In May 2025, the United States agreed to let the United Arab Emirates import 500,000 of Nvidia’s most advanced AI chips a year and to back a 5-gigawatt computing campus in Abu Dhabi. The terms looked closer to an instrument of statecraft than a commercial transaction, which is roughly how Mike Winston has come to describe compute.

Winston comes at this as a markets practitioner rather than a policy academic. He’s a CFA charterholder who spent five years running a billion-dollar merger arbitrage book at Millennium Partners, founded Sutton View Capital in 2012, and now chairs a company building data centers. When he talks about computing power, he frames it the way a strategist would talk about coal in 1850 or oil in 1950: as the input that decides which societies set the terms for everyone else.

Why Mike Winston Treats Compute as Infrastructure, Not a Sector

He set out the argument in an April 2026 interview. Societies that hold large amounts of compute and pair it with strong scientific and engineering cultures, he said, will tend to dominate those that don’t, and he sees the gap as existential as much as economic.

The mechanism he describes is cumulative. With enough computing power, a society can solve first-order scientific problems that have resisted answers for decades or centuries. Solving those problems then exposes a new layer of second-order problems that were invisible before, because you couldn’t see them until the first layer cleared. A society stuck below that threshold falls behind on the known questions and never even learns what the next ones are.

In his framing, compute stops being one technology sector among many and becomes something closer to roads, electricity, or a power grid. The claim is partly about sequence: get the infrastructure right and the discoveries compound on top of it; get it wrong and the deficit compounds instead. History, in his reading, has tended to reward the asymmetry. From the Age of Exploration forward, the societies with superior technology have tended to prevail over those without it, and rarely gently.

The Physical Layer Behind the Argument

One reason the thesis can be examined at all is that compute, unlike an idea, has a physical address. Training and running advanced AI models depends on specialized chips, principally the graphics processing units that, as Duke University’s Deep Tech initiative describes, sit at the heart of AI data centers because their architecture runs thousands of computations in parallel. Those chips are made through a supply chain so specialized that no single country can produce, at the quality and volume a modern economy needs, on its own. Fabrication concentrates heavily in Taiwan, where TSMC has refined the manufacturing process to a degree competitors have struggled to match.

That physical concentration is part of why governments treat compute as a lever. A 2025 report from the Center for a New American Security, written by senior fellow Janet Egan, makes the point directly: unlike algorithms and know-how, compute is physical, with a narrow, specialized supply chain, which makes it far more controllable through policy. Washington has leaned on that fact since 2019, layering on semiconductor equipment restrictions and then advanced-chip export controls that were tightened in 2022, 2023, 2024, and 2025.

The figures behind the buildout help explain why Winston tends to use the word “infrastructure” literally. The 2022 CHIPS and Science Act committed $52 billion to domestic semiconductor research and manufacturing, and the Semiconductor Industry Association projects U.S. fabrication capacity will roughly triple between 2022 and 2032. China has moved on a comparable scale, with its national chip fund approaching $100 billion. The Gulf deals announced in 2025 carried price tags to match, including more than $80 billion in cross-border AI investment tied to Saudi Arabia.

There’s a counterargument worth taking seriously. When the Chinese lab DeepSeek released its R1 reasoning model in early 2025, roughly four months after OpenAI’s o1 and at comparable performance, it suggested that algorithmic efficiency might let a country do more with fewer chips. CNAS addresses this head-on: efficiency gains are real, and the compute required to train a model of equivalent capability has been falling by half roughly every eight months. But advancing the frontier still demands exponentially more computing power, which means whoever holds the deepest compute base keeps the ability to run the experiments that produce the next breakthrough. Efficiency makes it cheaper to match yesterday’s models. Reaching tomorrow’s still takes more compute than anyone else has, which on this reading is where the lead is decided.

Compute and Political Values

The part of Winston’s thesis that separates it from a standard industrial-policy argument is the link he draws to political values. He argues that technological leadership and political systems are intertwined, and that the stakes rise sharply if societies indifferent to personal freedom secure the compute advantage first.

The concern isn’t unique to him. The CNAS report frames U.S. strategy in nearly the same terms, arguing that American leadership should “crowd out China’s expanding sphere of technology influence and ensure the AI transition is underpinned by trusted democratic technologies.” The fear on both Winston’s side and the analysts’ side is the same: that compute concentrated in authoritarian hands becomes self-reinforcing, because the nation with the most capable infrastructure can train the best systems, deploy them at scale, and fund the research that widens its own lead. Once that cycle starts, latecomers find it progressively harder to close the distance.

Winston pushes the logic a step further than most policymakers will say out loud. If compute decides which scientific frontiers a society can even perceive, he argues, then the political character of whoever controls the largest compute base eventually shapes the rules everyone else lives under. He frames it as a claim about civilization rather than quarterly returns.

The Investor’s Bet Behind the Thesis

That raises a practical question: what is Mike Winston, investor, actually doing about it? So far, he’s put capital where the argument points, at the infrastructure layer rather than the model layer.

In 2025 he co-sponsored AI Infrastructure Acquisition Corp. (NYSE: AIIA), a blank-check company where he is chairman and chief executive. Its mandate, per the company, is to target firms advancing artificial intelligence and the next-generation data center infrastructure that AI depends on, spanning high-performance computing, semiconductors, and the broader digital infrastructure chain. The positioning is deliberate: rather than bet on which model wins, AIIA aims at the physical capacity every model requires.

The same orientation runs through Jet.AI (NASDAQ: JTAI), the company Winston founded as Jet Token in 2018 and now chairs, which has repositioned around AI data center development. Jet.AI is advancing shovel-ready data center sites in Manitoba, the Maritimes, and Moapa, Nevada, with combined planned capacity exceeding a gigawatt, and it treats access to reliable power as the binding constraint on the entire enterprise. AIIA and Jet.AI are distinct companies with separate tickers and mandates, but they share a premise: the scarce and decisive resource is the powered campus the model runs on.

That emphasis on power lines up with the electricity figures. A 2024 Department of Energy report prepared by Lawrence Berkeley National Laboratory found that U.S. data centers drew about 4.4% of the country’s electricity in 2023 and could draw between 6.7% and 12% by 2028. Consumption climbed from 176 terawatt-hours in 2023 toward a projected 325 to 580 terawatt-hours within five years, and load growth that already tripled over the past decade is set to double or triple again. Chips can be ordered and shipped in months. The electricity to run them, and the grid connection to deliver it, takes considerably longer to secure.

That mismatch has shifted where the constraint sits. For much of the AI boom the scarce input was the GPU, and increasingly it’s power. CNAS notes that domestic permitting and energy availability now limit America’s ability to build large-scale AI data centers, even as nondemocratic states with regulatory flexibility, capital, and ready power race to build frontier-scale facilities of their own. New campuses can wait years for a grid interconnection, which gives a developer who secures power-ready land early an advantage that outlasts any single chip allocation.

This is the logic behind Jet.AI’s power-first site selection. The company says it screens locations for reliable, scalable power and acreage ahead of connectivity or proximity to existing tech hubs, and its Manitoba footprint sits in a province whose grid runs almost entirely on hydroelectric power. An investor who believes compute is civilizational infrastructure would, logically, go looking for land and electricity before anything else, which is the direction Winston has taken.

What the Thesis Asks Markets to Take Seriously

The fuller version of Winston’s view goes past the now-common claim that AI is important. He holds that compute capacity is itself a measure of national power, that the measurement is widening into a self-reinforcing lead, and that the values of whoever holds the lead will eventually be encoded into the technologies the rest of the world adopts.

Reasonable people can dispute the edges. Efficiency breakthroughs like DeepSeek’s may compress the advantage of sheer scale more than Winston allows. Export controls have leaked, with smuggling networks supplying restricted chips, which complicates the assumption that the supply chain stays controllable. And the history of technological determinism is littered with predictions that mistook a steep early curve for an unbroken one.

The structure of the bet is harder to dispute. Governments are now treating chips and data centers as instruments of statecraft, committing tens of billions of dollars and writing export rules around them. Mike Winston reached a conclusion much like the national security establishment’s. Where an analyst would write it up, he’s put money into the infrastructure layer he believes will decide the outcome.

DISCLAIMER – “Views Expressed Disclaimer – The information provided in this content is intended for general informational purposes only and should not be considered financial, investment, legal, tax, or health advice, nor relied upon as a substitute for professional guidance tailored to your personal circumstances. The opinions expressed are solely those of the author and do not necessarily represent the views of any other individual, organization, agency, employer, or company, including NEO CYMED PUBLISHING LIMITED (operating under the name Cyprus-Mail).