2 min read
AI and Decarbonization: Compute as an Asset Class, Scope 3 Insetting and LLM Simulations
NVIDIA turned AI compute into an investable asset class. The same blueprint can fund Scope 3 decarbonization via insetting, and LLMs can help steer decisions.
AI and climate usually show up in the same sentence when people talk about data center energy use. I'm more interested in the other direction: what the AI industry can teach decarbonization, and what AI models can do to help.
NVIDIA's blueprint: AI compute as an investable asset class
NVIDIA is mobilizing $500 billion by turning AI compute into a standardized, investable asset class. Jensen Huang shared it as "NVIDIA AI Factory Compute Is Becoming an Investable Asset Class".
This is the exact blueprint needed for industrial decarbonization.
Why Scope 3 decarbonization fails to attract capital
Today, Scope 3 emissions remain a hard challenge. Companies treat decarbonization as an isolated compliance cost, which fails to attract long-term capital. We need a global market that treats GHG abatement as investable assets without the risks of greenwashing.
How carbon insetting closes the capital gap
Carbon insetting solves this capital gap. It stops asking individual corporations to fund massive supplier retrofits alone. Instead, insetting turns supplier upgrades (like fuel switching and electrification) into standardized, fungible assets. This allows multiple buyers to co-finance and de-risk projects through shared offtake agreements.
For instance, Carbon3 is building exactly this: moving beyond basic carbon accounting to turn Scope 3 reductions into a repeatable, globally tradable asset class.
Can LLMs help steer industry and policy decisions?
Carbon3 put the question to the models directly, in a video called "We Asked the Smartest AI models about Insetting". My take:
If LLMs like GPT 5.6 Sol already managed to solve over ten long-standing open mathematical problems, I'm pretty convinced they can help steer industry and policy leaders to make the right decisions.
That doesn't mean handing decisions to a model. It means using models that can reason through very hard problems to test the options before people commit capital to them.
Simulating a Scope 3 market with AI
That's also the idea behind the AI-run Scope 3 market simulations Carbon3 published: let AI agents play out how a market for supplier decarbonization could behave, before anyone builds it for real.
These simulations have cost us a lot of tokens. But compared to finding out the hard way in a real market, tokens are cheap.