"Jensen Huang says the biggest limit to AI is not chips but electricity, predicting that tech giants will build small nuclear reactors to power future data centers."
When Nvidia CEO Jensen Huang sat down with Joe Rogan, he delivered one of the most striking predictions in tech today: the real bottleneck for artificial intelligence is not computing power — it is electricity. And to solve it, big tech companies may soon operate their own small nuclear reactors right next to massive data centers.
Why Is Energy Now the Bigger Problem Than Chips?
For years, the AI race was about building faster and more powerful GPUs. That era is shifting. Huang explained that while chip production continues to improve, power supply has become the critical limit. Training large AI models and running real-time inference systems consume enormous amounts of energy.
Modern “AI factories” — huge data centers packed with GPUs — now require electricity at the scale of entire cities. Grids around the world are struggling to keep up.
“Energy is the bottleneck for AI growth,” Huang said during the podcast.
Will Tech Giants Build Nuclear Reactors?
Huang predicts that major technology companies will build localized small modular nuclear reactors (SMRs) capable of producing hundreds of megawatts of clean energy. These reactors would:
- Power nearby AI data centers directly
- Reduce strain on national power grids
- Provide stable, 24/7 carbon-free electricity
- Feed excess power back into surrounding communities
This move would effectively turn large tech firms into mini power utilities while ensuring nonstop power for AI workloads.
Are Nuclear Deals Already Happening?
This is not just theory. Real investment is already flowing.
Google and Kairos Power signed an agreement to supply up to 500 MW of advanced nuclear energy by 2035. The first commercial reactor is expected around 2030 as part of Google’s carbon-free data center strategy.
Another collaboration between Kairos Power and the Tennessee Valley Authority will route energy from the Hermes 2 reactor (about 50 MW) into regional grids, while Google claims the clean-energy credits for data centers in Tennessee and Alabama.
Major projects at a glance
| Company | Partner | Planned Capacity |
|---|---|---|
| Kairos Power | 500 MW by 2035 | |
| TVA + Google | Kairos Power | Up to 50 MW (Hermes 2) |
Why Is AI Power Demand Exploding?
Analysts warn that global data-center electricity demand is rising at an unprecedented pace.
- Goldman Sachs projects steep growth in data center consumption through 2030.
- The International Energy Agency estimates total global demand could reach the high hundreds of terawatt-hours annually — more than double today’s levels.
This growth is driven mainly by:
- Training massive foundation models
- Real-time AI inference at scale
- Always-on global cloud services
Huang argues that without high-density baseload energy — especially nuclear — AI expansion will clash with climate commitments and grid limits.
The GPU Story That Started Everything
During the same conversation, Huang reflected on the surprisingly humble origins of today’s AI revolution.
In 2012, researchers at the University of Toronto trained AlexNet, the neural network that launched modern deep learning, on just two Nvidia GTX 580 gaming GPUs using CUDA-based software.
That success convinced Nvidia to pivot from gaming graphics toward AI computing. By 2016, the company launched the DGX-1 supercomputer. Huang even personally delivered early units to OpenAI and Elon Musk — a move that helped define Nvidia’s leadership in the AI hardware market.
What Does This Mean for Builders and Investors?
If nuclear energy becomes the backbone of AI infrastructure, several shifts could follow:
- Cheaper long-term compute costs for cloud AI services
- Rapid expansion of large-scale model training
- New clean-energy investments targeting AI demand
- Increased opportunities across AI SaaS platforms and GPU marketplaces
Developers building AI applications could benefit from more stable compute supply, while investors may track both semiconductor stocks and advanced nuclear companies.
FAQs
Is nuclear power safe for data centers?
New small modular reactors are designed with passive safety systems that reduce meltdown risk and require less on-site staffing than traditional plants.
Will nuclear energy lower AI costs?
In the long run, yes. Stable baseload power may reduce reliance on expensive peak grid electricity and improve price stability.
Which companies may benefit most?
Nvidia, cloud providers like Google and Microsoft, and nuclear technology firms working on SMRs are likely key beneficiaries.
