NVIDIA · 2026-05-07 · major
NVIDIA-IREN $2.1B Partnership Targets 5 GW of AI Infrastructure, Plus $3.4B Cloud Services Deal
NVIDIA gets a five-year right to buy up to 30M IREN shares at $70 ($2.1B max) while signing a separate $3.4B managed-GPU cloud contract, anchoring a planned 5 GW DSX-aligned data-center buildout led by IREN's 2 GW Sweetwater, Texas campus.
NVIDIA stakes IREN with up to $2.1B and signs a $3.4B GPU-cloud contract to build 5 GW of AI factories, flagshipped in Sweetwater, Texas.
Key specs
| Investment right | $2.1B |
|---|---|
| Cloud services contract | $3.4B |
| Shares | 30M @ $70 |
| Infrastructure scale | 5 GW |
| Sweetwater campus | 2 GW |
What is it?
A two-part deal between NVIDIA and IREN, an Australian-listed data-center operator. NVIDIA receives a five-year right to purchase up to 30 million IREN shares at $70 each — up to $2.1B, subject to regulatory approval — and signs a five-year, $3.4B managed-GPU cloud services contract that lets NVIDIA tap IREN-operated capacity for its own AI and research workloads.
How does it work?
Future joint deployments are expected to focus on IREN's 2 GW Sweetwater, Texas campus, planned as the flagship site for NVIDIA's DSX AI-factory architecture (rack-, cluster-, and site-level reference designs). The $3.4B cloud contract is anchored on roughly 60 MW of existing air-cooled Blackwell systems at IREN's Childress, Texas campus. Together the two arrangements give NVIDIA both equity exposure and dedicated managed capacity from a single operator.
Why does it matter?
The deal extends NVIDIA's pattern of taking equity in cloud and data-center partners (CoreWeave, Lambda) that buy its silicon, while locking in roughly 5 GW of long-term DSX-aligned capacity at a moment when frontier AI training is constrained more by power and steel than chips. For IREN, it converts a Bitcoin-miner-turned-AI-host into one of NVIDIA's flagship infrastructure partners outside the hyperscalers.
Who is it for?
Infrastructure buyers, NVDA/IREN investors, anyone tracking the AI-factory buildout.
Try it
Read the joint announcement at nvidianews.nvidia.com.