AI infrastructure now requires 100x more power than the previous year, making electricity a critical bottleneck for scaling.
Nvidia CEO Jensen Huang stated this week at Nvidia's GPU Technology Conference that "the amount of computation needed is easily 100 times more than we thought we needed at this time last year." This marks a striking reversal from the concern that gripped the tech and investment world following the emergence of DeepSeek, the Chinese low-cost AI platform, earlier this year. At that time, investors worried that AI would require drastically fewer semiconductor chips and less power, threatening the billions of dollars Wall Street expected to flow toward bottom lines. Instead, Huang emphasized that "this last year, this is where almost the entire world got it wrong." Rather than restraining AI investment, DeepSeek's arrival accelerated it, forcing the AI community to transition from earlier Large Language Models focused on generation to "reasoning" AI models with explicit reasoning capabilities.
The shift to reasoning models fundamentally increases computational demands. These models spend more time "thinking" about a problem before delivering an answer, breaking each prompt down into steps—a process particularly suited for complex problems. As Huang noted, "Users won't want to wait 10 times longer to get an answer that relies on 10 times more data." This computational surge drives explosive demand for datacenter infrastructure and electrical power. The International Energy Agency's "Electricity 2024" report projects that electricity consumption from data centers, artificial intelligence, and the cryptocurrency sector could double by 2026, with data centers being "significant drivers of growth in electricity demand in many regions." SemiAnalysis.com reports the IEA estimates 90 terawatt-hours (TWh) of power demand from AI datacenters by 2026—equivalent to about 10 Gigawatts—with AI datacenter capacity demand crossing above 10 GW by early 2025.
The boom in AI cluster demand has created extreme stress on electricity grids and generation capacity, with AI buildouts heavily limited by lack of datacenter capacity, particularly for training where GPUs must be co-located for high-speed chip-to-chip networking. The surge has benefited power and infrastructure companies across the sector. Beyond datacenters, Huang identified robotics as the next frontier, declaring "The time has come for robots," and noting a growing shortage of human labor. Nvidia's physical AI leverages both "slow-thinking" capability—allowing robots to perceive and reason about their environment—and fast-thinking capability for action. "Everyone, pay attention. This could very well be the largest industry of all," Huang said. Research firm GlobalX estimates the potential market opportunity for general-purpose humanoids at almost $3 trillion by 2035, based on assumptions of 15% household penetration and a $10,000–$15,000 price point, with Tesla CEO Elon Musk and industry stakeholders believing there could be over 1 billion humanoids on Earth by the 2040s. GlobalX further estimates the industrial humanoid market—for automating intensive production tasks—at nearly $2 trillion over the next decade.
Huang's characterization of robotics as "the largest industry of all" reflects the scale of opportunity emerging across the AI ecosystem. Companies including Tesla, Meta, Apple, Alphabet, Nvidia, and OpenAI are advancing aspects of humanoid technology. As the conference continues with "Quantum Day," where industry leaders explore the future of quantum computing, Barron's notes that quantum computing is "set to take center stage at Nvidia GTC, signaling that the emergent technology may no longer be decades in the future; rather, the future is approaching."