More than half of Texas drinks from the ground. The ground is running out. The state's plan is to run out slower. There is another way, and it has already been built.
Every November, when the irrigation pumps in the Belding farm belt southwest of Fort Stockton shut down for the winter, Comanche Springs comes back. Water rises in the old municipal pool. It runs through the cold months, and when the pumps restart in spring it stops.
In June of 1899 the spring flowed at 66 cubic feet per second, enough to irrigate six thousand acres and give a West Texas town its name as "Spring City"1. The decline began in May 1947, when large irrigation wells were drilled in the Belding area. By March 1961 the flow had ceased2. It has been on the pumps' schedule ever since: dead in the growing season, alive in the winter. A 2021 study from Texas State University put a figure on what year-round flow would take: hold pumping in the surrounding area to between 26,000 and 35,000 acre-feet a year and the spring runs at ten cubic feet per second every day3.
Fort Stockton is a small case of a large pattern. In 1975, a Texas Water Development Board geologist named Gunnar Brune published a survey of 281 major and historically significant springs in Texas. Sixty-three of them had already failed4. His explanation was that flowing artesian wells had released the pressure in the aquifers and pumping had then drawn water levels down, "as much as 700 feet" in places, so that streams which once gained water from the ground began losing it, and "thousands of small springs have dried up"5. When the Meadows Center at Texas State went back to Brune's list in 2024, the share of his springs that were dry had grown from eleven percent to thirty. Over the same fifty years the cumulative amount of groundwater pumped in Texas had nearly tripled, from roughly 250 million acre-feet to 700 million6.
Most Texans picture their water as a lake. For more than half the state it is a well. Texas used about fifteen million acre-feet of water in 2023, and fifty-six percent of it came from the ground7. San Antonio draws sixty percent of its supply from the Edwards Aquifer8. Lubbock is about sixty percent groundwater9. El Paso is more than half groundwater in a normal year and leans harder on it as the Rio Grande fails10. Bryan and College Station run entirely on the Carrizo-Wilcox11. Outside the cities there is usually no other source at all. The lifeline of Texas is underground, and it is going down.
The record is not in dispute. Under the High Plains, the Ogallala has fallen an average of forty-four feet since 1950; the single largest decline measured anywhere in the eight-state aquifer, 265 feet, is in Texas13. The U.S. Geological Survey puts the storage lost from the Texas portion at 169.6 million acre-feet14, more than five times the capacity of every reservoir in the state combined17. Texas pumps about 4.4 million acre-feet a year from the Ogallala, more than ninety percent of it for irrigation15, against recharge on the order of a million acre-feet a year16.
The oldest decline is under Dallas–Fort Worth. In 1890, a well drilled into the Trinity at Fort Worth would push water ninety to a hundred feet above the ground. The flowing wells stopped by 1914, and the water level in the Trinity under Fort Worth has since fallen about 850 feet18; along the I-35 corridor the regional planning group records declines of more than a thousand feet19. Houston pumped its own water table down three to four hundred feet, and the ground followed it: roughly ten feet of subsidence at Baytown by 197920, and the Katy area still sinking about an inch a year today21.
The most recent decline is in the Hill Country, and it is the fastest. Across thirty-five recording wells in the Trinity Aquifer between Kerrville and Wimberley, the median rate of decline over the past five years is 2.3 feet a year, with four to six feet a year in Kendall, Hays, and Kerr counties; twelve of those wells stood at record depth this summer22. In March 2026 the Hays Trinity Groundwater Conservation District reported every one of its monitored wells below its historic low and imposed a forty percent curtailment23. Jacob's Well at Wimberley, measured at six cubic feet a second in 1924, stopped flowing for the first time in recorded history in 2000, went dry again in 2008, 2011, 2013, 2022, and 2023, and has recorded zero flow in every year since, including six straight months from last October to this April24. And the Edwards Aquifer, the one two and a half million people drink from12, recharged in 2020 through 2024 at an average of 177,000 acre-feet a year. The drought of record, 1952 through 1956, the benchmark the whole Texas water system is designed around, averaged 168,20025. Comal Springs, the largest in the state, posted its lowest annual flow since 1956 last year26.
What happens under the ground shows up on top of it. Springs are where the water table reaches daylight, and the rivers of central and west Texas are fed by them; Brune's survey found streams that once gained water from the ground turning into streams that lose it5. The Nueces did not reach Corpus Christi Bay on 340 days last year63. The state has written inflow standards for its bays that say how much river the estuary needs to stay alive; for the Nueces the floor is about 116,000 acre-feet a year, and in a dry year it gets almost none82. The lifeline is not only the water under the state. It is the rivers that come out of it and the bays they run to.
On July 23, 2026, the Texas Water Development Board adopted the 2027 State Water Plan. It projects that the state's existing water supply will fall from 15.5 million acre-feet a year in 2030 to 14.0 million in 2080, "primarily due to depletion of aquifers." In a repeat of the drought of record the shortfall is 3.6 million acre-feet a year in 2030 and 5.8 million in 2080. The plan's cost is $174 billion. Seawater desalination is 3.7 percent of its answer. If the needs go unmet, the plan estimates the economic damage at $91 billion a year now and $177 billion a year by 208027. The regional plan for the Hill Country allows the Trinity to fall another thirty feet by 206028. The management rule on the High Plains is to keep half of the water that was there in 2010 for another fifty years29. Taken together, the water policy of Texas is to manage the decline of its groundwater so that it runs out more slowly.
That is a reasonable policy if there is no new water. There is. Texas has 367 miles of coast36 on the one supply that does not run out, and the water it is losing sits one hundred fifty to seven hundred miles inland. The state does not have a shortage of water. It has water in the wrong place.
Israel faced the same arithmetic and answered it. During a drought that ran from 1998 to 2002, with a shortfall of half a billion cubic meters in a single year, the government resolved to take its drinking water from the Mediterranean, tendered the first plant to private builders on a twenty-five-year contract in 2000, and in 2002 set a program of four hundred million cubic meters a year30. More than eighty percent of the water in Israel's cities now comes from the sea31. Its newest plant delivers water for about forty cents a cubic meter, under five hundred dollars an acre-foot34, a third of the sixteen hundred dollars an acre-foot San Antonio pays for the groundwater it pipes in from Burleson County35. Pumping from the Sea of Galilee, which passed five hundred million cubic meters in its heaviest years, fell to a few tens of millions a year32. And in October 2025, Israel began pumping desalinated seawater back up its national water carrier into the Sea of Galilee to refill it33.
What the Texas Aqueduct Project proposes puts the plants on the coast and, in time, on the water. The Gulf off Texas already holds the infrastructure of a century of offshore oil and gas, sixty-eight platforms in federal waters and some nineteen thousand miles of pipeline on the seafloor, with leases, surveyed shore crossings, and landfalls already permitted48. The world's only operating floating nuclear plant has been on the grid since the end of 201946, and in July 2026 the federal offshore regulator and the NRC signed an agreement to work out how to license reactors on the outer continental shelf47. Just as important, Texas law already favors this layout. Under the seawater desalination chapter the Legislature wrote in 2015, an intake more than three miles offshore needs no water-right permit and a brine outfall more than three miles offshore gets no contested-case hearing; discharge into a bay is barred on that track49. Corpus Christi's desalination plant, which the city council rejected in September 2025 and again on September 1, 2026, after the state had committed some $757 million to it, was a bay-discharge plant, and bay brine is what killed it50. Deep, well-mixed Gulf water, in and out, is the configuration the law was written for. The salt goes back where it came from: the concentrate, about twice as salty as seawater, is returned miles offshore through diffusers that mix it to near-ambient within a short distance, the way Israel's plants discharge to the Mediterranean and the way the Port of Corpus Christi designed its own offshore outfall80.
The Texas requirement can be stated in one number. For the water table to stop falling in every zone of the state, the water delivered plus the pumping retired has to exceed the drawdown. Summed across the eight major aquifer zones, that is about 4.74 million acre-feet a year: 1.55 trillion gallons, or 6,551 cubic feet per second flowing every hour of the year37, roughly three times the canal that carries the Colorado River across Arizona39. Seventy-two percent of that number is the Ogallala. The other seven zones together need about 440 billion gallons a year38, and every one of them is documented to recover when pumping stops.
Making the water is the settled part. Seawater reverse osmosis is a mature industry; the best plants run at three to three and a half kilowatt-hours per cubic meter, within a factor of three of the theoretical minimum40. What the plants need is firm power, and a single 300-megawatt reactor supplies enough for roughly half a million to seven hundred thousand acre-feet a year of desalination41, about what San Antonio, Corpus Christi, and the Rio Grande Valley together would draw. The whole requirement, if every gallon were made from the sea, is about 4,600 megawatts of continuous power, fifteen or sixteen such plants; the replace-and-retire path is under 900 megawatts, three81. The reactors are no longer theoretical. The first new commercial reactor in a generation went into construction in Wyoming in April 202642. Dow is building a four-unit plant on the Texas coast at Seadrift, with its final safety evaluation due in November43. The Nuclear Regulatory Commission finalized its new licensing framework in April and is under an executive order to decide new applications within eighteen months44. Texas created an advanced nuclear office last year with $350 million behind it45.
From the coast the water moves inland along corridors the state has already surveyed and condemned for its highways, transmission lines, railroads, and pipelines, and it goes to three kinds of places. It goes into reservoirs that sit empty with their dams and intakes intact: Twin Buttes at San Angelo was six percent full this month, Choke Canyon twenty-three percent, and Medina Lake fell to two percent last year51. It goes into the creek beds over the Edwards recharge zone, where rivers sink into the aquifer and come out at Comal and San Marcos. And it goes directly into the aquifers through injection wells, which Texas already does: San Antonio has about 155,000 acre-feet of Edwards water banked underground in southern Bexar County52, and El Paso has returned more than 92,000 acre-feet of treated water to its aquifer since 198553.
Where the water arrives, the pumping stops. Cities and industry switch to delivered supply. Farmers who choose to are paid to retire irrigation wells, on the model the Edwards Aquifer Authority already runs, which pays irrigators to leave water in the ground during droughts and leases pumping rights for storage54, and on the model Arizona adopted in 1980 when the federal government made groundwater reform the condition of its canal55. Every household well and stock tank in Texas stays as it is. State law already exempts them from district permitting56, and the purpose of the project is that they keep working.
Texas has already run this experiment twice, on its two biggest declines. Houston pumped the ground under it down three to four hundred feet and sank; when the Subsidence District pushed the city's core onto surface water, the water level under east-central Harris County rose 215 feet and the sinking in the old core stopped by the early 1990s77. North of Dallas, when Euless, Bedford and Arlington left the Paluxy for lake water in the 1970s, the levels came back, and a Woodbine well in Grayson County has risen forty-six feet since 199978. The aquifers that are still falling in those regions are the ones under the suburbs that have not switched yet. The ground does not need persuading. It needs the pumps turned off, and something else to drink from.
The costs are large and they should be stated. Replacing every gallon of overdraft in Texas with desalinated water would cost on the order of $200 to $430 billion. Replacing the water for cities and industry and paying farmers to retire the irrigation pumping instead comes to about $56 to $79 billion in capital plus roughly $1.4 billion a year in retirement payments, and produces the same water tables57. Delivered to San Antonio, Gulf water at today's first-of-a-kind reactor prices works out to roughly $2,700 to $3,900 an acre-foot with the plant and the pipe paid off over forty years, against the $2,200 the city already pays all-in for Vista Ridge; the running cost alone, power and upkeep, is about $70079. The gap is the price of building the first plant, and it is exactly the gap that Proposition 4 money and the AI companies' power contracts exist to close. First-of-a-kind reactors are expensive; Ontario's fleet is running about C$17,000 a kilowatt and the Wyoming plant about $11,600 with half federal funding58. The Ogallala will not refill in a human lifetime; at any surplus Texas could plausibly deliver, restoring 170 million acre-feet is a century-scale project, and the goal there is to stop the fall59. Lifting water 3,200 feet onto the High Plains takes more energy than making it60. Those facts set the order of construction: the coast and the Balcones cities first, where the lift is a fifth of the desalination energy and the buyers already pay Vista Ridge prices; the Hill Country and the Highland Lakes next, where the springs are; the High Plains last, by storage and retirement rather than by pipe.
On the High Plains that is the hard part of the proposal, and it should be said plainly. More than ninety percent of Ogallala pumping is irrigation15, and no pipe from the coast will ever water it. What the proposal offers there is a choice a farmer does not have today: to be paid, at the rate the federal government already pays on the Colorado River76, to retire pumping early, while the district's own rule is to run the aquifer down to half of what was there in 2010 and stop when it is gone29. Nobody is told to quit. Everybody is told the truth about the gauge, and offered a price.
For scale, the state's own plan is $174 billion and manages decline27.
Put on a calendar, the proposal runs like this. A pilot plant on the Coastal Bend, taking Gulf water offshore and putting it into Choke Canyon and Lake Corpus Christi, delivers first water by 2030. The first reactor-and-desalination complex on the coast and the trunk line to San Antonio, the Edwards recharge zone, and Uvalde are running by the mid-2030s, and that is when the Balcones springs begin to answer. The Hill Country, the Highland Lakes, and the I-35 corridor to Dallas–Fort Worth follow by the early 2040s. The High Plains come last, by 2050, through stored water and paid retirement rather than a pipe to every field. Each stage sells water to pay for the next. By 2050 every city and every industry in Texas is drinking from the Gulf, and the only wells still pumping are the ones on private land, for the people who live there.
The Texas Aqueduct Project is not a company, and it does not propose to build any of this itself. It is a case for a partnership that Texas is unusually placed to make. The energy is built for the data centers; the water and the retired pumping come back to the communities that host them. The companies building artificial intelligence are asking the state for more firm electricity than it has ever added at once: ERCOT's queue of large new loads passed 400 gigawatts in April 2026, and nearly ninety percent of it is data centers66; OpenAI's Stargate alone has more than four gigawatts planned in the state, at Abilene and in Shackelford and Milam counties70. Those facilities use far less water than power. Texas data centers draw about 25 billion gallons a year today and could reach 161 billion by 203067, a real number but a small one beside the state's fifteen million acre-feet. Five gigawatts of data centers need roughly 30,000 to 64,000 acre-feet of water a year but about 39 terawatt-hours of electricity, nine times what a million acre-feet of desalination consumes68. The companies have already shown they will sign contracts running to 2042 and beyond for firm nuclear power69. The proposal is to put that power on the coast, where the same reactors that serve a data center can make water, and to write into every contract a water reservation and a charge of a dollar or two per megawatt-hour that pays to retire groundwater pumping. On 39 terawatt-hours that is $39 to $79 million a year, enough to retire 100,000 to 200,000 acre-feet of pumping annually at the rate the federal government already pays on the Colorado River76.
No single company can do this, and none should be expected to. A network that spans eight aquifers and the length of the coast is the kind of work a government exists for: not to build it, but to set the strategic plan, hold the regulatory backstop, and bring capital-driven companies into a market for water that pays them to deliver it. The country has done this before, and each time it put the United States ahead of the world. The transcontinental railroad was built by private companies on federal land grants and government bonds that were repaid with interest71. The Manhattan Project spent about two billion dollars in three years72. NASA's budget reached 4.4 percent of federal spending in 1966 on the way to the moon73. Closer to this problem, Hoover Dam was repaid from power sales by 198774, and Arizona is still repaying the federal government for the canal that saved Phoenix and Tucson's groundwater75. In every case the government set the goal and backed the risk, and companies did the work and were paid by what they delivered.
The reason to start now is that the first installment is already paid. In November 2025, seventy percent of Texas voters approved Proposition 4, which dedicates a billion dollars a year to the Texas Water Fund through 2047 and requires at least half of it to go to new supply. The amendment names marine desalination and "infrastructure for water transport and integration" as eligible uses, and it bars the fund from financing pipelines that move fresh groundwater from one part of Texas to another61. The reactors are in the ground. The Edwards Aquifer Authority has shown since 1993 that a pumping cap backed by paid forbearance and a substitute supply holds through a drought of record62.
What the state gets back can be measured, and it should be measured at the well. A San Antonio summer in a 1950s-class drought without a Stage 5 cut, the forty-four percent reduction the city reached in May 202564. A Katy that stops sinking21. A well in Kendall County that never has to be deepened, because the Trinity under it is held at zero net decline instead of planned down thirty feet28. Rice under water on the Matagorda canals in a year when the Highland Lakes would otherwise have cut it off65. The Nueces reaching Corpus Christi Bay, which it failed to do on 340 days in 202563. Comanche Springs in July3. The water table, read against the level it held in 1950, is the scoreboard.
The numbers behind this article, the sources for every one of them, and the water levels in all eight aquifer zones as of this month are on this site. If you run a groundwater district, a utility, or an agency that measures one of these aquifers, we would like to hear from you, especially where you think a figure is wrong. If you serve in the Legislature, the ask is one sentence: measure Texas water at the well, and certify replacement supply before anyone is asked to pump less. And if you are a Texan who turns on a tap, find out where your water comes from, and put your name on this. Then go to Fort Stockton in November and watch what happens when the pumps turn off.