How 5 Companies Use NVIDIA AI to Accelerate Clean Energy
NVIDIA AI speeds clean energy projects at ThinkLabs, Atomic Canyon, Redwood Materials, TerraPower and Commonwealth Fusion Systems across grid and nuclear.
Summary
On September 21, 2026, at New York Climate Week, NVIDIA highlighted five AI clean energy projects. NVIDIA Inception member ThinkLabs AI, founded by Josh Wong, uses CUDA digital twins and agents to integrate variable generation and unblock grid connections. Southern California Edison cut each application review from 30 to 45 days to two minutes with its simulation agent. Trey Lauderdale's fellow Inception company Atomic Canyon powers Neutron, a nuclear knowledge workbench, and NIVA, a fleet assistant built with the Institute of Nuclear Power Operations, Electric Power Research Institute and Nuclear Energy Institute. PG&E's Diablo Canyon uses its technology; Lauderdale says 50 to 60 year old nuclear methods cannot scale.
Redwood Materials, another Inception member, uses 100% recycled electric vehicle batteries and in house electronics for off grid AI factories, bringing capacity online in months instead of years. NVIDIA Blackwell intelligence follows data center demand; fewer transformers and inverters and no uninterruptible power supply cut costs, while new batteries remain an option. CTO Colin Campbell says it handles AI training fluctuations as electricity demand outruns grid construction.
Nuclear supplies 20% of US electricity. TerraPower CEO Chris Levesque says founders including Bill Gates added emissions avoidance 20 years ago. Its emission free Natrium reactor uses an NVIDIA Omniverse digital twin to cut future plant siting and delivery from years to months. Liquid sodium enables atmospheric pressure, no offsite water or electricity, and passive emergency cooling. Commonwealth Fusion Systems uses Omniverse libraries and OpenUSD to shrink SPARC tokamak experiments from years to weeks. Cofounder Brandon Sorbom says carbon free fusion passively shuts down. High temperature superconductors make SPARC 40x smaller; ARC, its first plant, planned for Chesterfield County, Virginia, in the 2030s, is over 10x smaller and targets cost parity with renewable and nonrenewable power.
Positives
- Southern California Edison reduced each grid interconnection review from 30 to 45 days to two minutes using ThinkLabs AI.
- 100% recycled electric vehicle batteries let Redwood Materials bring off grid AI power capacity online in months instead of years.
- Natrium needs no offsite water or electricity for stability and can cool itself during an emergency without intervention.
- Omniverse and OpenUSD compress Commonwealth Fusion Systems' SPARC experiments from years into weeks.
- High temperature superconductors make SPARC 40x smaller, while the planned ARC power plant is over 10x smaller.
Risks & concerns
- AI electricity demand is accelerating faster than new grid infrastructure can be built.
- Outdated infrastructure, lengthy research timelines and upfront costs continue to slow large scale clean energy adoption.
- Grid variability is increasing uncertainty, requiring utilities to make investment and reliability decisions using probabilistic risk models.
- Trey Lauderdale says nuclear operating methods developed over 50 to 60 years will not scale for current demand.
- Commonwealth Fusion Systems does not expect its first ARC plant to connect to the grid until the 2030s.