Mike Winston, Investor and Jet.AI Executive Chairman, on the Power Shortage Driving AI Infrastructure

The grid is the constraint. Interconnection queues in the United States now run eight to ten years in some jurisdictions, long enough that data center developers have concluded they cannot wait for utility-scale power. The companies moving fastest have built parallel infrastructure, mobile, modular, and jet-engine-derived, to bridge the gap between when AI clusters need power and when the grid can provide it.

ProEnergy sold 21 aero-derivative gas turbines to just two data center clients. The total capacity exceeds one gigawatt of bridging power. The turbines are second-life commercial jet engines: CF6-80C2 cores stripped from retired 767s and Airbus A310s, overhauled, mounted on steel skids, and converted to run on natural gas. Each unit produces 48 megawatts. Each can be on-site and generating power within 30 days. GE Vernova, which manufactures the LM2500XPRESS, a competing aero-derivative unit producing 35 megawatts, is quoting three-to-five-year lead times for new orders.

This is the market Mike Winston, investor, founder, and Executive Chairman of Jet.AI (NASDAQ: JTAI), entered when the company divested its aviation operations to flyExclusive in early 2025.

Grid Queues and Why the Wait Is Measured in Years

Landon Tessmer, vice president of commercial operations at ProEnergy, has observed permitting delays of eight to ten years in some regions, delays that begin before a single kilowatt reaches a data center facility. IEEE Spectrum, October 2025. Contact any major OEM about a new LM6000 or Siemens Energy SGT-A35 and the response is the same: three to five years minimum, and that estimate has been moving in one direction. Tessmer put the ProEnergy alternative plainly: a PE6000 can be delivered in 2027.

Manufacturing constraints drive part of the shortage. A single LM2500 unit contains thousands of hand-assembled components, many produced from high-alloy castings requiring precision heat treatment and specialized testing that cannot be trivially scaled. But the deeper issue is that demand accelerated faster than any production capacity could reasonably anticipate. The hyperscalers began announcing multi-billion-dollar AI infrastructure programs within months of each other, and the equipment that powers those facilities, at the grid level and behind the meter, was not available at the required scale.

OpenAI’s infrastructure partner, Crusoe, ordered 29 LM2500XPRESS units to supply nearly one gigawatt of generation for the Stargate campus outside Abilene, Texas. Each unit produces approximately 35 megawatts with a five-minute fast-start capability and can operate entirely independent of the grid. Crusoe built a power generation portfolio alongside its compute infrastructure. That integration, controlling the power as well as the compute, is the model that the most aggressive AI infrastructure developers are converging on.

Aero-Derivative Supply: The Hardware Angle in the Mike Winston Investor Thesis

Three conditions explain the aero-derivative moment. New gas turbines from OEM manufacturers are years away at any commercially useful scale. Repurposed aviation hardware is not.

ProEnergy’s Tessmer estimates approximately 1,000 CF6-80C2 cores will be retired from commercial aviation service over the next decade, a quantifiable supply of proven mechanical platforms that can be overhauled to as-new condition and converted for stationary power generation. ProEnergy has fabricated 75 PE6000 packages since 2020. Another 52 are under assembly or on order. Each package includes the gas turbine, generator, inlet cooling, emissions control, and integrated switchgear. A new entrant, FTAI Power, launched by FTAI Aviation in December 2025, is converting CFM56 aircraft engines into 25-megawatt turbines targeted at AI and cloud data center loads, with commercial production beginning in 2026. The field is expanding, but supply remains tighter than demand by a wide margin.

Three conditions support the structural case: an oversupplied pool of retired aircraft engines, an undersupplied market for deployable power generation, and data center demand measured in gigawatts with no credible near-term ceiling. The International Energy Agency has projected that data centers could consume 4% of global electricity by 2030, more than Japan uses today. According to IEEE Spectrum, large data centers already exceed 100 megawatts of demand, and the latest AI facilities being designed for training-scale workloads are planned at more than one gigawatt.

All three conditions are in place simultaneously and show no sign of resolving within the window that matters for current infrastructure investment decisions.

Mike Winston’s Background: Reading Market Structures Before They Price In

Mike Winston, CFA, spent his career in disciplines that require forming a view before the market reaches consensus. He joined Credit Suisse First Boston in 1999 as an equity research analyst covering the telecom sector, on a team that Institutional Investor ranked first in its annual All-America Research ranking. Five years at Millennium Partners followed, co-managing a $1 billion merger arbitrage and event-driven book through Catapult Capital Management. He founded Sutton View Capital in 2012 and went on to co-lead an activist litigation against the Dole Foods board that secured a 35% increase in total consideration for shareholders after demonstrating that the CEO had systematically suppressed the company’s stock before attempting a management buyout.

The merger arbitrage background is relevant context for the data center thesis. Event-driven investing is built on evaluating situations the market has already decided are settled. The discipline lies in determining when the consensus is wrong, holding a position against uncertainty, and understanding intrinsic value well enough to know what an asset is worth if conditions shift. “Markets are driven by greed and fear,” Winston has said. “In the long term, value prevails.” Success in that discipline requires being right about the second while surviving the first.

The move into company-building that produced Jet.AI was framed, in Winston’s own account, around the same logic: identifying a structural tailwind early and building toward it. Jet Token launched during the blockchain wave in private aviation. When regulatory constraints and the COVID disruption shifted conditions, the company pivoted through AI-enabled aviation tools, developing agentic booking software, route optimization for carbon reduction, and dynamic pricing for charter operations. That work produced a firsthand view of the constraint that turned out to matter most: compute infrastructure.

“We saw firsthand the scale of transformation AI would bring,” Winston said in a 2026 interview. “That led us to data centers, where the infrastructure opportunity is significant. Given my background in real estate finance and telecom, it was a natural transition. Today, we’re extending that into power generation using aero-derivative engines.”

Jet.AI’s Portfolio: Nevada, Canada, and the Power Studies (JTAI)

Jet.AI has signed a letter of intent for a 50-megawatt initial project within a 120-acre campus in Moapa, Clark County, Nevada, with a site plan capable of phased expansion to one gigawatt of total capacity. Power studies for the Nevada site are currently underway. A second front is the hyperscale data center joint venture in Canada: a 385-acre site in Winnipeg, Manitoba (the Midwestern Canada campus), along with a companion project in Maritime Canada. As of early 2026, the Canadian JV is advancing through its third milestone, which centers on validating energy access and grid feasibility, a prerequisite for environmental permitting and project financing.

The financial model is specific and Winston has stated it directly to shareholders. Data centers cost approximately $10 million per megawatt to build. Each megawatt generates roughly $1 million in net operating income, a 10% yield on construction cost. At a market capitalization rate of 6%, a 50-megawatt facility built for $500 million produces a stabilized value of approximately $800 million, creating $300 million in value on a debt-financed equity basis. Financed at 80% debt, the equity contribution is $100 million. The general partner’s portion of that equity, on exit, generates returns well above what comparable-risk debt would produce. “These data centers are the bedrock of the AI economy,” Winston wrote to shareholders, “and their demand will only grow.”

Jet.AI reported full-year 2025 results showing net income of $4.6 million, compared to a net loss of $12.7 million in 2024, with approximately $13.7 million in cash as of March 2026 and no debt, supported by a $250 million shelf facility. The income figure reflects unrealized investment gains from the company’s AI Infrastructure Acquisition Corp stake. The project pipeline is the operational story: Nevada power studies, Canadian milestone completion, and the pending flyExclusive transaction close.

The Infrastructure Thesis and What Makes It Durable

Winston holds a specific view about compute that he has stated in investor and public forums: societies that control large amounts of compute and maintain strong scientific and engineering cultures will dominate those that do not. The logic is sequential. With sufficient compute, first-order scientific problems that have resisted solution for decades become tractable. Once solved, they expose a new layer of second-order problems, ones that couldn’t be formulated before the first layer was cleared. Societies that don’t control the compute never see the second layer.

“Technological leadership and political values are deeply intertwined,” he has said. The implication for investors is that compute infrastructure built for AI training and inference has a demand profile measured in decades, not cycles. Power, the binding constraint on compute at scale, sits at the foundation of that thesis.

The aero-derivative angle is not the most obvious entry into that argument. It requires knowing that commercial jet engines retire on a predictable schedule, that their cores can be overhauled to generate stationary power at competitive cost, and that the alternative, waiting for new OEM capacity, means years of additional lead time in a sector where the difference between energized and unenergized represents years of revenue. Jet.AI spent years operating in aviation, building AI tools against real hardware constraints, before the company’s leadership concluded that the supply of retired aviation equipment represented something more than a used-parts market.

Whether Jet.AI executes at the pace its current project milestones require is a legitimate question. The Nevada power study is ongoing. Canadian permitting is early-stage. The flyExclusive transaction, which will clarify the company’s capital structure after the aviation divestiture, was expected to close by April 30, 2026. Each of those variables carries execution risk appropriate to a company of JTAI’s current scale.

The grid queue is documented. The turbine backlog is documented. The demand trajectory from the hyperscalers is documented. The retirement schedule for CF6-80C2 engines is documented. What is not yet settled is whether Jet.AI can position its projects early enough in the queue to capture the returns available to developers who bring power to AI infrastructure before the grid does.

Mike Winston has spent his career making exactly that kind of bet: that the gap between where the market consensus lands and where the evidence points is large enough to matter, and that the investors who see it first hold a position most others never get.

Disclosure: This article discusses Jet.AI, Inc. (NASDAQ: JTAI). Readers should conduct their own due diligence before making investment decisions. This piece reflects publicly available information and does not constitute investment advice.