Central Asia Emerges as Next Battleground for AI Infrastructure

A New Front in the Global Compute Race

While the world watches hyperscalers pour billions into Virginia and Singapore, a quieter infrastructure contest is unfolding across Central Asia. Uzbekistan and Kazakhstan, two landlocked economies historically known for cotton and oil, are now vying to host the compute capacity that powers artificial intelligence workloads across a region stretching from the Caspian to the Chinese border.

The stakes are straightforward: whoever builds first and scales fastest can anchor cross-border traffic, attract co-location tenants, and shape how enterprises and governments in neighboring markets source inference and training capacity. At DailyTechWire, we've tracked similar races in Southeast Asia and the Gulf; Central Asia's version carries the added complexity of Soviet-era grid infrastructure, geopolitical proximity to both Russia and China, and a regulatory environment still finding its footing on data sovereignty.

Two projects have emerged as early movers. In Tashkent, a Saudi-backed facility called TAS-1 is being developed by DataVolt, with the first phase slated to come online before year-end. Across the border in Kazakhstan, a separate campus linked to Nvidia is targeting 125 megawatts of capacity by 2027. Neither has disclosed pricing, but the timeline suggests both are racing to sign anchor tenants before regional demand fully materializes.

The Tashkent Bet: Saudi Capital Meets Uzbek Energy

DataVolt's TAS-1 facility represents the most concrete progress so far. The company has not published detailed technical specifications, but the involvement of Saudi backing points to the Gulf's broader strategy of seeding compute infrastructure in emerging markets where land and power costs remain below tier-one Asian hubs.

Uzbekistan offers two advantages: relatively cheap electricity, much of it still generated from natural gas, and a government eager to diversify beyond commodity exports. The country has been courting foreign investment in digital infrastructure as part of a wider push to formalize its shadow economy and attract technology-sector capital. That policy shift has included new data-localization rules and incentives for co-location providers, though enforcement and regulatory clarity remain uneven.

The year-end target for TAS-1's first phase is aggressive. It implies that site preparation, power provisioning, and cooling systems are already well underway. If DataVolt meets the deadline, Tashkent will leapfrog Almaty and Astana to become the first city in Central Asia to host a large-scale, purpose-built AI-ready data center. That timing advantage could prove decisive in locking in early customers, particularly regional enterprises and government agencies reluctant to route sensitive workloads through Moscow or Beijing.

Kazakhstan's Nvidia Connection and the 125 MW Target

Kazakhstan's project carries a different profile. While details remain sparse, the involvement of Nvidia suggests the facility is being designed with GPU-dense workloads in mind. The 125-megawatt target by 2027 would put the campus in the same league as mid-tier hyperscale sites in Southeast Asia, though still well below the multi-hundred-megawatt campuses now common in Northern Virginia or West Texas.

Kazakhstan has been positioning itself as a China alternative for rare earths and critical minerals; extending that logic to compute infrastructure makes strategic sense. The country has more developed financial markets than Uzbekistan, a larger economy, and better integration with international logistics networks. It also faces less regulatory unpredictability, though corruption and bureaucratic friction remain obstacles for foreign investors.

The 2027 timeline is more conservative than Uzbekistan's year-end push, which may reflect a more deliberate approach to grid upgrades and tenant onboarding. It also leaves room for phased expansion: 125 megawatts is enough to serve regional demand in the near term, but the campus could scale further if cross-border AI adoption accelerates faster than current forecasts suggest.

Why Central Asia, Why Now

Three forces are converging to make Central Asia a plausible compute hub. First, latency: for workloads originating in Kazakhstan, Uzbekistan, Kyrgyzstan, and Turkmenistan, routing through Singapore or Mumbai adds 80 to 150 milliseconds of round-trip delay. That penalty matters less for batch training jobs, but it becomes a friction point for real-time inference, video processing, and interactive applications. A regional data center can cut that latency by half or more.

Second, data sovereignty: governments across the former Soviet space are tightening rules around where citizen data can be stored and processed. Hosting infrastructure in-country or in a neighboring market with treaty agreements can simplify compliance and reduce the risk of sudden regulatory shutdowns.

Third, cost arbitrage: power and land in Tashkent or Almaty remain cheaper than in established Asian hubs. That gap is narrowing as Central Asian governments raise electricity tariffs and as new supply comes online in places like Malaysia and Indonesia, but for now the economics favor greenfield builds in less saturated markets.

The question is whether demand will materialize fast enough to fill the capacity. Central Asia's combined GDP is smaller than Vietnam's, and enterprise AI adoption is still in early innings. If regional workloads grow more slowly than projected, both Uzbekistan and Kazakhstan risk ending up with stranded assets or having to compete on price for workloads that could just as easily run in Istanbul or Dubai.

The Grid Challenge No One Is Talking About

Both countries inherited Soviet-era transmission infrastructure, much of it built in the 1970s and 1980s. Portions have been upgraded, but large swaths remain vulnerable to voltage instability, especially during peak summer cooling demand. A 125-megawatt data center represents a significant new load, equivalent to a mid-sized industrial plant. If the local grid cannot absorb that demand without triggering brownouts or requiring expensive dedicated substations, operating costs will climb and uptime guarantees will become harder to meet.

Neither DataVolt nor the backers of Kazakhstan's Nvidia-linked campus have publicly detailed their power-provisioning strategy. The silence suggests those negotiations are still ongoing or that the solutions involve a mix of on-site generation, grid upgrades funded by the host government, and contractual risk-sharing that neither side wants to advertise.

This is not unique to Central Asia. Every emerging market that tries to leapfrog into data center infrastructure hits the same wall: the grid was not built for this. The difference is that in Singapore or Seoul, utilities have decades of experience managing large, lumpy industrial loads. In Tashkent and Almaty, that institutional knowledge is thinner, and the margin for error is smaller.

What Happens If Both Succeed

If both projects come online on schedule and fill capacity at a reasonable pace, Central Asia will have validated a new tier of compute geography. That outcome would likely trigger a second wave of investment: Chinese cloud providers exploring cross-border expansion, Turkish telcos looking for co-location partnerships, and possibly Russian operators seeking to route workloads outside the direct reach of Western export controls.

It would also put pressure on neighboring markets. Azerbaijan, Georgia, and Armenia have all floated data center ambitions but have moved more slowly. If Uzbekistan and Kazakhstan establish themselves as regional anchors, those countries risk being relegated to edge nodes or missing the window entirely.

The less optimistic scenario is that one project stumbles, demand undershoots, and the winner-takes-all dynamic leaves the region with a single viable hub and a half-empty facility. In that case, the race will have been more about first-mover risk than first-mover advantage.

Either way, the starting gun has been fired. Central Asia's compute future will be shaped in the next eighteen months, and the region's position in the broader AI supply chain will depend on whether ambition, capital, and infrastructure can converge faster than the regulatory and technical headwinds can slow them down.