The Three-Year Window: Powering AI Without Locking In Gas

The decision was made on a Tuesday at four in the afternoon, in a conference room on the forty-second floor of a glass tower in San Francisco. The young analyst had laid two binders on the long table. One was three inches thick, labeled SOLAR + STORAGE. The other was less than half an inch, labeled COMBINED CYCLE GAS. The thick binder contained the interconnection queue position, the four-year transmission study schedule, the easement negotiations across six counties, the renewable energy credit assumptions, and the long-duration battery procurement letter of intent. The thin binder contained a turbine delivery slot from a major OEM and a commercial operation date of Q3 2028.

The managing director glanced at both. He looked at the clock above the door. The clock said 4:07. The contract with the AI customer required first power by the end of 2028. The analyst already knew, by the way the managing director was looking at the clock, what was about to happen.

“Run the thin one,” he said.

She nodded. She did not say what she was thinking, which was that the thick binder had better numbers on every page that mattered — cost, emissions, water, jobs over twenty years — and that the only thing the thin binder had going for it was a calendar.

---

Powering AI data centers has become the largest energy investment decision of the decade, and that decision is being made under a brutal time constraint that is driving more than a hundred billion dollars of new capital toward natural gas turbines. The Stratos Project in Utah is the most visible example, but it is not the exception. It is the pattern.

As of Early July 2026, but Changing Fast

The three-year window has become even clearer—and more consequential.

Duke Energy has agreed to terminate its Carolina Long Bay offshore-wind lease through a settlement with the U.S. Department of the Interior. Under the agreement, nearly $129 million will be reinvested in additional electric-power capacity in the Carolinas. Duke has identified grid infrastructure, nuclear generation, and natural gas among the possible destinations for that capital.

The Duke agreement follows larger federal settlements involving TotalEnergies and Invenergy. TotalEnergies is redirecting $928 million previously associated with offshore-wind leases toward the Rio Grande LNG project, Gulf of Mexico oil production, and domestic shale gas. Invenergy’s $765 million agreement redirects capital toward natural-gas power plants in five Midwestern states, together with geothermal projects in the western United States.

The emerging signal is therefore more significant than the familiar observation that natural gas can often be developed faster than clean generation. Public policy is now influencing which projects remain available, which capital investments move forward, and which technologies are positioned to meet rapidly accelerating AI-era electricity demand.

The three-year window is not merely a construction timetable. It is a policy window—and the infrastructure selected within it may shape electricity costs, water use, emissions, and regional resilience for the next thirty years.

The Stratos Decision

In early May 2026, Utah’s Military Installation Development Authority and the Box Elder County Commission approved the land-use framework for the Stratos Project — a 40,000-acre AI data center campus backed by Kevin O’Leary, branded “Wonder Valley” in the developer’s marketing materials, and sized for 9 gigawatts at full buildout. For scale, the entire state of Utah averages about 4 gigawatts of electricity use. One facility, when complete, will run at more than twice the load of an entire state. A second proposed project, Joule Power, is targeting 12 gigawatts. Both are designed to run off-grid, on dedicated natural gas turbines built behind the meter, tapping the Ruby Pipeline at a daily draw that approaches or exceeds the pipeline’s existing capacity.

State estimates put the emissions increase from Stratos alone at roughly 50% above Utah’s current statewide total. Water demand is projected at 16.6 billion gallons per year, drawn from the watershed that feeds the Great Salt Lake — already down roughly 70% from historical levels. This is the Food-Energy-Water-Compute Nexus violated in real time, in one of the most water-stressed basins in the country.

Public resistance has been substantial. After nearly 4,000 formal protests were filed with the Utah Division of Water Rights, the developers withdrew their water rights change application on May 5, 2026 — and announced their intent to refile under additional supporting documentation. The next day, Utah House Bill 60 took effect, narrowing the state engineer’s authority to weigh broad public welfare concerns in water-right reviews. Citizens used the process as designed. The process was rewritten twenty-four hours later.

The deeper issue is that Stratos is not an outlier. SoftBank has announced a 10 GW gas-fired campus in Ohio. Meta has committed to seven new gas plants for a 7 GW Louisiana facility. The pattern is now industry standard: when AI capex demands first power within 24 to 36 months, the only generation asset that can be delivered on that timeline is a gas turbine. The rest of the decision — climate, water, public health, long-term economics — is being absorbed by whatever county happens to host the project.

From Delay to Buyout: When the Clean Option Leaves the Table

The new Duke Energy buyout in North Carolina adds a second version of the same pattern. It is not only that gas developers move faster than clean developers. It is that clean capacity can be removed from the future supply stack altogether. When a 1.3 to 1.6 GW offshore wind opportunity is retired before construction while the reinvestment path includes natural gas, the market is not simply choosing gas over clean power. Policy is narrowing the menu before the market gets to choose.

The broader buyout portfolio makes the point harder to dismiss. TotalEnergies agreed to exit U.S. offshore wind and invest $928 million in LNG, Gulf of Mexico oil, and shale gas projects before its offshore wind leases were terminated and reimbursed. Invenergy accepted a $765 million buyout to terminate four offshore wind leases off New York, California, and Maine, with the Department of the Interior saying the funds would support natural-gas plants in Indiana, Wisconsin, Iowa, Kansas, and Missouri, plus geothermal projects in the West. Duke brings the pattern from multinational developers to a major U.S. regulated utility.

Duke Energy’s own generation planning shows the same pattern from the inside. In its biennial carbon plan filed with the North Carolina Utilities Commission, the utility has petitioned to build 9 gigawatts of new gas capacity, arguing the state’s mandated 70 percent emissions cut by 2030 is no longer achievable without risking reliability. Ratepayer and environmental advocates have said Duke’s long-term capacity-planning software was set with artificial limits on renewable expansion, which had the effect of steering the optimization model’s preferred resource mix toward combined-cycle gas turbines rather than wind, solar, and storage. Separately, the U.S. Department of Energy has directed close to $96 million in Defense Production Act grants toward refurbishing Duke coal units at Belews Creek and other North Carolina and Kentucky plants, using grid-reliability and national-security authority to extend the service life of existing coal capacity.

That is why the three-year window is not merely a construction-timeline problem. It is a capital-allocation problem. If the clean project spends years in the queue and then faces political buyout risk, while the gas turbine has private land, pipeline access, known financing documents, and federal policy support, the outcome is not a neutral market decision. It is an engineered time advantage.

There is a counter-signal worth noting. FERC approved PJM’s expedited interconnection track in June 2026, allowing up to ten large or uprated capacity projects per year to pursue interconnection agreements within ten months and operation within three years if they meet certain readiness and state-siting criteria. That is exactly the kind of procedural reform that could let clean firm power compete on the same clock as gas. The policy question is whether that pathway becomes the rule, or whether it remains a narrow exception while buyouts and behind-the-meter gas define the buildout.

The Wrong Question: Gas vs. Nuclear

The framing that has captured most of the policy conversation is gas versus nuclear. Small modular reactors (SMRs) are presented as the clean firm answer to AI’s appetite for power. The framing is appealing, and it is wrong for the decade we are actually in.

SMR marketing typically claims 3 to 4 years from groundbreaking to commercial operation. Reality has been closer to 10 to 12 years from project announcement to first kilowatt-hour. NuScale, the most advanced U.S. developer, watched its first commercial project collapse over cost in late 2023. TerraPower’s Natrium project in Wyoming has slipped its commercial operation date to 2030 or later. The fuel supply chain for next-generation reactors is largely Russian-dominated, and U.S. domestic production will not reach commercial scale before 2028 or 2029. None of this is fatal to nuclear. SMRs are real, they will matter, and the United States will need them by the mid-2030s. But the first commercial U.S. SMR will probably energize around 2030 or 2031 if everything goes well. That is at least five years too late to influence which fuel the next 50 GW of hyperscale AI capacity locks in. Anyone framing the choice as gas versus nuclear has already conceded the next decade to gas. The gas industry, understandably, prefers this framing.

The Right Question: Gas vs. WWS+Battery

The serious alternative to gas in the 2026 to 2030 window is wind, water, and solar paired with battery storage — the WWS architecture that Stanford’s Mark Jacobson has been modeling for fifteen years and that the cost and deployment curves have finally caught up to.

The speed advantage is real. Utility-scale solar projects energize in 12 to 18 months from permit to commercial operation. Onshore wind, 18 to 24 months. Battery storage, often under 12 months. A solar-plus-storage facility sized to power a 1 GW data center can be physically built faster than the data center itself.

Lazard’s 2024 levelized cost analysis put unsubsidized utility-scale solar at $29 to $92 per megawatt-hour and onshore wind at $27 to $73. New combined-cycle gas came in at $45 to $108 — and that range is rising as gas prices have decoupled from the cheap-gas era of the 2010s. Four-hour lithium battery storage has dropped below $150 per megawatt-hour and continues falling.

The firmness question — can renewables actually deliver 24/7 power for an AI workload — is where the conversation usually stops and shouldn’t. Modern WWS systems can deliver firm power if they are designed for it. Texas’s ERCOT grid is running on 30% or more renewables most days. California regularly hits 100% renewable supply during midday hours. Form Energy’s iron-air long-duration batteries, with 100-hour discharge, are now in commercial deployment. The combination of 4-hour lithium for daily cycling, multi-day iron-air or flow batteries for weather variability, and geographic diversification across a few hundred miles handles the firmness problem at a total cost still below new gas.

The technology works. The cost works. The construction timeline beats gas. So why are the deals not getting done?

Why Gas Keeps Winning Anyway

Three reasons, and none of them are about the engineering.

First, permitting asymmetry. A gas turbine on private industrial land with an existing pipeline tap can be permitted in months. A utility-scale renewable project requires an interconnection study that takes three to seven years in most regional grid operators. PJM Interconnection’s queue had more than 3,300 GW of projects waiting in 2024 — larger than the entire installed capacity of the U.S. grid. The gas plant doesn’t queue. It just builds.

Second, off-grid logic. When a developer builds behind the meter, they escape the interconnection queue, the renewable portfolio standard accounting, and most utility scrutiny. The Stratos model — own the land, own the generation, never touch the public grid — is partly an engineering choice and largely a regulatory-arbitrage choice. The same arbitrage is available to a clean-powered behind-the-meter campus, but the developer has to want it.

Third, capital structure familiarity. Project finance for gas turbines is a forty-year-old asset class. Banks know the documents, insurers know the risks, the deal closes on a known template. WWS plus storage plus long-duration backup requires a slightly different deal structure — and that purely informational friction is enough to tip many decisions back toward gas in the conference room on the forty-second floor.

Fourth, policy buyout risk. A renewable project can have a good resource, a viable market, and a real regional need — and still be paid to exit if the federal policy environment turns against it. That changes the investment signal for every clean developer watching the queue. It also makes the gas option look artificially safer, not because it is cheaper over thirty years, but because the rules are being written around its speed and capital is being redirected toward assets that can clear the calendar test.

These are policy problems and market-design problems. They are not laws of physics. They can be fixed in the time available, if they are recognized as the actual bottleneck.

Conclusion: Sooner, Not Later

The choice in front of the country right now is not gas versus nuclear. Nuclear arrives later. The choice is gas versus clean firm power that exists today, paired with the policy reforms that would let it be deployed at AI’s speed.

The reforms are unglamorous and concrete: finish FERC Order 2023’s interconnection queue reforms; expedite siting and easement for clean firm projects with the same urgency that gas turbines receive; stop using federal settlements to buy back viable clean-energy leases unless the replacement supply is demonstrably cleaner, cheaper, and faster; create standard project finance templates for solar-plus-storage-plus-long-duration packages so the deal closes as easily as a gas plant; and require any new behind-the-meter generation over a certain threshold to publish full lifecycle emissions, water, and heat data so the public conversation has facts to work with. The opposite reforms — narrowing public objection rights, fast-tracking gas while slow-walking clean alternatives — are also possible, and Utah’s HB 60 shows what they look like when they happen.

The Perpetual Sustainability™ premise applies. Systems must regenerate, or they fail. A 9 GW gas plant built in 2026 is a 30-year asset, locking in fossil generation through 2056. A 9 GW solar-plus-storage campus built in 2026 is a 30-year asset that gets cleaner as the grid around it gets cleaner. The infrastructure decisions made in the next three years will determine whether AI accelerates the energy transition or delays it by a decade. The window is open. It will not stay open.

---

Two years later, the analyst — promoted now, her own deals to run — laid out a different pair of binders. The one labeled SOLAR + STORAGE + LONG-DURATION was still three inches thick, but the four-year transmission study had become an eighteen-month interconnection slot under the new federal rules. The renewable energy credit assumptions had firmed up. A consortium of insurers had standardized the project finance documents. The turbine delivery slot in the other binder was still available, but it was no longer the only path to first power by the contract deadline.

This time the managing director did not look at the clock above the door. He read the first binder for a long time before he spoke. The analyst noticed the difference and understood what it meant. The math had not changed. The rules had. That was all it had ever taken.

Dynamic Links

Internal — PerpetualInnovation.org

• The Food-Energy-Water-Compute Nexus: Adding the Fourth Letter — the framework this article applies
• Earth Day 2026: Fossil Fuels, Hormuz & the One-Way Trip — the broader argument on stranded fossil assets
• Pi-rdAI — Regenerative Dynamic AI Operating System
• Pi-Sustain — Sustainability Strategy Hub

External — High-Authority Sources

• IEA — Energy and AI (2025) — global data center electricity projections
• Lazard — Levelized Cost of Energy+ (2024) — current unsubsidized cost comparisons across generation types
• Lawrence Berkeley National Laboratory — 2024 U.S. Data Center Energy Usage Report — federal baseline on U.S. data center load growth
• FERC Order 2023 — Interconnection Queue Reform — the queue reform that determines clean energy deployment speed
• Stanford — The Solutions Project (WWS) — Jacobson group’s wind-water-solar roadmap
• Reuters — Duke Energy to terminate North Carolina offshore wind lease (2026) — current example of a clean-energy lease buyout and reinvestment shift
• Reuters — TotalEnergies abandons U.S. offshore wind and shifts investment to oil and gas (2026) — example of a large buyout-style settlement and fossil-fuel capital pivot
• Reuters — Invenergy offshore-wind lease termination and reinvestment agreement (2026) — example of redirected capital toward Midwestern gas plants and Western geothermal
• Reuters — PJM expedited interconnection track approved by FERC (2026) — current counterexample showing how faster interconnection could help non-gas resources compete on time
• Reuters — PJM moves toward managing data center demand (2026) — current grid-operator response to AI/data-center load growth
• Sierra Club — Duke Energy reduces proposed rate increase after public pressure (2026) — state-level example of ratepayer pushback on gas-driven rate cases
• PR Newswire — DOE announces Defense Production Act grants for Duke Energy coal plants (2026) — federal coal-fleet refurbishment funding cited in the carbon-plan subsection

Suggested GenAI Prompts

1. I like using stories to convey complex or challenging topics. Tell a story that conveys both the fact and the counter-factual about this topic: a county that approves a behind-the-meter gas-powered AI data center versus one that approves a behind-the-meter solar-plus-storage data center of the same size. Make sure the narrative is grounded in verifiable data and be prepared to cite reliable sources for every factual claim embedded in the story.
2. What are the top 4 to 6 things a [state legislator / county commissioner / utility regulator] should ask about a proposed AI data center before approving it? [Optional: Our jurisdiction is in (state/region), our biggest current energy source is (gas / coal / nuclear / renewables), and our water situation is (stressed / adequate).]
3. Walk me through, in plain language, how interconnection queue reform under FERC Order 2023 actually changes the timeline for getting a clean energy project built. What is the bottleneck before reform, and what does it look like after?
4. What are the easiest first steps a small business or community group could take to push for clean energy choices in a local AI data center decision? [Optional: We are in (state/region) and the proposed project is (size).]
5. Build a side-by-side comparison of three power options for a hypothetical 1 GW AI data center coming online in 2029: behind-the-meter natural gas, behind-the-meter solar-plus-storage-plus-long-duration battery, and grid-supplied power with a future SMR. Score each on time-to-energization, levelized cost, lifecycle emissions, water consumption, and 30-year stranded asset risk.

AI Disclosure and Attribution

This article was created with assistance from Claude Opus 4.7 (2026, May) as part of the Pi-rdAI Rapid Strategic Planning ecosystem. Additional background research and drafting support came from Gemini 3.5 (2026, May) and ChatGPT 5.5 (2026, May). The July 2026 update was created with assistance from Gemini 3.5 Deep Research (2026, Jul), ChatGPT 5 (2026, Jul), and Claude Sonnet 5 (2026, Jul). Feature image generated using ChatGPT image (2026, July) based on the article.Content development and review by Dr. Elmer B. Hall — Strategic Business Planning Company (SBPlan.com) and PerpetualInnovation.org.

Copyright © 2026 Strategic Business Planning Company. All rights reserved.