AI, Data Centers, and Indigenous Water Rights at Bear Butte
Indigenous speakers at World Peace and Prayer Day at Bear Butte raised concerns about AI and data centers consuming Native data, land, and water. They highlighted issues with data sovereignty, generative AI's "pan-Indianism," and the significant local water strain caused by data centers, particularly in places like Loudoun County, Virginia.
Last updated 25 June 2026
AI, Water & Data Sovereignty
Checking the AI water and data warnings from Bear Butte
At World Peace and Prayer Day at Mathó Pahá (Bear Butte), South Dakota, on 20–21 June 2026, Indigenous speakers warned that artificial intelligence and its data centers are taking Native data, land and water. Most of what they said tracks the public record: the data-control concern is real, generative AI does flatten distinct traditions into a generic image, and data centers do draw heavily on water in places already short of it. Two points need sharpening — the scale of national water use, and how the technology actually works.
TL;DR
- Joseph White Eyes (Cheyenne River Lakota) said data entered into AI search and chat tools stops being the user’s own. Provider policies vary, but prompts are routinely logged and may be used for training unless a user opts out.
- His “pan-Indianism” point holds: generative image models are trained on scraped web data and return averaged, composite depictions that blur culturally distinct ceremonies.
- US data centers directly consumed about 17.4 billion gallons of water in 2023; a federal projection puts that at 38–73 billion gallons by 2028.
- That national total is small against US agriculture’s roughly 26.4 trillion gallons (2024). The strain is local: Loudoun County, Virginia data centers used about 10% of the county’s water in 2023.
- The speakers’ firmest claim is the least covered: data centers are being sited on and near tribal land without proper consultation.
“The information no longer belongs to us”
Speaking on the harm of data collection and AI, White Eyes described what happens when people use AI companies’ search engines and chat boxes. As documented by Censored News, he said: “The moment we engage with them, the information no longer belongs to us.”
“The moment we engage with them, the information no longer belongs to us.”
The direction is right; the absolute is too strong. Prompts typed into consumer AI tools are logged, and default settings often permit their use for model training. But handling depends on the product: many providers offer training opt-outs, conversation deletion, and enterprise or API tiers with no-retention terms. “No longer belongs to us” overstates a real loss of control rather than a total forfeiture.
The underlying concern has a name in the research literature — Indigenous data sovereignty, the principle that Native nations should govern data about their people, lands and knowledge. That question of governance is not settled by a vendor’s privacy toggle, and it is the part of White Eyes’ point that survives intact.
Generative AI and “pan-Indianism”
White Eyes used a concrete example: asking a generative model for an image of a Sundance. He described the system assembling its idea of the ceremony from whatever it has scraped — “it builds that model based off of what they look at” — and the result, he said, is that “we start seeing things like pan-indianism show up on Facebook.”
On the substance, he is largely correct. Generative image models are trained on large, scraped web datasets and produce composite, averaged outputs. Asked for a culturally specific ceremony, they tend to return a generic “pan-Indian” pastiche that erases the differences between nations — a well-documented failure mode of these systems.
One mechanical clarification. A model does not browse the web at the moment of a prompt unless it is given a search tool; the scraping and averaging happen during training, and the homogenization is baked into the model’s weights before any user types anything. What surfaces in a Facebook feed is a separate recommendation system. Those are wiring details. The core claim — that the technology flattens distinct traditions into one image — stands.
The water question: big numbers, local pain
The Censored News account, echoing White Eyes and Paiute speaker Josh Dini — who tied the concern to an active drought — says data centers use enormous amounts of water. Directionally, yes. The scale needs context.
US data centers directly consumed about 17.4 billion gallons of water in 2023, per a 2024 Lawrence Berkeley National Laboratory federal report, which projects 38–73 billion gallons by 2028 ▸ Projection. Set against US agriculture’s roughly 26.4 trillion gallons in 2024, direct data center use is a rounding error nationally — on the order of 0.07% ● Derived (17.4 billion ÷ 26.4 trillion; note the one-year gap).
| Category | Water | Year | Source |
|---|---|---|---|
| US data centers (direct consumption) | ~17.4B gal | 2023 | LBNL / federal |
| US data centers (projected direct) | 38–73B gal | 2028 | LBNL (projection) |
| Loudoun County, VA data centers | ~1.6B gal (~10% of county) | 2023 | Reported |
| US agriculture | ~26.4T gal | 2024 | USDA |
The honest version of the speakers’ point is narrower and stronger than “enormous.” The harm is not aggregate volume; it is concentration and timing. Where data centers cluster, they can take a large share of a single community’s supply — about 10% of Loudoun County, Virginia’s water in 2023 — and they do it during drought, against farms and households. That is the case the data supports.
Pollution and minerals
The Censored News summary adds that data companies are polluting the land, water and air and driving demand for more critical minerals. Two distinctions matter here.
A data center is servers and cooling, not a smokestack; its direct emissions are minimal. The footprint is mostly upstream — grid electricity, where one peer-reviewed estimate puts added US AI-server emissions at 24–44 million metric tons of CO₂e per year by 2030 ▸ Projection — plus chip fabrication and warmer, saltier water returned to local systems. Attributing the pollution to the building understates how much of it sits in the supply chain.
The critical-minerals claim is directionally true: chips, copper and grid buildout all scale with the AI boom. The heaviest mineral demand, though, comes from the power and transmission infrastructure feeding the centers more than from the processors themselves.
Where the speakers are on firmest ground
The strongest claim at Bear Butte is not a number. Paiute speaker Josh Dini said AI data centers are being pushed without proper consultation with the tribes, even as a drought deepens. That is a governance question, not a contested statistic, and it is the part mainstream AI coverage most often skips.
Siting decisions for water- and power-hungry facilities increasingly land on or beside tribal land and shared watersheds. Whether tribes get free, prior and informed consent on those projects is a live policy issue — and on that, the speakers are describing a real gap, not overstating one.
Water and energy figures and projections change as new facilities come online and reporting improves. Verify against the cited primary sources before relying on any single number. Figures labeled projection are forecasts; the derived ratio is calculated here from the two sourced totals.
Sources
- Brenda Norrell, “World Peace and Prayer Day: The Power of Spirit,” Censored News, 23 June 2026. bsnorrell.blogspot.com
- MOST Policy Initiative, “Data Center Water Use,” April 2026 (citing the 2024 Lawrence Berkeley National Laboratory federal report). mostpolicyinitiative.org
- WaterVerge, “Data Centers and AI Water Consumption,” May 2026 (Loudoun County figures). waterverge.com
- Tom’s Hardware, AI water use analysis, June 2026 (USDA 2024 agriculture water figure). tomshardware.com
- “The Environmental Impact of AI Servers and Sustainable Solutions,” arXiv preprint, 2026 (CO₂e projection). arxiv.org
Last updated 25 June 2026
Comments
Loading comments…