U.S.-China AI Rivalry Tests Countries Seeking Chips, Investment and Independence

Kazakhstan puts a price on the AI balancing act

Kazakhstan signed agreements on June 15, 2026, for an artificial intelligence infrastructure project that could attract up to $10 billion, including a planned data center in Ekibastuz equipped with Nvidia processors. Its diplomatic ambitions are just as large: secure access to American technology while helping shape an AI organization promoted by China.

The Central Asian country joined the U.S. technology partnership Pax Silica in June. Three weeks later, its AI minister, Zhaslan Madiyev, went to Shanghai for a signing ceremony that made Kazakhstan a founding member of China's World AI Cooperation Organization.

For a country of about 21 million people, the attraction is concrete. American partnerships offer advanced chips, computing infrastructure and potential jobs. Chinese partnerships bring investment, more affordable AI technology and a voice in an organization whose other 28 founding members include Pakistan and Russia.

Washington is warning that participation in competing AI initiatives could put a country's status as a trusted partner at risk. Beijing is presenting its organization as an inclusive alternative. Countries seeking both increasingly have to consider whether a technology purchase or diplomatic agreement could affect access to the other side.

Kazakhstan's position captures a wider struggle stretching from Indonesia and Singapore to Saudi Arabia and Europe. Governments want the benefits of AI without giving another country decisive control over the processors, software and computing services their businesses and public institutions will need.

What the $10 billion project actually involves

Kazakhstan's agreements involve Firebird, an American startup backed by Nvidia, and state operator Kazakhtelecom. The planned Data Center Valley would be built in Ekibastuz, an energy hub in the northeast where access to inexpensive electricity is a major attraction.

The first phase is valued at roughly $5 billion, including $1 billion from Kazakhtelecom. It aims to open commercially in 2027 with a site rated at 125 megawatts. Megawatts describe the scale of its electrical power requirements, a central concern because large AI computing facilities consume substantial amounts of electricity.

The headline $10 billion represents potential investment across several agreements. The strategic cooperation framework sets out intentions without creating a binding investment obligation, while the Kazakhtelecom and Firebird term sheet is binding. Financing, power delivery and chip shipments still need to come together for the first phase.

Expansion plans call for at least 300 megawatts, with ambitions eventually reaching toward one gigawatt, or 1,000 megawatts. The project could house as many as 100,000 Nvidia graphics processing units, including GB300 and Vera Rubin technology. These processors, usually called GPUs, perform many calculations at once and are central to training and running advanced AI.

The proposed 300 megawatts would be 2.4 times the initial 125 megawatts. That gives a sense of the expansion being discussed, although the larger figures remain development goals.

Why Washington and Beijing want partners

The United States is organizing Pax Silica around trusted technology partners and secure supply chains. For participating countries, the potential benefits include closer technology cooperation, manufacturing opportunities and a place in the networks that supply AI infrastructure.

The U.S. State Department has warned that countries significantly involved in initiatives advancing competing visions of superintelligence could undermine their standing as trusted Pax Silica partners. Superintelligence describes a proposed future form of AI with abilities beyond those of humans across a wide range of tasks.

China's embassy in Washington has argued that the two powers should use their respective AI strengths cooperatively and that the technology should be accessible beyond major powers. Beijing is also seeking partners while facing U.S. restrictions on access to advanced AI chips and equipment used to manufacture them.

Kyle Chan, a scholar at the Brookings Institution's John L. Thornton China Center, sees different diplomatic starting points. Washington already has a strong alliance network, while Beijing is working to attract countries to an alternative grouping. Countries able to work with either side therefore have bargaining power.

A short timeline of Kazakhstan's strategy

Kazakhstan's moves bring together diplomatic membership, investment agreements and plans to sell computing services abroad. The key developments span several dates:

• 2025: Chinese direct investment in Kazakhstan reached a record $2.8 billion, according to local officials.
• June: Madiyev visited Washington as Kazakhstan joined Pax Silica.
• June 15, 2026: Agreements for the proposed AI infrastructure project were signed in Astana.
• Three weeks after the Washington visit: Madiyev attended the Shanghai signing ceremony for China's AI organization.
• 2027: The first 125 megawatts of Data Center Valley are targeted to begin commercial operations, alongside Kazakhstan's planned start of computing exports.

Officials aim for $3 billion in annual computing exports. In practical terms, customers abroad would pay to use processors located in Kazakhstan rather than install all the necessary equipment themselves. Kazakhtelecom head Bagdat Mussin has described 2027 as the hoped for starting point.

Former Kazakh diplomat Dana Masalimova, an independent scholar affiliated with Harvard University's Davis Center for Russian and Eurasian Studies, says the country is spreading its risks rather than accepting a choice between two powers. She also describes its position as vulnerable and highly risky.

Kazakhstan has electricity and mineral resources that make it attractive to technology investors. Its challenge is to turn those advantages into lasting income and useful domestic services while keeping access to both American and Chinese partnerships.

Indonesia and Singapore seek room to maneuver

Indonesia has joined China's AI initiative while continuing to pursue American technology partnerships. Its nickel and copper resources give it assets relevant to technology supply chains, although mineral wealth alone cannot provide the chips, software and expertise needed for advanced AI.

Mari Pangestu, vice chair of Indonesia's National Economic Council, explained the country's preference while speaking to reporters at a University of Hong Kong business school event. She said Indonesia wanted to benefit from both powers:

Simply put, we should be able to be part of both sides and choose what is best for the system.

Singapore is considering joining China's AI organization even as American companies expand their presence there. Anthropic is establishing a new Asia office in the city state, adding another American AI business to its technology sector.

Indonesia and Singapore illustrate different sources of bargaining power. Indonesia offers resources and a large market. Singapore offers an established location for international technology operations. Both are trying to preserve commercial relationships across the rivalry rather than organize their entire AI strategy around one supplier country.

Saudi Arabia combines American chips with Chinese models

Saudi Arabia is pursuing advanced Nvidia computing infrastructure while also adopting technology built on Chinese open models. Saudi AI startup Humain launched a platform using those models in September.

Hardware and models serve different purposes. Chips perform the calculations, while an AI model is the trained software that uses those calculations to generate answers, analyze information or carry out other tasks. A country can therefore purchase American processors and run software developed elsewhere on them.

Open models give users access to key components of the trained system so they can run or adapt it themselves, subject to licensing conditions. That can offer governments and companies greater control over deployment and reduce dependence on a service operated entirely by a foreign supplier.

Saudi Arabia's approach helps explain why the emerging blocs are difficult to separate neatly. Governments may prefer one country's hardware, another country's models and their own domestic data centers. Diplomatic pressure can complicate a combination that otherwise makes commercial sense.

Europe and Japan face different limits

France is pursuing greater technological independence through Mistral, one of the most ambitious AI developers outside the United States and China. Its ability to develop models at home gives France an option that many other governments lack.

Britain and Germany have large economies and substantial military capabilities, but Carnegie Endowment for International Peace scholars Anton Leicht and Sam Winter-Levy describe them as lacking comparably powerful domestic frontier models. Frontier models are systems operating near the leading edge of AI capability.

The scholars argue that U.S. partnerships may be crucial to securing data center access and chip agreements. For such countries, decisions made in Washington and Beijing can shape access to advanced systems even when their own governments have limited influence over those decisions.

Japan remains closely aligned with Washington on strategic and security matters, including advanced semiconductors, trusted AI and standards. Takahiro Tsuchiya, a professor at Kyoto University of Foreign Studies, nevertheless expects room for inexpensive Chinese models in applications carrying lower risks.

France's domestic model development and Japan's selective use of foreign technology represent different routes toward greater autonomy. Both still operate within an industry where the most advanced computing resources and AI systems are concentrated in a small number of companies and countries.

China extends its reach through financing

China's international AI role extends beyond the sale or release of models. Research by AidData, a laboratory at William & Mary in Virginia, found Chinese government grants and loans supporting projects and activities using or enabling AI in at least 77 countries with low or middle incomes.

Major recipients identified in that research include Nigeria, Zambia, Ghana, Bangladesh, Senegal, Cameroon and Kenya. Financing and technical assistance give countries another route to building the infrastructure and skills needed to use AI.

Alberto Lemma, a researcher at ODI Global, argues that China's open approach to models can help countries reduce their dependence on the United States. Affordable software matters especially where governments and businesses have limited budgets for computing services.

Research from the Pew Research Center also points to more positive views of China among people in countries with middle incomes on AI and its regulation. Such public attitudes form part of the setting in which governments choose technology partners.

The United States retains an advantage in the most advanced AI systems, while China is working to narrow gaps in chip production and AI capability. Countries choosing Chinese financing or software still have reasons to seek American technology for their most demanding computing needs.

What turns an AI agreement into useful services?

Kazakhstan's domestic plans show why governments want more than membership in an international organization. Madiyev, who is deputy prime minister and minister for AI and digital development, says the country intends to develop five levels of AI activity, from energy and chip infrastructure to language models and applications.

Madiyev puts the potential global AI economy above $50 trillion, an estimate of opportunity rather than current revenue. He sees particular promise in practical applications that help businesses operate more efficiently and people complete everyday tasks.

Freedom Holding chief Timur Turlov says his company plans a data center partnership with Nvidia and an AI assistant inside its superapp. The proposed assistant would help customers navigate services ranging from investments and insurance to shopping, hotels and flights. The company already uses AI in administration, reporting and process management.

At AI & Digital Bridge 2026, Sofreh Capital founder and chief executive Shervin Pishevar also expressed an intention to open offices and invest in Kazakhstan. Kazakhstan's National Information Technologies company and Georgia's Digital Governance Agency signed a memorandum covering cooperation on digital government, public services and possible joint technology projects.

For investors, the immediate decisions concern where money will flow into processors, electricity, cloud computing and data centers. Kazakhstan's project shows why the terms matter: a binding agreement, a financing plan and an operating facility represent different stages of development. Firebird says its Armenian AI factory was built in about six months and was due online in July, offering a comparison for its construction ambitions.

Alok Mehta, director of the Wadhwani AI Center at the Center for Strategic and International Studies, says countries are weighing the strengths and weaknesses of competing technology systems while seeking AI sovereignty. In everyday terms, that means having enough control over infrastructure, software and access to keep essential services running on terms a country can accept.

Kazakhstan's next concrete milestone is the planned commercial opening in 2027. Delivering powered facilities, obtaining chips and attracting paying customers will determine how much of its diplomatic balancing becomes a functioning industry.

Key Points

• Kazakhstan joined the U.S. Pax Silica partnership and then became a founding member of China's AI organization three weeks later.
• Its AI infrastructure agreements could attract up to $10 billion, with a first phase valued at roughly $5 billion.
• A 125 megawatt facility in Ekibastuz is targeted to open commercially in 2027.
• Washington has warned that significant participation in competing AI initiatives could affect trusted partner status.
• Indonesia, Singapore and Saudi Arabia are pursuing combinations of American and Chinese technology relationships.
• AidData identified Chinese grants and loans supporting AI activity in at least 77 countries with low or middle incomes.
• Access to financing, electricity, advanced chips and usable software will shape how much independence countries can achieve.