AI accelerator chip: Interventions
Try this in What if Switch on levers and see what changes in India by 2036.
Interventions
Changes that would cut emissions or material use, by lever and life stage, with effect per unit, readiness, adoption, evidence and who acts
| Lever | Life stage | Change | Effect per unit | Readiness | Adoption | Who acts | Source |
|---|---|---|---|---|---|---|---|
| Use less | Use | Keep accelerators busy: raise the share of peak arithmetic that training achieves (model FLOPs utilisation) | Raising utilisation from 20% to 60% cuts the GPUs needed by about 67% (A100 study) [15]. | Deployed (software and scheduling) | Not found | Model labs, cloud operators | [15] |
| Circularity | Use | Run each accelerator for longer before replacing it | Extending lifespan from one to three years gives savings comparable to raising utilisation; both together cut GPU needs by up to 93% [15]. Assumed lifespans range from 3 years [17] to 6 years [3]. | Deployed | Not found | Cloud operators, buyers | [15][17][3] |
| Efficiency | Use | Move to newer generations that do more work for each watt | Google says Ironwood gives 2x the performance per watt of its predecessor [40]; NVIDIA says the HGX B200 can be as much as 15x more energy efficient for AI inference (its own claim) [50]. Embodied carbon per exaflop of Google's TPUs fell from 114 to 38 gCO2e between v4i and v6e [3]. | Deployed | Rolling out with each generation | Chip designers, cloud operators | [40][50][3] |
| Change route | Use | Power the data centre with carbon-free electricity | For TPU v6e, operational intensity falls from 316 to 118 gCO2e per exaflop under market-based accounting, a 63% cut (calculated) [3]. | Deployed (Google procurement) | Google's carbon-free energy procurement | Cloud operators, utilities | [3] |
| Design | Components design | Stack compute and memory in 3D rather than spreading chips out on a large interposer | When compute units and HBM capacity are doubled, embodied emissions rise 89% with 3D integration against 141% with 2.5D (imec and VUB) [11]. | Research | Not found | Chip designers, packaging firms | [11] |
| Efficiency | Making | Abate fluorinated gases in the fab and cut the gases used in etching | TSMC reports a company-wide abatement rate of 96% for fluorinated gases [30]; imec's TAP process cut CF4 use by 98% [12]. | Deployed (abatement); piloted (new etch) | 96% abatement at TSMC in 2025 [30] | Foundries, equipment makers | [30][12] |
| Electrify | Supply chain | Supply fabs with renewable electricity | Electricity can make up to 60% of the carbon footprint of an advanced N2 logic wafer [12]. | Deployed | Renewable electricity was 11.4% of TSMC's Taiwan electricity in 2025 (2,940 GWh) [30] | Foundries, utilities | [12][30] |
| Business model | Use | Pool public GPUs and rent them out, so more users share fewer idle chips | No saving per unit found; the IndiaAI portal has onboarded over 38,000 GPUs to rent at an affordable rate [23]. | Deployed | 38,000 GPUs on the Indian portal [23] | Governments, cloud operators | [23] |
as of 2026-11-16
Sources
- Life-cycle emissions of AI hardware: a cradle-to-grave approach and generational trends Schneider et al. (Google), arXiv preprint 2502.01671, 2025
- 3D Integration Reduces Embodied Emissions Impact Of GPU Scaling (imec and VUB) Semiconductor Engineering, 2026
- How can we reduce environmental impact in chip manufacturing? imec (first published in Semiconductor Digest, July/August 2025 issue), 2025
- From FLOPs to Footprints: The Resource Cost of Artificial Intelligence (arXiv:2512.04142v1) arXiv preprint, 2025
- More than Carbon: Cradle-to-Grave environmental impacts of GenAI training on the Nvidia A100 GPU (arXiv:2509.00093v3) arXiv preprint, 2025
- IndiaAI Mission expands AI ecosystem with affordable compute and startup support Digital India, Government of India, 2026
- TSMC 2025 Sustainability Report Taiwan Semiconductor Manufacturing Company (TSMC), 2026
- Google offers 7th-gen Ironwood TPUs for AI, with AI-inspired comparisons The Register, 2025
- NVIDIA HGX B200 Reduces Embodied Carbon Emissions Intensity (19 September 2025) NVIDIA Technical Blog, 2025