The chipmaker highlights AI’s potential to cut waste and stabilize grids. The challenge: ensuring its own technology doesn’t overwhelm the power system.

At Climate Week NYC, NVIDIA spotlighted how artificial intelligence could accelerate the clean energy transition. The company points to research suggesting AI could save 4.5% of global energy demand by 2035 by optimizing transport, buildings, and industrial processes. Its Earth-2 climate platform aims to give utilities more precise forecasts for storms and renewable output, helping grids adapt faster to extreme weather.

NVIDIA is betting on “sustainable computing” as its next growth story. The company says its latest GPUs have reduced carbon intensity by 24% compared to prior generations and notes that its offices and data centers now run on 100% renewable energy. It has also partnered with startups designing AI-optimized, energy-efficient data centers, which it argues could unlock gigawatts of grid capacity.

The Other Side of the Ledger

Those claims come against a backdrop of rising concern over data center demand. Training large AI models is highly energy-intensive, and the boom in new facilities is straining grids from Virginia to Texas. Analysts caution that while AI can help reduce waste, it also risks adding significant new load — forcing utilities to weigh new gas plants alongside clean energy investments.

AI will almost certainly play a role in the energy transition, from better grid management to faster climate modeling. The question is whether its growth can be scaled responsibly, with benefits that outweigh the costs.

For energy consumers, that balance will determine whether AI is remembered as part of the solution — or part of the problem.

The Bottom Line

NVIDIA’s climate pitch reflects a broader truth: AI has real potential to cut waste and strengthen resilience, but its own energy footprint can’t be ignored. The technology may well help solve parts of the climate challenge — but only if its rapid expansion doesn’t deepen the pressures already facing the grid.