Your Water, Your Power, Your Money — the series. The 2030 promise is not failing because the ambition was too large. It is failing because the architecture was never built to carry the load being placed on it. The build-out is marketed as if clean energy, nuclear restarts, small modular reactors, carbon commitments, and grid modernization will all arrive in time to absorb AI-scale demand. The verified record shows the opposite: the public story is organized around 2030, while the physical substrate remains years behind, structurally underbuilt, and increasingly dependent on behind-the-meter gas, extended-life fossil generation, and household-level burden. This is Ghost Load™ at grid scale.
In 60 Seconds
The build-out is marketed against a 2030 finish line. The institution holding the most-cited 2030 commitment — Microsoft — has, by its own published number, moved 23.4 percent away from its goal since the goal was made. The nuclear and clean-energy substrate promised to power the build-out covers approximately 2 to 3 percent of the actual forecast demand. The remaining 97 to 98 percent is absorbed by behind-the-meter natural gas, extended-life coal, and direct cost transfers to households who did not cause the load.
The Date Was Set by Policy, Not by Steel
Almost every promise about powering the AI build-out converges on 2030 — the year Microsoft says Three Mile Island feeds its data centers, Google says its first SMR comes online, corporate net-zero commitments come due, state roadmaps finish, and federal reliability horizons end. The convergence is not an accident: 2030 is far enough away to let the build-out continue uninterrupted and close enough to sound serious. The single cleanest piece of evidence that 2030 will not deliver: Microsoft pledged in 2020 to become carbon negative by 2030, and in its FY2024 sustainability report disclosed emissions have risen 23.4 percent above the 2020 baseline. Microsoft names the cause itself: data-center construction, embodied carbon in building materials, and AI hardware.
One Term You Need: Ghost Load™
Ghost Load™ is the gap between what an institution collects from you and what actually reaches the service you thought you were paying for. The formula is G = L − N — Total Load minus Necessary Load. Necessary Load is what it actually costs to deliver the thing; Ghost Load is everything above that line: administrative friction, financial extraction, capacity sold but not delivered, costs reassigned to people who did not cause them.
The Three Showpiece Projects
Three Mile Island — one reactor, maybe by 2027. Announced September 20, 2024: a $1.6 billion Constellation commitment, a $1 billion DOE loan, a 20-year Microsoft PPA, 835 megawatts, target accelerated from 2028 to 2027. Status as of late May 2026: no fuel loaded; work to date is hiring, inspections, procurement, and water-system restoration; the PJM transmission upgrades that would let the 835 MW actually reach the data centers are not yet built. A credible 2027–2031 delivery, not a current source of electricity — and it is one reactor.
Google + Kairos Power — demo reactors, commercial by 2030. The October 2024 agreement contemplates six to seven SMRs totaling 500 MW, first commercial deployment targeted for 2030. What is actually built or permitted: Hermes 1 (35 MW thermal, not power), Hermes 2 construction permits (November 2024), and a non-nuclear Engineering Test Unit (vessel installed July 2025). A 35 MWt reactor running as a power plant would deliver only about 12–15 MWe. Currently zero SMRs are connected to the U.S. grid.
Stargate — operational, but burning gas. The OpenAI/SoftBank/Oracle/MGX venture: a $500 billion four-year commitment, 10 GW target. As of late April 2026, the Abilene flagship runs ~0.3–0.6 GW; five other U.S. sites are not operational. The decision that quietly undoes the clean-energy narrative: at least three of the seven U.S. sites use on-site natural gas plants because the interconnection queue is too slow — reported as behind-the-meter generation rather than data-center fossil emissions.
The Water-to-Electricity Trade-Off
At least six of the seven Stargate sites are committed to closed-loop liquid cooling, which avoids the evaporative water draw that triggered most documented community water conflicts. It is real — and it is the only documented concession the public pressure of 2024–2025 has won. But it does not eliminate the extraction. Thermodynamically, heat cannot disappear; closed-loop cooling cuts water consumption but requires massive mechanical chillers, which spikes electricity demand. The community saves the aquifer by shifting the same extractive load onto the local substation.
Jevons Paradox
Next-generation GPUs are more efficient per calculation at the chip level — and that is a textbook case of the Jevons Paradox: efficiency gains tend to increase total resource consumption because the gains get spent on more capacity. Rack density has jumped from roughly 15 kW per rack to over 100 kW per rack in the AI-optimized generation. The efficiency gain is entirely absorbed into operator capability; the strain on the host community's grid increases exponentially.
What This Adds Up To
If every Stargate site delivers at full capacity by 2029, Three Mile Island restarts on time in 2027, Kairos delivers a first commercial SMR by 2030, and all the major nuclear deals deliver on schedule, the total dedicated low-carbon nuclear capacity coming online for U.S. data centers by 2030 is on the order of 1.5 to 2 gigawatts. The forecast new data-center demand for that same window is 70 to 90 gigawatts in PJM and ERCOT combined. The promised nuclear substrate covers approximately 2 to 3 percent of the demand it has been marketed to cover. The remaining 97 to 98 percent is the substitution gap — behind-the-meter natural gas, extended-life coal, and socialized transmission upgrades whose costs do not appear in the original announcement. The 2030 narrative requires the substitution gap never to be summed.