How Chinese AI Labs and Hardware Vendors Go Global at Once

The macro picture

Chinese tech companies are quietly taking different routes to the same place: a larger share of the global AI economy. AI labs, optical-module makers, and warehouse-robot vendors are all expanding overseas in 2026, but the way each one is doing it reveals more about the structure of the AI value chain than any aggregate trade number can.

Customs data for the first half of 2026 shows the obvious version of the story. Integrated-circuit exports nearly doubled in value to $177.3 billion, up 96 percent year on year, helped by global chip price inflation. Exports of automatic data-processing equipment rose 41 percent to$138.1 billion. Industrial-robot shipments climbed 18 percent to $929 million across 141 countries.

Those numbers are real, but they understate what is happening on the ground. The bigger story is in corporate filings, which show Chinese firms taking meaningful share in both the software and hardware layers of the global AI build-out.

AI labs: software that scales, enterprise that does not

The clearest split is between AI labs that sell software and AI labs that try to sell into large enterprises. Shanghai-based MiniMax generated $57.7 million, or 73 percent of its total revenue, outside mainland China in 2025, up from$21.3 million the year before. By the end of 2025, it had served more than 214,000 enterprise customers and developers from over 100 countries, and its flagship M2 model became the first Chinese model on OpenRouter to cross 50 billion in daily token consumption.

Enterprise-focused peer Zhipu AI tells a different story. Its 2025 annual report says revenue was "substantially derived" from Chinese customers, with on-premise deployments generating 534 million yuan, or 74 percent of sales. The Goldman Sachs thesis on Chinese AI models is that performance has caught up to Western models, but on-premise enterprise work still requires a local support network that only domestic labs can build cheaply.

The divide is structural. Lightweight consumer apps and developer-facing cloud services can ship globally with a small international team. Heavy enterprise deployments need implementation engineers, compliance reviewers, and integration partners in every region. Chinese labs that pick the first path can grow fast. The ones that try to do both end up growing slowly in both.

Optical modules: the boring, indispensable layer

If AI labs are the visible part of the Chinese export story, optical modules are the load-bearing wall. Zhongji Innolight and Eoptolink now sit at the top of the global optical-transceiver market, supplying 400-gigabit and 800-gigabit components to Amazon, Nvidia, and other US hyperscalers. Both companies generated more than 90 percent of their 2025 revenue overseas.

That position was not accidental. US and Japanese competitors largely exited the optical-transceiver market by 2020, and Chinese vendors stepped into a vacuum at exactly the moment cloud providers started building the AI clusters that needed those parts. Zhongji's US market alone delivered more than 57 percent of its revenue last year.

Warehouse robots: priced out at home, marked up abroad

Chinese mobile-robot vendors are exporting for a different reason. Research firm Interact Analysis says the saturated and highly competitive Chinese market is forcing leading players to look outward, and the price arbitrage is real. An autonomous forklift that sells for $15,000 to$20,000 in China can command $40,000 to$60,000 in Europe and North America, even after localization and support costs.

For vendors with international operations, the gap is already showing up in revenue mix. One leading Chinese warehouse-robot supplier generated 75 percent of its 2025 revenue outside mainland China, and almost 80 percent of its new orders came from overseas, with the Americas growing at more than 50 percent. Domestic-only competitors are reporting flat or declining margins.

What ties it together

The throughline is not nationalism, and it is not a unified industrial policy. It is the simple fact that the global AI build-out is the most capital-intensive infrastructure project since the mobile internet, and Chinese vendors are competitive in every layer where the moat is engineering and execution rather than regulatory capture.

The next test is whether Chinese AI labs can keep their software advantage as US export controls tighten, and whether the optical and robotics companies can hold their share once former US and Japanese competitors decide the market is worth re-entering. For now, the export numbers are moving in the right direction, and the reasons behind them are more durable than any single trade negotiation.