Water policy is moving toward a more precise question: not simply whether a data center uses water, but how much, for what equipment, and who has to disclose it.
Texas and Rhode Island offered two different answers in 2026. Texas put a water-use study into the state budget. Rhode Island considered bills that would have required large data centers to report operational water details directly. The distinction matters because a study, a proposed mandate, and live per-inference attribution are three different layers of transparency.
Texas: Rider 6 makes measurement a state assignment
Texas Rider 6 directs the Public Utility Commission of Texas (PUCT) to collect information for a study of water use by data centers and virtual currency mining facilities. The work is organized through PUCT Project No. 59281, including the commission's energy and water use survey guidance.
That is meaningful infrastructure for policymaking. A state cannot plan water supply, transmission, or future industrial load very well if it has no consistent view of which facilities are drawing water and how their cooling systems work. Rider 6 starts by building that view across a fast-changing class of large energy users.
But Rider 6 should not be confused with a per-request disclosure rule. It is a study and data-collection assignment: the state is assembling a baseline for planning and oversight. It does not, by itself, put a live water number in the response to every inference, nor does the rider turn a facility-wide estimate into a request-level measurement.
The Texas approach is therefore upstream. First establish a comparable inventory and a public-policy baseline; then decide which reporting, siting, or conservation requirements the evidence supports.
Rhode Island: reporting rules proposed, then stopped in committee
Rhode Island's proposals took a more direct facility-reporting route. Senate Bill 2776 and House Bill 7331 would have applied to large data centers at or above a 50-megawatt threshold.
The bills proposed annual reporting to the Rhode Island Department of Environmental Management. The information would have included average daily and peak daily water withdrawal, the cooling technology in use, and the facility's recycling or reuse practices. They also would have allowed the agency to require a water-efficiency, conservation, or recycling plan when the facts warranted one, alongside safeguards related to future infrastructure and site restoration.
That is a different policy instrument from Texas Rider 6. Instead of asking the state to assemble the first statewide dataset, the Rhode Island bills would have put a recurring reporting obligation on qualifying operators and given the environmental agency a role in reviewing the results.
The outcome needs a precise description. The Senate measure was referred to committee and the House measure followed the same committee path. Both bills died in committee after being held for further study. Neither reached the governor, so neither was ever vetoed. The official Rhode Island bill-status portal and the Senate committee vote record are the relevant status trail; the bill-text pages above are the primary source for what the proposals actually said.
The comparison
| | Texas Rider 6 | Rhode Island S2776 / H7331 | | --- | --- | --- | | Policy shape | State-led study and data collection | Proposed operator reporting requirement | | Who supplies the first data | PUCT gathers a statewide picture of data center and virtual currency mining water use | Qualifying data centers would report operational water details to the environmental agency | | What the record is meant to support | Planning, oversight, and a more consistent baseline | Recurring facility disclosure plus possible efficiency or conservation plans | | Status in 2026 | A live state data-collection effort under Rider 6 | Both proposals died in committee and were never vetoed | | What it does not provide | Per-inference water accounting | An enacted mandate or live request-level measurement |
The comparison also shows why the word water use can hide several different measurements. A facility may report total withdrawal. A regulator may build a regional planning estimate. An operator may disclose cooling technology and reuse. None of those figures automatically tells a customer how much water a single workload was attributed on a particular device.
CarbonLayer's transparency takeaway
CarbonLayer's view is that these layers should connect, not compete. Public policy needs facility-level facts. Communities need understandable reporting about local water demand. Buyers also need a number close to the work they are commissioning.
That is why CarbonLayer publishes water fields on every successful POST /api/inference response: waterMl, savedWaterMl, baselineWaterMl, and the device's waterIntensityMlPerKwh. The per-inference water transparency write-up explains the accounting and the boundary between measured site coefficients and modeled per-request attribution. Fleet roll-ups are available through /api/savings, with regional device context in /api/network.
The labels are as important as the figures. A per-inference slice is not a facility flow meter. A facility coefficient is not a claim that every neighboring site has the same cooling system. Naming those boundaries makes the number useful instead of making it sound more certain than it is.
Texas Rider 6 is a case for building the public baseline. Rhode Island's S2776 and H7331 were a case for requiring recurring facility reports, even though both died in committee. CarbonLayer's contribution is the request-level layer: show the water attribution alongside the inference, preserve the measured-versus-modeled distinction, and let the next policy conversation start with numbers people can inspect.