IREN co-founder Daniel Roberts laid out a vertically integrated AI infrastructure thesis in a lengthy X thread on Friday, arguing that the binding constraint on the AI build-out is no longer silicon but physical infrastructure — power, land, cooling and data center capacity. "AI demand grows exponentially. Infrastructure doesn't," Roberts wrote, framing IREN's strategy around three layers: owned physical infrastructure, NVIDIA GPU-based compute, and enterprise software and operational tooling. "Layers 1 and 2 are where the overwhelming majority of IREN's value is being created today," he said.
The company, formerly Iris Energy, has expanded from Bitcoin mining into AI infrastructure across Texas, British Columbia, Oklahoma, Spain and Australia, and Roberts said IREN has secured roughly 5 gigawatts of grid-connected capacity globally. The thread also surfaced IREN's $3.4 billion, five-year AI cloud contract tied to NVIDIA Blackwell GPU deployments in Texas — one of the largest direct GPU-of-cloud deals between a non-hyperscaler and NVIDIA disclosed to date.
Why it matters
Roberts' argument lands in a market that has spent two years treating NVIDIA as the bottleneck. The pitch is that power interconnect queues, substation build times, and land with the right fiber and water cooling are now the gating resource, and that vertically integrated operators who own those assets capture the spread that pure GPU resellers cannot. The 5GW figure, if real, would position IREN among the larger non-utility power footprints held by a single AI-neocloud name.
Market impact
IREN shares gained roughly 10% on Thursday, and sister company WhiteFiber (WYFI) jumped 22% the same day after announcing a separate five-year AI compute agreement worth more than $160 million in the Paris region using NVIDIA GPUs, adding another 5% in Friday premarket. The two read together as a validation of the vertically integrated thesis from public market flow — though the multi-year capex to actually use 5GW of secured capacity is the open question the next several quarters will answer.
Frequently asked questions
-
Who is Daniel Roberts and what is his role at IREN?
Daniel Roberts is co-founder of IREN (formerly Iris Energy). He outlined a vertically integrated AI infrastructure thesis in a lengthy X thread on Friday, framing the company as a full-stack AI platform spanning power, data centers, GPUs and enterprise software.
-
What does IREN mean by '5GW of grid-connected capacity'?
According to Roberts' thread, IREN has secured roughly 5 gigawatts of grid-connected power capacity globally across sites in Texas, British Columbia, Oklahoma, Spain and Australia — the physical foundation the company is betting becomes the binding AI constraint.
-
What is the $3.4B NVIDIA deal tied to Blackwell GPUs?
IREN disclosed a five-year, $3.4 billion AI cloud contract tied to NVIDIA Blackwell GPU deployments in Texas. It is one of the largest direct GPU-of-cloud agreements between a non-hyperscaler and NVIDIA disclosed to date.
-
How does WhiteFiber (WYFI) differ from IREN?
WhiteFiber provides AI cloud and high-performance compute services using third-party data center infrastructure, while IREN focuses on owning and operating the underlying infrastructure itself. WhiteFiber announced a separate five-year, $160M+ AI compute deal in the Paris region using NVIDIA GPUs.
-
Why is the 'infrastructure, not chips' argument significant for AI investors?
Roberts' argument reframes the AI build-out as a power-and-real-estate race rather than a silicon race. If physical constraints — power interconnect queues, substation build times, suitable land — are the true bottleneck, vertically integrated operators with secured grid capacity could capture a margin spread that…
CoinDesk