Bold claim: AI’s insatiable energy appetite is the single biggest hurdle standing between today’s innovations and widespread, reliable deployment. That’s the framing Google Cloud CEO Thomas Kurian used at Fortune’s Brainstorm AI event in San Francisco, where he outlined a three-part plan to power AI without choking the grid.
Kurian reminded attendees that Google Cloud has been planning for AI energy needs long before large language models exploded onto the scene. The core worry was energy itself—the bottleneck that could slow or stall AI progress as much as chips and data center capacity. He said, in simpler terms, the power must be ready and steady before training a massive model. To that end, Google Cloud is pursuing three strategic threads.
First, diversify the energy mix. In practice, this means not relying on a single power source. Kurian argued that certain energy types aren’t suitable for the moment of peak AI training, when a training job can create a dramatic energy spike. A mixed, resilient energy portfolio is essential to avoid supply gaps during critical compute bursts.
Second, maximize efficiency. Google Cloud doesn’t just plug in power; it optimizes consumption. The company uses its own AI-driven control systems to monitor and manage thermodynamics and energy flow inside data centers, seeking to extract as much value as possible from every kilowatt. In short, smarter cooling and smarter electrical management can shrink energy waste while keeping performance high.
Third, invest in new energy technologies. Kurian hinted at pursuing fundamental, possibly groundbreaking approaches to generating energy for AI, though he didn’t disclose specifics. The idea is to explore novel methods that could decouple AI growth from traditional power constraints.
This energy-centric strategy arrives as data center capacity is poised to expand significantly. Industry figures estimate roughly a 46% rise in global data center capacity over the next two years, with billions of watts added to the grid—while AI workloads themselves grow in scale and complexity. In parallel, NextEra Energy and Google Cloud announced an expanded partnership to develop new U.S. data center campuses that will include new power generation facilities, signaling a tangible move from plan to practice.
Industry leaders warn that energy availability is a critical bottleneck alongside chips and model improvements. Building out data centers is another potential chokepoint; Nvidia’s Jensen Huang has famously contrasted the U.S. timeline for constructing AI-focused facilities with China’s rapid build cycles, suggesting a three-year gestation for a U.S. AI supercomputer versus a weekend for constructing a hospital abroad. This dynamic underscores why energy strategy matters as much as algorithms and accelerators for the AI race.
If the energy puzzle can be solved through diversification, efficiency, and innovative generation, the path to scalable AI becomes clearer. But questions remain: will these strategies deliver reliably at scale, and how will policymakers, utilities, and consumers respond to the growing demand for AI energy? Share your thoughts on where the biggest risk or opportunity lies in this energy-AI intersection.
