A rough estimate of doing as much generative AI coding as the CEO of Nvidia recommends : an increase of 2.7x of the global annual emissions footprint for the average person
-
A rough estimate of doing as much generative AI coding as the CEO of Nvidia recommends : an increase of 2.7x of the global annual emissions footprint for the average person
By Boris Gamazaychikov
According to the CEO of NVIDIA, engineers should be spending >$250k annually on AI coding. By my estimates, that translates to 7-11 tons of CO2e. Compared to the global average individual carbon… | Boris Gamazaychikov | 20 comments
According to the CEO of NVIDIA, engineers should be spending >$250k annually on AI coding. By my estimates, that translates to 7-11 tons of CO2e. Compared to the global average individual carbon footprint (6.6 tCO2e/year), the upper end would raise someone’s footprint 2.7x. That amount of AI coding might seem unrealistic, but Jensen is influential in shaping the industry, and some businesses are already using leaderboards and incentives to encourage more AI usage. In the face of such massive usage, there's still a lot that can be done: - use smaller models - leverage greener hosting locations - integrate energy/carbon data into usage tracking systems (proxies at first for closed models) - choose open models (inherently more transparent) - pressure AI providers on transparency and sustainability Still, more mindful usage of AI is upstream to all of this - contrary to the tokenmaxxing trend. Shoutout Vijay Gadepally for the idea Assumptions: [this is a rough estimation and it's sensitive to the many underlying assumptions] - Model: Claude Opus 4.6 - Pricing and token distribution for coding from Anthropic (80% cached input, 16% output): https://lnkd.in/ehbXur72 - Energy per query for this model from Nidhal Jegham using an updated version of this methodology: https://lnkd.in/eJvB3-ZS - - Range represents variations among possible hardware configurations (DGX H200/H100, DGX B200). - - "Large" prompt configuration confirmed by Anthropic token distribution. - Assumed PJM carbon intensity (418 gCO2e/kWh in 2025 per Electricity Maps) as largest concentration of Anthropic clusters is in that region per Alex Lanin (https://lnkd.in/eFhpVTgs) - Assumed AWS global average PUE of 1.15 https://lnkd.in/ez697P9K - Individual carbon footprint https://lnkd.in/eGGeGXPd | 20 comments on LinkedIn
LinkedIn (www.linkedin.com)
-
(6.6+11.3)/6.6 = 2,7(12)
-
A rough estimate of doing as much generative AI coding as the CEO of Nvidia recommends : an increase of 2.7x of the global annual emissions footprint for the average person
By Boris Gamazaychikov
According to the CEO of NVIDIA, engineers should be spending >$250k annually on AI coding. By my estimates, that translates to 7-11 tons of CO2e. Compared to the global average individual carbon… | Boris Gamazaychikov | 20 comments
According to the CEO of NVIDIA, engineers should be spending >$250k annually on AI coding. By my estimates, that translates to 7-11 tons of CO2e. Compared to the global average individual carbon footprint (6.6 tCO2e/year), the upper end would raise someone’s footprint 2.7x. That amount of AI coding might seem unrealistic, but Jensen is influential in shaping the industry, and some businesses are already using leaderboards and incentives to encourage more AI usage. In the face of such massive usage, there's still a lot that can be done: - use smaller models - leverage greener hosting locations - integrate energy/carbon data into usage tracking systems (proxies at first for closed models) - choose open models (inherently more transparent) - pressure AI providers on transparency and sustainability Still, more mindful usage of AI is upstream to all of this - contrary to the tokenmaxxing trend. Shoutout Vijay Gadepally for the idea Assumptions: [this is a rough estimation and it's sensitive to the many underlying assumptions] - Model: Claude Opus 4.6 - Pricing and token distribution for coding from Anthropic (80% cached input, 16% output): https://lnkd.in/ehbXur72 - Energy per query for this model from Nidhal Jegham using an updated version of this methodology: https://lnkd.in/eJvB3-ZS - - Range represents variations among possible hardware configurations (DGX H200/H100, DGX B200). - - "Large" prompt configuration confirmed by Anthropic token distribution. - Assumed PJM carbon intensity (418 gCO2e/kWh in 2025 per Electricity Maps) as largest concentration of Anthropic clusters is in that region per Alex Lanin (https://lnkd.in/eFhpVTgs) - Assumed AWS global average PUE of 1.15 https://lnkd.in/ez697P9K - Individual carbon footprint https://lnkd.in/eGGeGXPd | 20 comments on LinkedIn
LinkedIn (www.linkedin.com)
@ketan That's crazy economically too.. if every developer is costing the company $250k a year on top of their salary then the company wouldn't last long.
-
A rough estimate of doing as much generative AI coding as the CEO of Nvidia recommends : an increase of 2.7x of the global annual emissions footprint for the average person
By Boris Gamazaychikov
According to the CEO of NVIDIA, engineers should be spending >$250k annually on AI coding. By my estimates, that translates to 7-11 tons of CO2e. Compared to the global average individual carbon… | Boris Gamazaychikov | 20 comments
According to the CEO of NVIDIA, engineers should be spending >$250k annually on AI coding. By my estimates, that translates to 7-11 tons of CO2e. Compared to the global average individual carbon footprint (6.6 tCO2e/year), the upper end would raise someone’s footprint 2.7x. That amount of AI coding might seem unrealistic, but Jensen is influential in shaping the industry, and some businesses are already using leaderboards and incentives to encourage more AI usage. In the face of such massive usage, there's still a lot that can be done: - use smaller models - leverage greener hosting locations - integrate energy/carbon data into usage tracking systems (proxies at first for closed models) - choose open models (inherently more transparent) - pressure AI providers on transparency and sustainability Still, more mindful usage of AI is upstream to all of this - contrary to the tokenmaxxing trend. Shoutout Vijay Gadepally for the idea Assumptions: [this is a rough estimation and it's sensitive to the many underlying assumptions] - Model: Claude Opus 4.6 - Pricing and token distribution for coding from Anthropic (80% cached input, 16% output): https://lnkd.in/ehbXur72 - Energy per query for this model from Nidhal Jegham using an updated version of this methodology: https://lnkd.in/eJvB3-ZS - - Range represents variations among possible hardware configurations (DGX H200/H100, DGX B200). - - "Large" prompt configuration confirmed by Anthropic token distribution. - Assumed PJM carbon intensity (418 gCO2e/kWh in 2025 per Electricity Maps) as largest concentration of Anthropic clusters is in that region per Alex Lanin (https://lnkd.in/eFhpVTgs) - Assumed AWS global average PUE of 1.15 https://lnkd.in/ez697P9K - Individual carbon footprint https://lnkd.in/eGGeGXPd | 20 comments on LinkedIn
LinkedIn (www.linkedin.com)
Remember when we were told that an ai can do tasks more energy efficient than a human because it works so much faster?
Ketan Joshi (@ketan@climatejustice.social)
Attached: 1 image A rough estimate of doing as much generative AI coding as the CEO of Nvidia recommends : an increase of 2.7x of the global annual emissions footprint for the average person By Boris Gamazaychikov https://www.linkedin.com/feed/update/urn:li:activity:7454529889153658880/
Climate Justice Social (climatejustice.social)
-
Remember when we were told that an ai can do tasks more energy efficient than a human because it works so much faster?
Ketan Joshi (@ketan@climatejustice.social)
Attached: 1 image A rough estimate of doing as much generative AI coding as the CEO of Nvidia recommends : an increase of 2.7x of the global annual emissions footprint for the average person By Boris Gamazaychikov https://www.linkedin.com/feed/update/urn:li:activity:7454529889153658880/
Climate Justice Social (climatejustice.social)
@catraxx@tech.lgbt @ketan@climatejustice.social wasn't there a linkedin/reddit posting some time ago, taht a guy just let his "AI assistent" play SimCity (or soemthing similiar) to burn tokens and make mangement happy.
And he used up so much he got a promotion and a raise? -
@catraxx@tech.lgbt @ketan@climatejustice.social wasn't there a linkedin/reddit posting some time ago, taht a guy just let his "AI assistent" play SimCity (or soemthing similiar) to burn tokens and make mangement happy.
And he used up so much he got a promotion and a raise? -
A rough estimate of doing as much generative AI coding as the CEO of Nvidia recommends : an increase of 2.7x of the global annual emissions footprint for the average person
By Boris Gamazaychikov
According to the CEO of NVIDIA, engineers should be spending >$250k annually on AI coding. By my estimates, that translates to 7-11 tons of CO2e. Compared to the global average individual carbon… | Boris Gamazaychikov | 20 comments
According to the CEO of NVIDIA, engineers should be spending >$250k annually on AI coding. By my estimates, that translates to 7-11 tons of CO2e. Compared to the global average individual carbon footprint (6.6 tCO2e/year), the upper end would raise someone’s footprint 2.7x. That amount of AI coding might seem unrealistic, but Jensen is influential in shaping the industry, and some businesses are already using leaderboards and incentives to encourage more AI usage. In the face of such massive usage, there's still a lot that can be done: - use smaller models - leverage greener hosting locations - integrate energy/carbon data into usage tracking systems (proxies at first for closed models) - choose open models (inherently more transparent) - pressure AI providers on transparency and sustainability Still, more mindful usage of AI is upstream to all of this - contrary to the tokenmaxxing trend. Shoutout Vijay Gadepally for the idea Assumptions: [this is a rough estimation and it's sensitive to the many underlying assumptions] - Model: Claude Opus 4.6 - Pricing and token distribution for coding from Anthropic (80% cached input, 16% output): https://lnkd.in/ehbXur72 - Energy per query for this model from Nidhal Jegham using an updated version of this methodology: https://lnkd.in/eJvB3-ZS - - Range represents variations among possible hardware configurations (DGX H200/H100, DGX B200). - - "Large" prompt configuration confirmed by Anthropic token distribution. - Assumed PJM carbon intensity (418 gCO2e/kWh in 2025 per Electricity Maps) as largest concentration of Anthropic clusters is in that region per Alex Lanin (https://lnkd.in/eFhpVTgs) - Assumed AWS global average PUE of 1.15 https://lnkd.in/ez697P9K - Individual carbon footprint https://lnkd.in/eGGeGXPd | 20 comments on LinkedIn
LinkedIn (www.linkedin.com)
@ketan The war for territory is just a distraction from tech companies truing to annex our ability to think without them.
I wonder is it's possible to use the accretion model from astronomy to tell when a tech company accumulates enough value to be irredeemably evil. -
R relay@relay.publicsquare.global shared this topic
