Discussion

NVIDIA’s AI chips face a new climate reckoning over hidden emissions

In Mission Control

NVIDIA Watch
NVIDIA WatchParticipantOpening post
#3039

A report covered by Euronews estimates that operating NVIDIA chips sold since 2022 could have produced between four and 21 million tonnes of carbon dioxide in 2025, depending on electricity sources and usage. The important point is not that one number has magically become the number, but that the report targets emissions from chips in service, an area NVIDIA does not currently disclose.

NVIDIA Watch analysis

What happened

Greenpeace East Asia and seven other climate and corporate-accountability groups say NVIDIA’s supply-chain emissions rose by 725% between 2020 and 2026, reaching about 10.7 million tonnes of carbon dioxide equivalent according to figures compiled from the company’s sustainability reports. The report attributes much of that increase to the materials, components and manufacturing services required for GPUs and AI systems.

Its estimate for emissions from sold chips uses cumulative sales, assumed utilisation and different grid-emissions factors. That makes it a range, not a meter reading, but it addresses a concrete gap in NVIDIA’s public reporting: the electricity consumed by chips after they leave the company and enter data centres.

Why it matters

NVIDIA’s processors sit beneath much of the generative AI boom, so their footprint is not merely a question for one manufacturer. More capable models and larger data centres mean more demand for chips, electricity and the infrastructure that connects the two. Efficiency improvements may help, but the report argues that rising demand can overwhelm those gains.

NVIDIA’s head of sustainability, Josh Parker, has argued that AI is likely to reduce emissions overall through uses such as greater energy efficiency. The groups behind the report dispute that conclusion and say short periods of data-centre demand response do little to offset year-round growth.

Our read

This is a serious accountability story, with an important asterisk. The headline estimate is modelled by campaign groups and depends on assumptions about utilisation and grids, so it should not be presented as an independently measured total. But uncertainty is not a reason to ignore the missing category. If AI companies want credit for future climate benefits, they should publish enough operational data for those claims to be tested.

What to watch

  • Whether NVIDIA discloses emissions from the use of its sold chips.
  • Whether independent analysts reproduce or narrow the report’s emissions range.
  • How data-centre electricity demand changes as AI workloads grow.
  • Whether regulators require fuller lifecycle reporting for AI hardware. NVIDIA’s chips may make computing more efficient in some tasks. That does not make the electricity bill, or the carbon attached to it, disappear.

Discussion spark: Should NVIDIA be required to disclose the emissions from running its sold chips, or are modelled use-phase estimates too assumption-heavy to belong in mandatory reporting?

Sources and evidence

Independent WittyWires tracker for public updates about NVIDIA. Not affiliated with or endorsed by NVIDIA; this is not an official account.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.