Central Asia’s AI Window

Δημοσιευμένα 2025-10-10 17:00:59

182

Central Asia’ plentiful critical minerals make it an asset to Washington’s bid for AI dominance.

On September 22, 2025, during his visit to the United States, President Kassym-Jomart Tokayev presided over a signing between Pittsburgh-based Wabtec and Kazakhstan Temir Zholy (KTZ) for 300 freight locomotives—a multi-year package valued at $4.2 billion. President Donald Trump amplified the deal as “the largest Railroad Equipment Purchase in History.”.

Look under the hood, though, and the triumphalism fades. Wabtec has been assembling Evolution-series locomotives in Astana since 2009, investing heavily and producing more than 600 units for KTZ and export. The new contract largely aggregates activity that was already underway, rather than opening a fresh lane of US-Kazakhstan cooperation. It does not materially deepen vertical ties (new technologies, supply-chain co-location) or horizontal ones (regional market access), nor does it shift Central Asia’s strategic balance vis-à-vis China or Russia. In short, it’s good optics with limited leverage.

What this moment really exposes is Washington’s recurring preference for splashy announcements over durable economic statecraft. Central Asia—and Kazakhstan in particular—wants consistent frameworks that compound, not one-off headlines.

With the US-China AI competition underway, the smarter play runs through power and materials—nuclear fuel, copper, magnets, and grid-ready corridors. This is where Central Asia can be a swing partner rather than a photo-op. In July, the White House launched an AI Action Plan, setting an explicit goal of global leadership and organizing policy around three key pillars: accelerating innovation, building the infrastructure and energy to power it, and leading internationally on AI diplomacy and security.

The plan pairs deregulation with “build-out” realism, streamlined permitting for data centers and fabs, grid upgrades, and a National AI Research Resource pilot to broaden access to compute and data. So far, the scoreboard favors the United States. The most capable frontier models are overwhelmingly American. Nvidia, meanwhile, controls roughly 90 percent of the data-center GPU market that powers training and inference, an edge that US firms have parlayed into a first-mover scale advantage.

But advantage is not destiny. China has proven it can compress the distance between innovation and industrialization, and January’s “DeepSeek scare” was a reminder that smart optimization and scale can shock markets even without a breakthrough architecture. US tech stocks stumbled, Nvidia shed value, and investors got a preview of what a faster-than-expected catch-up could look like.

Strip away the hype, and two hard constraints determine the outcome: the energy required to run AI and the materials needed to build it. The International Energy Agency projects that global electricity demand from data centers will more than double to 945 TWh by 2030, with the United States accounting for the largest share of the increase. In the United States, the Department of Energy estimates that data centers already consumed 4.4 percent of US electricity in 2023 and could reach 6.7–12 percent by 2028.

Here’s the competitive wrinkle: America’s innovation edge means many companies are building parallel AI stacks—competing labs, duplicated clusters, and overlapping campuses that all need firm power. That’s healthy for ideas, but it’s brutal on grids. China, by contrast, is attempting to centralize and orchestrate computing at scale through its “Eastern Data Western Computing” (EDWC) program—eight national hubs and ten data-center clusters linked to cheaper inland power, a reminder that Beijing can utilize planning tools to smooth out redundancy in ways US markets cannot.

If the constraint is power, the clear answer is nuclear, which can provide continuous, carbon-free, and highly land-efficient energy. This is no longer theoretical. Microsoft signed a 20–year Power Purchase Agreement (PPA) to revive Three Mile Island Unit 1 (835 MW) for AI data centers; AWS acquired a 960 MW nuclear-powered campus tethered to Pennsylvania’s Susquehanna plant; and Equinix just announced more than 1 GW of advanced-nuclear procurements, including a 500 MW deal with Oklo and preorders for transportable microreactors. Small modular reactor (SMR) pilots are also moving forward. Standard Power plans to deploy nearly 2 GW of NuScale SMRs in Ohio and Pennsylvania, aiming to serve data centers beginning in 2029.

But if nuclear is the scalable, climate-aligned answer, uranium is a strategic chokepoint. With Russian imports banned and domestic fuel services being rebuilt, Kazakhstan, which already produces over 40 percent of the world’s mined uranium, becomes the swing partner that lets US AI scale without energy-security risk. The alternative—ceding leverage to Moscow in the nuclear fuel cycle—should sound uncomfortably familiar to anyone who watched the outcome of Europe’s past dependence on Russian energy.

If energy is the first hard limit to AI, critical materials are the second. The United States still relies on China across most stages of the rare-earth and magnet supply chain, and that dependence spills from the commercial tech stack into the defense industrial base. DOE and USGS data show that China dominates not only mining but also separation, refining, and NdFeB magnet manufacturing—the very stages that turn oxides into motors, actuators, fans, and drives—components that touch everything from server halls to missiles.

Beijing has not been shy about leveraging its influence. Since 2023, it has implemented export controls on gallium, germanium, graphite, and antimony, and this spring tightened its management of rare-earth exports, periodically restricting volumes or slowing down license issuance.

AI itself is less REE-intensive than, say, EVs—but AI at scale is metals-intensive. Data centers need copper by the tens of thousands of tons per site for busbars, switchgear, cabling, and networking; they also pull in aluminum, electrical steel, and rare-earth magnets in high-efficiency motors, fans, and HDDs. Analysts at Trafigura estimate AI data centers could add up to 1 million metric tons of copper demand by 2030, on top of an already tight market.

That is how Central Asia and the new US-brokered South Caucasus corridor, which plugs into the Middle Corridor, can become a strategic channel for the United States and allies to diversify away from the Chinese processing monopoly. Central Asia produces or processes approximately 30 of the 50 US-listed critical minerals, including 10 of the 12 for which the United States was 100 percent import-reliant in 2023.

The USGS has mapped 384 rare-earth and rare-metal occurrences across the five states—a regional endowment built for scale, not one-offs. Uzbekistan’s Almalyk copper complex is midway through its expansion to produce 300,000 tons of copper cathode per year by 2030. Tajikistan, meanwhile, is the world’s second-largest antimony producer, just as China has implemented export restrictions on this defense-critical metal.

The framework is already there, if underused. Through the C5+1 Critical Minerals Dialogue, Washington and the five capitals can move from MOUs to term sheets and standardize, fund, and pace projects along the Middle Corridor. They can lock in a small set of off-takes for AI-critical materials—uranium and copper, yes, but also electrical steel, aluminum, rare-earth magnets, and graphite—and the market will take care of the rest.