Net positive. From day one.
AI infrastructure has real environmental costs. Based on our own research, covering them is inexpensive: a small fraction of what we spend on compute itself. So 110% coverage isn't a stretch goal or a someday promise. We pledge to cover it from day one, saving toward it as we go and refining the accounting, savings, and spending for offsets and research as our understanding improves.
The commitment is to offset 110% of your usage's impact, on whatever infrastructure it runs on, not a claim tied to one vendor. We do the best accounting the public data allows, and offset the rest. AWS is where we run today, and their published data lets us be as precise as possible. As we add other providers, we will extend the same discipline, even where the data is less complete.
Energy: already covered. We add more.
AWS achieved 100% renewable electricity matching in 2023, seven years ahead of their 2030 goal. By running on AWS, 100% renewable-matched energy is built into our cost of doing business. We add 10% additional verified carbon offsets on top through Gold Standard or Verra VCS certified projects. For every ton of CO2 attributable to our infrastructure, we fund the removal of 1.10 tons. More importantly, we also commit to funding solutions to reduce electric strain on local power grids.
100%
AWS renewable energy
+ 10%
Comitas carbon offsets
= 110%
net positive
Water: we close the gap today, not in 2030.
Data centers consume water for cooling. AWS has committed to being water-positive by 2030, but they're not there yet. As of 2026, AWS is 75% of the way to their water-positive goal, up from 53% reported in 2024. We cover the remaining gap plus 10% more through water restoration partnerships in the same watersheds where our infrastructure operates. Here, too, we commit to funding research and solutions to use less water in datacenters to begin with.
Where the money goes
We're looking for water restoration projects that add water back to the watersheds where our infrastructure operates, not programs that simply pay local water users to consume less. For our primary region (Oregon) and as we expand to others, we vet each partner against that standard before funding anything.
We haven't finalized our own funding agreements with specific partners yet. As we do, region by region, we'll name them here, along with what we've funded and the certificates behind it.
How we measure and report
Carbon: the AWS Customer Carbon Footprint Tool provides our CO2e emissions per region. We read the number and buy 10% more in verified offsets. No estimation required. We also commit funding solutions that reduce strain on local power grids.
Water: AWS publishes Water Usage Effectiveness (WUE) per region: liters of water consumed per kilowatt-hour of IT load. We multiply our energy consumption (from the Carbon Footprint Tool) by the regional WUE to calculate our water footprint. Then we restore the gap plus 10%. We also commit to funding solutions to reduce water consumption in datacenters.
Frequency: quarterly, matching the cadence of AWS's reporting data.
Disclosure: emissions, offset purchases, water restoration funding, partner names, and verification certificates, published quarterly on this page.
Per-user impact: each customer will see their own running total: the actual dollars they've paid that went toward offsets and the mission, accumulated since they joined.
Margin for error: our own research shows offsetting AI's environmental impact costs a small fraction of what we spend on compute. That's exactly why we pledge 110%. The extra 10% is a deliberate overestimate, built in from the start to make under-coverage unlikely.
Model providers
As we add model providers beyond AWS, we extend this commitment to cover those workloads. We can't measure what providers don't publish, but we research and report what's available: each provider's stated environmental commitments, the verified results where they exist, and the gap where they don't. Comparison creates pressure. We use it.
| Provider | Public commitment | Target | What's actually been reported |
|---|---|---|---|
| AWS | Net-zero carbon (The Climate Pledge) | 2040 | Renewable-matched since 2023; total emissions rose 16% in 2025; separately, 75% of the way to its 2030 water-positive goal , source |
| Microsoft | Carbon negative, water positive | 2030 | Water positive achieved 2025; emissions rose 25% in FY25 , source |
| Google Cloud | 24/7 carbon-free energy, every hour | 2030 | Some sites near ~90%; company states AI buildout is outpacing grid decarbonization , source |
| Oracle Cloud | 100% renewable electricity; net-zero | 2030 / 2050 | 92% renewable coverage on cloud infrastructure , source |
| Meta | Net-zero value chain; water positive | 2030 | 30+ water restoration projects; some individual sites report high local water draw , source |
| OpenAI | No public commitment found | — | No corporate climate disclosure published , source |
| Anthropic | No full sustainability report found | — | Joined a $915M carbon-removal coalition; hasn't published its own footprint , source |
| xAI | No public commitment found | — | Documented through litigation over unpermitted gas turbines at its Memphis-area data center, not through any sustainability disclosure of its own , source |
As of August 30, 2026. Compiled from each provider's own published sustainability reporting and, where noted, independent reporting. Not a claim about intent, only about what each company has and hasn't disclosed. We'll keep this updated as providers publish new data.
The bottom line
The extra 10% ensures that every user will be a net positive for the environment. The platform fee is tuned over time for sustainable growth and continued public value, not to maximize what gets extracted from it.
Put your usage on the right side of this pledge.
Every workload you run becomes an environmental net positive.
Request access →