AI vs Carbon Efficiency: The Hyperscaler Gap

AI is scaling fast. Carbon efficiency isn’t keeping up.

We’ve analysed 2,617 IT companies in the Earthmark database, and one metric is starting to matter in both earnings and ESG conversations: carbon intensity (tCO2e/$1M revenue).

𝗪𝗵𝗮𝘁’𝘀 𝗵𝗮𝗽𝗽𝗲𝗻𝗶𝗻𝗴

  • IT sector emissions: +147% YoY (2.85M → 7.05M tCO2e).
  • Primary driver: hyperscale AI training + data centre buildout.
  • Signal: revenue growth is outpacing carbon efficiency gains.

We've processed the latest reports and disclosure to focus on how this looks with some of the Big Tech players.

𝗧𝗵𝗲 𝗵𝘆𝗽𝗲𝗿𝘀𝗰𝗮𝗹𝗲 𝗽𝗿𝗲𝗺𝗶𝘂𝗺

• 𝗔𝗺𝗮𝘇𝗼𝗻: 112.8tCO2e/$1 M revenue. Leads total volume with 80.8M tCO2e in emissions, reflecting the dual impact of expanding AWS AI infrastructure and global logistics fulfilment (8.7 × sector benchmark).
  • 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁: 72.0tCO2e/$1M — ~5.6× above sector weighted avg (12.9).
  • 𝗚𝗼𝗼𝗴𝗹𝗲: 46.8 tCO2e/$1M — energy +34.5% YoY (44M MWh); emissions at 18.8M tCO2e.
  • Takeaway: AI infrastructure = structurally higher carbon intensity (for now).

𝗥𝗲𝘃𝗲𝗻𝘂𝗲 𝘃𝘀 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆

  • 𝗡𝗩𝗜𝗗𝗜𝗔: 32.1 tCO2e/$1M. ~99.8% Scope 3; revenue growth masks footprint.
  • 𝗔𝗽𝗽𝗹𝗲: 36.8 tCO2e/$1M. Strong revenue efficiency + 96.5% recycling rate.
  • Takeaway: intensity can look “healthy” while absolute emissions surge.

𝗧𝗵𝗲 𝗿𝗲𝗮𝗹 𝗰𝗼𝗻𝘀𝘁𝗿𝗮𝗶𝗻𝘁: 𝗦𝗰𝗼𝗽𝗲 𝟯

  • 85–99.8% of emissions sit in supply chains (chips, GPUs, steel, concrete).
  • This is where most AI-related carbon sits—and where visibility is weakest.

𝗪𝗵𝗮𝘁 𝘁𝗵𝗶𝘀 𝗺𝗲𝗮𝗻𝘀 (𝗰𝗼𝗺𝗺𝗲𝗿𝗰𝗶𝗮𝗹𝗹𝘆)

  • Vendor selection: carbon intensity is becoming a procurement KPI, not just disclosure.
  • Risk: “100% renewable” claims don’t reflect real-time AI workloads or embodied carbon.
  • Opportunity: 24/7 carbon-aware compute + circular hardware = differentiation.