For the past two years, the AI race has been defined by one question: who can get enough Nvidia Corp. (NASDAQ:NVDA) chips? According to IREN Ltd. (NASDAQ:IREN) co-founder and co-CEO Daniel Roberts, investors may soon be asking a very different question.
Speaking recently to Bloomberg Tech about the state of AI infrastructure, Roberts said a company looking to build a 1-gigawatt AI factory today likely would not get its first compute online until 2030.
The Next AI Bottleneck
The comment highlights a growing challenge facing the AI industry as hyperscalers, model developers and data-center operators race to expand capacity.
While much of Wall Street’s attention remains focused on Nvidia’s GPUs, securing the chips is increasingly becoming just one piece of the puzzle.
Bringing a large-scale AI campus online requires access to power, transmission infrastructure, substations, permits and years of development work. As demand for AI compute continues to surge, the industry’s biggest constraint may no longer fit inside a server rack.
Why Power Stocks Are Entering The AI Conversation
Roberts’ comments help explain why investors have increasingly turned their attention to companies sitting at the intersection of power and AI infrastructure.
Names such as Bloom Energy Corp. (NYSE:BE), which provides on-site power solutions, are emerging as AI-adjacent plays as data centers seek reliable electricity sources.
Meanwhile, infrastructure developers including IREN, TeraWulf Inc. (NASDAQ:WULF), Applied Digital Corp. (NASDAQ:APLD) and Nebius Group N.V. (NASDAQ:NBIS) are increasingly being valued not just for compute capacity, but for access to power itself.
The AI Arms Race Is Changing
The first phase of the AI boom was about securing GPUs.
The next phase may be about securing megawatts.
That shift could reshape how investors think about AI winners. While Nvidia remains at the center of the AI ecosystem, companies controlling power generation, grid access and energy infrastructure are becoming increasingly important as developers pursue larger training clusters and inference workloads.
In a market obsessed with chips, Roberts’ 2030 timeline serves as a reminder that building AI infrastructure involves much more than buying hardware.
Sometimes the hardest part is simply getting enough electricity to turn it on.
Photo by Below the Sky via Shutterstock
Recent Comments