDIA really starts shipping systems into those campuses, crossing $70 becomes a question of execution.
The real message
NVIDIA has been balancing two paths: remain a chip seller, or move deeper into cloud services.
Building its own cloud would antagonize AWS, Azure and Google Cloud, its three largest customers. Staying out of cloud leaves hyperscalers to capture the service margin on NVIDIA hardware.
The compromise is to use equity and long-term contracts to cultivate a group of NVIDIA-friendly mid-sized clouds: CoreWeave, Lambda, IREN and others. They cannot directly challenge the largest hyperscalers, but they can serve NVIDIA's own research demand, startups and companies that do not want to rely on the big three clouds.
Seen together, the two numbers in this deal -- the $2.1 billion option and the $3.4 billion purchase commitment -- say something plain: NVIDIA wants to turn IREN into a de facto dedicated cloud partner.
Who comes next? CoreWeave is already there. Nebius has been moving. Cerebras has been emphasizing its NVIDIA relationship in IPO materials. Huang appears ready to issue more than one ticket in this race.
Sources: NVIDIA and IREN Announce Strategic Partnership to Accelerate Deployment of up to 5 Gigawatts of AI Infrastructure (NVIDIA Newsroom); IREN inks AI infrastructure deal with Nvidia (CNBC); Why IREN Stock Jumped Today (The Motley Fool); CocoLoop; NVIDIA and IREN map AI data center buildout that could hit 5 gigawatts (StockTitan)