Descriptions:
IREN co-CEO Dan Roberts joins Bloomberg Technology at Dell Technologies World to discuss the company’s position as a global AI cloud infrastructure provider with data centers in Texas, British Columbia, and other regions. The conversation focuses on the physical bottlenecks constraining AI compute capacity—not silicon, but power, steel, and concrete. Roberts offers a candid timeline: starting a one-gigawatt AI data center from scratch today means waiting 18 to 24 months just to get a utility’s attention, with a realistic online date in the early 2030s.
IREN recently announced a five-gigawatt compute partnership with Nvidia, which also made a direct equity investment in the company. The two organizations are collaborating on Nvidia’s reference architecture for a gigawatt-scale factory in Sweetwater, Texas, with Dell serving as IREN’s primary server partner. Roberts attributes IREN’s relative advantage to an early strategic decision—made eight years ago—to site facilities in regional communities near abundant renewable energy rather than in metropolitan areas, avoiding both higher energy costs and the public backlash other operators face.
Roberts characterizes the current AI era as still in its “dial-up” phase, arguing that as inference latency drops and prompts become near-instant, demand will accelerate sharply, requiring far more physical infrastructure investment than the industry is currently planning. He identifies power grid connection as the single biggest constraint, noting that utilities are risk-averse by design and currently inundated with interconnection requests.
📺 Source: Bloomberg Technology · Published May 18, 2026
🏷️ Format: Interview
