An all-or-nothing approach to AI development in Central Asia won’t serve the United States in competition with China.

Wealthy Armenian-American businessmen, the US vice president, and the Kazakh steppe sound like elements of an old-school Soviet joke. But in 2026, they’re the building blocks of a growing push to establish the United States as the backbone of an artificial intelligence supersystem in Eurasia. That is, if the contradictions in US policy don’t inadvertently hand AI preeminence in the region to China.

Without their own localized computing power, emerging economies—which make up all of Eurasia—risk dependence on wealthier countries and higher costs for access to AI models. That combination could mean foregone productivity gains and slower economic growth as AI adoption accelerates worldwide.

US efforts to bridge this compute gap in Eurasia are de facto led by Firebird.ai, a San Francisco startup founded in 2025 by two Armenian-American tech entrepreneurs to provide top-quality US computing power to emerging markets, starting with a massive supercomputing hub in Armenia. Their project has brought together the best of the American private sector and the US government. NVIDIA executive and fellow Armenian-American Rev Lebaredian provided his company’s mid-tier Blackwell-200 chips to supply the requisite computing power. The US Department of Commerce granted export licenses for the B200 chips to Armenia in November 2025 for deployment in what was then a $500 million endeavor.

Commerce expanded that quota when Vice President JD Vance visited Yerevan in February 2026 and announced a license to export 50,000 more higher-performing chips. That commitment coincided with Firebird.ai and NVIDIA’s announcement of a second computing hub in Armenia, with investments reportedly valuing the project at $4 billion. Firebird announced the opening of its first computing hub on August 8, complete with a congratulatory message from NVIDIA CEO Jensen Huang.

Another notable guest at the announcement was Kazakhstan’s deputy prime minister for artificial intelligence and digital development, Zhaslan Madiyev. Just weeks earlier, he negotiated a $10 billion agreement with Firebird.ai and NVIDIA—similar to the Armenia deal but larger in scale—to build advanced AI infrastructure in Kazakhstan as part of the country’s “Data Center Valley” initiative that aims to make Astana Central Asia’s undisputed leader in compute power. While Firebird.ai and NVIDIA’s leadership clearly had an additional interest in helping Yerevan build out its computing capabilities given their Armenian heritage, the Kazakhstan deal appears to be based purely on business fundamentals. 

This should all be great news for the Trump administration, which has made AI development and secure technology supply chains key elements of its foreign policy. This approach is a sensible one as China continues to develop its state-backed AI models and advanced technologies. To its credit, in June, Kazakhstan joined the State Department’s signature Pax Silica initiative, which aims to forge US-led international partnerships on all segments of the advanced technology value chain, from critical minerals to semiconductors to frontier AI models. In addition, Astana joined many Pax Silica countries in signing the State Department’s AI Opportunity Statement, which outlines pro-innovation, industry-driven policies for AI development and partnerships. 

But Kazakhstan is also attempting to extend its “multi-vector foreign policy” to the AI sphere. In July, President Kassym-Jomart Tokayev joined a similar China-led pledge in Shanghai that promoted Beijing’s emphasis on developing more open but less secure AI models. Kazakhstan is the only country said to be signed up to both initiatives. 

Possibly motivated by Tokayev’s China turn, the State Department has reportedly drafted a letter with thinly veiled threats to remove countries deploying Chinese AI technology from Pax Silica. According to Reuters, the letter reads: “signature of the Pax Silica Declaration is not merely a membership subscription, but a commitment…It cannot be held alongside membership in duplicative initiatives whose expectations conflict with our own.” 

State Department officials have so far declined to comment on the letter, which remains unpublished. Yet it nevertheless laid bare some of the tensions in American AI policy. The AI Opportunity Statement pledges to “enable industry partnerships and investments in next-generation data centers,” yet the State Department letter would seemingly call into question Firebird.ai and NVIDIA’s Kazakhstan deal. Doing so could wipe out billions of dollars in investments, future profits, and intangible amounts of diplomatic goodwill. 

Even if US policy doesn’t unwind the Firebird.ai-NVIDIA-Kazakhstan deal, the uncertainty it introduces could have negative strategic and commercial effects. Upstream in the AI value chain, for example, American mining companies assessing Kazakhstan’s rich geological prospects might likewise hold back on investment decisions, potentially depriving the US of critical minerals flows down the road. The Trump administration’s commercially focused Eurasia strategy has so far led to more than $100 billion in commercial agreements, with more in the pipeline; introducing policy uncertainty from Washington risks undercutting that positive approach.

If truly forced to choose between the United States and China as its AI partner, Kazakhstan and many of its Central Asian neighbors would almost certainly choose Beijing. China’s physical infrastructure in the region is extensive (Kazakhstan was the first country to join the Belt and Road Initiative) and growing; Beijing maintains deep investment and financial ties with Astana and other regional capitals, and can flex diplomatic attention and coercion that the United States is unlikely to match. Zooming into the model level, Kazakhstan and its neighbors could conclude that China’s cheaper, customizable open-weight models offer a better productivity proposition in the near term than the US’ exquisite, closed models.

The US government is right to warn its partners about the security risks of deep AI cooperation with China. From an AI safety perspective, Chinese models tend to have fewer guardrails than top American models, raising the specter of deleterious outcomes and catastrophic risk. China is known to embed backdoors in digital applications and infrastructure products and will certainly do so with advanced compute. But demands that countries be “with us or against us” are too blunt a policy tool to advance US-led AI partnerships with emerging economies.

Instead, the United States should outcompete China by building out compute power and advanced technology supply chains. While American companies will not beat Chinese firms on price, the US government could do more to syndicate projects across the value chain. 

Building on Firebird.ai and NVIDIA’s infrastructure-plus-chips model, complementary projects could modernize network infrastructure, improve energy and water efficiency, offer advanced cybersecurity solutions, improve access to AI-powered applications, and provide AI literacy training. This would likely require greater interagency coordination with the US Development Finance Corporation, the Export-Import Bank of the United States, and the US private sector to buy down risks and financing costs so companies can make projects bankable. That’s a tall order, but the opportunity costs of forcing countries to choose between the United States and China and the security risks of Chinese tech warrant a more comprehensive approach.

US companies are rightly driving American AI development in Eurasia. Led by Firebird.ai’s compute infrastructure agreements in Armenia and Kazakhstan, the United States is positioned to provide high-quality advanced technologies to emerging economies. US government policies should reinforce these private sector gains by creatively packaging and marketing the US AI stack abroad. Doing so would make it easier for potential partners to adopt American technologies and cut deals with American companies, ideally at the expense of less-regulated Chinese options. Rather than driving countries into China’s arms, the US government should look to multiply the mutual benefits of American AI offerings abroad.

About the Author: Andrew D’Anieri

Andrew D’Anieri is associate director of the Atlantic Council’s Eurasia Center, where he facilitates the Center’s work on Ukraine and Russia and leads its Central Asia programming. He oversees the Russia Tomorrow paper series and contributes to the Center’s work on Ukraine’s reconstruction and business and governance in Eurasia. He has written for The Hill, The New York Post, The Diplomat, and The National Interest. He is a regular commentator on BBC World News and has also been interviewed by news outlets including Sky News, Newsweek, ABC News, and Ukraine 24. He is a member of the Eurasia Foundation’s Young Professionals Network.

The post How to Compete with Chinese AI in Central Asia appeared first on The National Interest.