China’s second Belt and Road may be built on AI

China’s recent push on artificial intelligence (AI) across the Shanghai Cooperation Organisation (SCO) and BRICS points to a larger economic strategy. At the SCO summit in Bishkek on 1 Sept, President Xi Jinping proposed an international AI application cooperation centre and 100 technology cooperation projects over the next three years. Less than two weeks later, at the BRICS summit in New Delhi, he called for an AI open-source community, AI-enabled industrialisation and deeper cooperation across technology supply chains.

Together with China’s July AI Cooperation and Development Action Plan and the launch of the World AI Cooperation Organization (WAICO) — which promises affordable computing services for developing countries, interconnected computing infrastructure, green computing facilities, multilingual datasets and international open-source communities — these initiatives suggest that Beijing wants to internationalise more than Chinese models. It is trying to expand an entire Chinese “AI stack” (a collection of technologies, frameworks and infrastructure components that facilitate using AI systems as defined by IBM), from applications and cloud services down to servers, accelerators and the semiconductor base beneath them. 

Externalising domestic industrial capacity

China may therefore be approaching a second Belt and Road moment. The parallel is economic rather than literal. After the 2008 financial crisis, China’s investment-heavy stimulus accelerated infrastructure construction and expanded capacity in steel, cement, machinery and engineering. The Belt and Road Initiative (BRI) later helped internationalise some of those capabilities by creating overseas demand for Chinese contractors, equipment, finance and standards. It served many other strategic purposes, but externalising domestic industrial capacity was one important economic function.

The sectors have changed. The underlying problem is becoming familiar. China has poured resources into AI models, computing centres, cloud infrastructure, semiconductor fabs, servers and domestic chip development. A 2025 IMF working paper estimated the fiscal-equivalent cost of China’s broader industrial policies at about 4% of GDP annually and found that policy-induced factor misallocation reduced aggregate productivity by about 1.2%. In AI infrastructure, some government-backed computing centres have operated at utilisation rates of only 20% to 30%, prompting Beijing to develop a national network to redistribute surplus computing power.

Unevenly sourced down the AI stack

Yet China’s AI stack is far from uniformly oversupplied. At the frontier, it still faces shortages. Advanced AI accelerators, high-bandwidth memory (HBM) and leading-edge manufacturing equipment remain bottlenecks, reinforced by US export controls. Chinese AI chipmakers have recently raised prices as HBM shortages increased production costs.

Further down the stack, however, China has rapidly expanded cloud capacity, data centres, servers, telecom equipment, mature-node semiconductors and increasingly capable domestic AI hardware and models. Its challenge is increasingly asymmetric: scarcity at the frontier alongside large state-backed capacity and uncertain commercial returns across much of the rest of the system.

That makes overseas demand strategically important. A viable AI ecosystem needs paying users, workloads, developers and recurring revenues. If Chinese models, servers and computing centres remain concentrated in a domestic market marked by intense price competition and duplicated investment, technological self-reliance risks remaining an expensive state-supported project. A sustainable stack requires demand at the top that can finance upgrading at the bottom.

Global South helps generate demand

The Global South offers Beijing a way to create that demand. China does not need to begin by persuading Brazil, Indonesia or African governments to buy Chinese semiconductors. It can begin with open-source models, local-language applications, training programmes, cloud services and industrial AI. Those applications create computing demand. Computing creates demand for data centres, servers, networking equipment and accelerators. Those systems pull through a wider semiconductor supply chain, including power management, connectivity, storage and edge-processing chips that do not require the world’s most advanced fabrication processes.

Demand can cascade down the stack. This gives China’s emphasis on “inclusive AI” a commercial logic. Many emerging-market applications — government services, industrial inspection, logistics, agriculture, smart cities and local-language inference — do not require frontier training clusters. Cost, availability, local deployment and financing can matter more than absolute computing performance. China can compete by combining open models with lower-cost hardware, telecom infrastructure, cloud platforms and project financing.

If that package gains scale, the returns move in the opposite direction. More users raise utilisation; larger installed bases increase hardware demand; higher volumes spread fixed costs; overseas revenues support research and development; and deployment generates learning that can improve subsequent generations of Chinese systems.

That is the commercial loop Beijing needs to close. The contrast with the US-centred AI ecosystem is instructive. The US sits at the centre of a frontier stack linking American chip designers and hyperscalers with Taiwanese foundries, South Korean HBM, Japanese materials and equipment, and Dutch lithography. It is an ecosystem capable of monetising technological scarcity through high margins at critical chokepoints.

China has stronger incentives to monetise scale. Its emerging proposition combines increasingly capable open models with affordable deployment and a progressively more domestic hardware base. Washington is concentrating frontier capabilities among trusted partners. Beijing is trying to expand the market beneath the frontier and use that market to finance its climb upward.

Here the Belt and Road analogy becomes especially useful. The first wave used overseas infrastructure demand to pull through China’s industrial supply chain: projects created business for contractors, machinery makers, steel producers and financiers. A digital second wave could use overseas AI demand to pull through China’s technology stack: applications generate cloud and computing demand; computing drives hardware demand; hardware supports semiconductors, equipment and standards.

New scaffolding: BRICS, the SCO and WAICO

The new model could also rely less heavily on sovereign lending. Cloud subscriptions, service contracts, joint ventures, local capital and corporate investment can create recurring commercial revenues. Software ecosystems add another advantage: developer networks, compatibility requirements and switching costs can make technological relationships more persistent than a completed railway or port.

BRICS, the SCO and the new WAICO can provide institutional scaffolding. 37 countries have agreed to establish the latter, strengthening Beijing’s effort to frame AI access and cooperation as a Global South issue. These platforms provide access, training networks, standards cooperation and political legitimacy.

Their members are unlikely to form an exclusive Chinese technology bloc. India, the UAE, Brazil and Indonesia have strong incentives to preserve access to American, European, Japanese and Korean technologies. Beijing does not need exclusivity. It needs Chinese technology to become a normal option whenever emerging economies build AI capacity.

The first Belt and Road helped turn capabilities created by China’s investment-heavy growth model into global infrastructure. China’s AI diplomacy may now be attempting a similar transformation for its technology economy.

The next phase of AI competition will therefore depend on more than who builds the most advanced processor or trains the strongest model. It will also depend on who can create the largest commercially viable ecosystem around the technology they can actually supply.

China’s second Belt and Road may be built on AI because the Global South could provide what its domestic technology drive increasingly needs: users, workloads, scale and revenues that flow through the stack — from models at the top to semiconductors at the bottom — and help finance the next climb up the technology ladder.

Related: From lender to partner: China’s BRI 2.0 is harder for the West to contain | High quality, high anxiety: Southeast Asia’s BRI paradox