Edition 017: The Cost
The cost of the data economy is not only the loss of autonomy and power that previous editions have described. It is also more systemic, involving the land, water, energy, and labor taken from places and people without the power to refuse.
In her book, Atlas of AI, Kate Crawford traces the material impacts of AI across the world, arguing that AI is, in its deepest material reality, a technology of extraction— minerals from the earth, labor from the people processing the data, and the data itself.[1] Author Karen Hao has shown that the largest AI companies operate by a logic indistinguishable from colonial empires, taking land and water and energy and labor from places without the political power to refuse and concentrating the benefit in places that already have everything.[2] Emily Bender and Timnit Gebru have warned that the environmental and financial costs of these systems fall disproportionately on communities that never benefit from them.[3]
All of these arguments are correct. What the arguments cannot do, because no argument can, is look at your life for you. That is what each of us must do—to turn and see where the extraction touches the ground we stand on. The arguments need each of us to recognize ourselves inside them.
***
A data center draws its electricity from a grid built for the people who used to live nearby, and its water from a watershed those people depended on. The cooling towers evaporate that water into the sky and most of it does not come back. The servers inside run AI models trained on the unconsented work of writers, artists, and ordinary people who put their lives online before anyone explained that those lives could be plundered for training data. The minerals inside those servers were mined by people whose names and conditions do not appear in any document the corporation is required to file. The carbon released to power the operation is borne by every community.
And the cost is not only in building the facility that houses the servers. It’s running the AI that runs on those servers. The largest AI companies will not say how much energy a single query to one of their models consumes, or which data center it is routed to, or what the carbon intensity of the grid serving that data center is at the moment the request is processed.[4] Eighty to ninety percent of the computing power devoted to AI is used not in training a model but in running it, day after day.[5] The Lawrence Berkeley National Laboratory, the federally funded research institution responsible for projecting US energy demand, has said publicly that the information the companies are willing to disclose is not enough to project what the AI revolution will require of the grid, or what it will emit into the atmosphere.[6] The AI corporations guard that information as trade secrets. But the system being built will reshape the energy supply of the country and the world at large. And the people who bear the cost will go on bearing it long after the buildings have been replaced by whatever comes next.
None of these costs appears in the deal signed. No law requires a disclosure about the energy being consumed. The financial filings cover quarterly earnings. The AI model card, if there is one, covers benchmark performance. There is no document in the entire regulatory structure of this country that requires the corporation to account for what it is taking from a place and people to build what it is building.
The system was built to see certain things and to be blind to others. What it sees, it counts. What it does not see, it does not owe. The fact that something cannot be priced does not mean it has no value. It means that the system has declined to value it.
***
Texas is an example of that pattern. The state is on pace to host the largest concentration of data center capacity in the world by 2030.[7] The state is losing roughly a billion dollars a year on the sales tax exemption that attracts the developers, and additional revenue beyond that through layered property tax abatements that the state backfills from general revenue.[8] ERCOT is forecasting electricity load growth that will require grid expansion paid for by residents, not by the hyperscalers. ERCOT projects a 79% projected price hike by 2027 for household electricity bills.[9] A quarter of the state is in extreme or exceptional drought, and the cooling water being drawn down is being drawn from the same sources the ranches and the towns depend on.[10]
By the number, this is success. Capital investment is up. Construction employment is up. The Comptroller can point to the gross product of various counties and show that the line has moved. What the line does not show is what the people in those counties are being asked to pay. Texas has the highest uninsured rate in the country, for adults and for children, and roughly one in four uninsured children in the United States is a Texan.[11] The Commonwealth Fund ranks the Texas health system 50th out of 50.[12] One in five Texas children, roughly 1.7 million of them, lives in a household that cannot reliably afford enough food.[13] The state that is on pace to host the largest concentration of data center capacity in the world is the same state that cannot reliably feed its children, insure them, or provide them with reliable healthcare.
The number registers the new construction and the Texas business boom. But the number does not register the people sleeping on the street or the children who went to school hungry that morning. The cost is externalized on the household electricity bill, on the graded field, in the depleted aquifers, and on the people the metrics were never built to see.
***
This is not an argument that every data center is a moral catastrophe, or that no one should build the infrastructure the digital economy requires. I am arguing something narrower and harder. When an institution takes something from a place, whether it is land, water, labor, or data, it owes that place and its people more than the price it paid at the closing table. The price at the closing table reflects only what the seller could bargain for under the conditions the buyer arranged. It does not reflect what is being lost. The aquifer is not being valued at what it is worth to the next generation of Texans who will need it. The grid is not being valued at what its expansion will cost the households whose bills will rise to pay for it, or at what that expansion will cost the environment, or at what the damage to the environment will cost the people who live in it. The state is not being valued at what it can no longer do for the children inside its borders. The price was set before the cost was known, and the cost will be known only after it has been paid.
The extraction continues. That is the sentence I keep thinking about and coming back to. From the extraction of human data to the extraction of human labor to the extraction of the earth itself, the cost is being borne by places and people without the power to refuse it. In Texas, the cost is borne by the taxpayers, monetarily and on the invisible side of the ledger. What the state counts as a boom is what the people inside it are paying for.
***
[1] Kate Crawford, Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence (New Haven: Yale University Press, 2021). https://yalebooks.yale.edu/book/9780300264630/atlas-of-ai/.
[2] Karen Hao, Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI (New York: Penguin Press, 2025). https://www.penguinrandomhouse.com/books/743569/empire-of-ai-by-karen-hao/
[3] Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell, "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜," in Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT '21), 610–623 (New York: Association for Computing Machinery, 2021). https://doi.org/10.1145/3442188.3445922
[4] James O'Donnell and Casey Crownhart. “We Did the Math on AI’s Energy Footprint. Here’s the Story You Haven't Heard.” MIT Technology Review, 20 May 2025, www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/. Accessed 19 May 2026.
[5]See n. 4.
[6]See n. 4.
[7] “Texas Data Center Market Is on Pace to Become the World's Largest by 2030.” The Texas Land Agent, 15 Apr. 2026, https://thetexaslandagent.com/texas-data-center-market-buildout-2030/. Accessed 19 May 2026.
[8] Paul Cobler and Apurva Mahajan. "Texas Losing a Billion Dollars a Year on Data Center Tax Break." The Texas Tribune, 8 Apr. 2026, https://www.texastribune.org/2026/04/08/texas-data-centers-sales-tax-break-billion-dollars/. Accessed 19 May 2026.
[9] “ERCOT Warns of Explosive Load Growth Driven by Data Centers.” Texas Scorecard, 16 Apr. 2026, https://texasscorecard.com/state/ercot-warns-of-explosive-load-growth-driven-by-data-centers/. Accessed 19 May 2026; Diana DiGangi, “Data Center Demand Spike Could Drive 79% ERCOT Price Hike in 2027: EIA.” Utility Dive, 16 Mar. 2026, https://www.utilitydive.com/news/data-center-demand-spike-could-drive-79-ercot-price-hike-in-2027-eia/814804/. Accessed 19 May 2026.
[10] Anna Y. Monroe, “Recent Report Shows Data Centers May Negatively Impact Texas' Water Supply.” Houston Public Media, 2 Mar. 2026, https://www.houstonpublicmedia.org/articles/technology/2026/03/02/544809/recent-report-shows-data-centers-may-negatively-impact-texas-water-supply/. Accessed 19 May 2026; "When the AI Cloud Comes for Texas Water." Texas Observer, Apr. 2026, https://www.texasobserver.org/texas-legislature-data-center-boom-water/. Accessed 19 May 2026.
[11] “Texas Has the Worst Uninsured Rate in the US Once Again — and Policymakers Hold the Key to Fix It.” Texans Care for Children, 25 Sept. 2025, https://txchildren.org/texas-has-the-worst-uninsured-rate-in-the-us-once-again-and-policymakers-hold-the-key-to-fix-it/. Accessed 19 May 2026.
[12] “Texas.” Commonwealth Fund, 2025 Scorecard on State Health System Performance, https://www.commonwealthfund.org/datacenter/texas. Accessed 19 May 2026.
[13] "Hunger in Texas." Feeding Texas, feedingtexas.org/learn-about-hunger/hunger-in-texas/. Accessed 21 May 2026.
