Version 8

Chinese AI providers are competing across model capability, open-weight releases, capital, and physical infrastructure simultaneously. Chinese models hold 61% of weekly token market share on OpenRouter [1], with China occupying 6 of the top 9 positions in a spring 2026 global usage ranking [2]. Two major open-weight releases in mid-June 2026 — MiniMax M3 [8] and GLM-5.2, which tops the Artificial Analysis Intelligence Index v4.1 [10] and drew a surprised reaction from the Vercel CEO over its coding ability [11] — coincide with DeepSeek's $7.4B raise at a $50B valuation [13] and disclosed plans for MW-to-GW scale data center construction [14]. A US-ASML dispute over whether a banned EUV lithography tool reached China remains unresolved: the US Commerce Secretary personally raised the concern with ASML leadership [17], and ASML formally denied ever shipping an EUV machine to China, calling the reports 'inaccurate and damaging to our reputation' [18].

Chinese AI providers are contesting multiple layers of the stack simultaneously — model performance, open-weight distribution, capital, and compute infrastructure. The EUV export controls dispute matters because if the US concern is founded, it would mean China's hardware access is less constrained than the export control regime implies, changing the long-term semiconductor parity calculation.

Through most of 2025, U.S. models led weekly token consumption on OpenRouter, the API routing platform widely used by developers and AI startups. From early 2026, Chinese models became the primary growth driver, reaching 61% of the platform's weekly token market share by mid-2026 [1]. A separate global AI model usage ranking from spring 2026 shows China holding 6 of the top 9 positions [2]. Observers differ on causes: Rohan Paul treats the OpenRouter gains as a structural competitive shift [3]; Arnaud Bertrand attributes the 2026 turn to a specific Chinese model release [4]; developer-community commentators argue that developers route to whatever model delivers the best cost-performance ratio at a given moment, making the gains real but conditional on price [5]. The gains trace at least partly to OpenAI's June 2024 cutoff of API access for Chinese developers, which Baidu, Alibaba, and Moonshot AI actively exploited to recruit displaced users [6][7].

The open-weight model pipeline in mid-June 2026 is dense. MiniMax M3, released June 13, has 428B total parameters with 23B activated, was available on HuggingFace on day 0 with open MXFP8 weights, and achieves 9x faster prefill than its predecessor via block sparse attention [8][9]. GLM-5.2, released June 16 by Z.ai under an MIT license, is a 753B-parameter mixture-of-experts model that tops the Artificial Analysis Intelligence Index v4.1 with a score of 51, expands its context window to 1 million tokens, and has attracted attention across developer communities — the Vercel CEO described its coding ability as 'almost shocking' [10][11]. Both releases carried day-0 open weights. Milk Road AI offers a dissenting view: open-source models will not win the AI race, and the market misread DeepSeek's early 2025 launch as validation of open-weight approaches when the real story was different [12].

At the infrastructure and financial layer, DeepSeek completed a $7.4B financing round at a $50B valuation on June 16, making it China's most valuable AI startup; founder Liang Wenfeng held approximately 90% pre-round and personally contributed around $3B [13]. DeepSeek had earlier disclosed job openings for MW-to-GW scale data center design, which SemiAnalysis interprets as a move from software-first to vertical integration of physical compute [14]. SemiAnalysis also noted that China established a Space Computing Industry Innovation Center — government-chartered and led by Beijing University of Posts and Telecommunications — before Elon Musk announced the AI1 satellite, a development that received little Western coverage [15]. A hardware teardown found SMIC's N+3 process reached TSMC N6-class logic density via DUV multi-patterning, but at higher cost and process complexity, with the resulting Kirin 9030 Pro trailing current flagship SoCs [16].

A separate dispute concerns export controls directly. US Commerce Secretary Howard Lutnick personally told ASML executives that the Trump administration believes an advanced EUV lithography machine may have reached China [17]. ASML denied this categorically in a public statement, saying it has never shipped an EUV machine to China and that the reports are 'inaccurate and damaging to our reputation' [18][19]. The disagreement is unresolved. US export controls apply to ASML's EUV machines in part because they incorporate US intellectual property [20]; ASML separately sells older DUV machines to China — a distinction central to the current dispute.