Synthesis history

Two substantive additions this pass. Tom's Hardware reports Chinese memory module vendors are switching supply chains from Samsung, Micron, and SK Hynix to CXMT chips [^40434], a domestic consolidation that builds CXMT'…

No new themes this pass. Additional reporting from The Next Web, Engadget, and daily.dev [^40140][^40141][^40142] confirms the DeepSeek chip development story already established by Reuters last pass. A CSIS policy anal…

Three substantive developments this pass. First, Reuters (via Ars Technica and Semafor) reports DeepSeek has been developing its own AI inference chip for approximately one year, hiring engineers and meeting with hardwa…

Academic research summarized in [^39549] adds a download-share data point: China surpassed the US in open model downloads in summer 2025 and held a 1.15B-to-723M lead by March 2026, with Qwen — not DeepSeek — as the pri…

New items this pass add limited substance. InferenceX's evaluation of DeepSeek V4 on Huawei Ascend NPUs [^37712] extends the Ascend software ecosystem story — the hardware is being benchmarked for frontier model inferen…

Two hardware developments are new this pass. Huawei is preparing to sell Ascend AI accelerator chips in South Korea in Q4 2026 — its first attempt to market AI compute outside China — which extends the story from Chines…

Three developments are new this pass. First, Meituan confirmed LongCat-2.0 was trained from scratch on 50,000 domestic Huawei Atlas-950 chips—removing prior uncertainty about whether domestic hardware could support larg…

A three-way export control debate is the main new development: Jensen Huang argues controls function as industrial stimulus for Huawei and cede operating-layer standards; Perplexity CEO Srinivas argues controls compress…

A J.P. Morgan report (via Rohan Paul) adds the most granular longitudinal baseline yet: Chinese firms held over 45% of OpenRouter traffic by April 2026, up from under 2% in late 2024, and Chinese models are up to 50x ch…

Two substantive new items define this pass. The Information's report (via Rohan Paul) that Anthropic's Mythos model preview prompted DeepSeek CEO Liang Wenfeng to pursue the $7.4B fundraise [^34781] adds new context to …

Two substantive new items define this pass. UBS survey data [^34600] showing 60% of enterprise AI budget monitors are shifting to Chinese models moves the adoption story from developer-traffic statistics to corporate bu…

Two new semiconductor hardware items from SemiAnalysis (June 23) expand the thread's scope beyond AI models and capital. CXMT — approaching China's largest semiconductor IPO — is shown to have built its DRAM technology …

Nathan Lambert's analysis calling GLM-5.2 the first credible open-weight agent and Zvi Mowshowitz's skeptical take — arguing heavy distillation from Claude Opus, a persisting frontier gap, and commercial niche limitatio…

ASML issued a formal public denial this pass, characterizing the US government's EUV concern as 'inaccurate and damaging to our reputation' [^32032][^32041] — a harder public stance than the private response noted previ…

Three substantive developments are new this pass. A US-ASML dispute emerged over whether a banned EUV lithography tool reached China — Commerce Secretary Lutnick personally raised the concern with ASML leadership; ASML …

DeepSeek's $7.4B raise at a $50B valuation on June 16 is the most significant new development: it makes DeepSeek China's most valuable AI startup and adds financial scale to the infrastructure investment thesis already …

New items this pass were mostly empty Twitter posts and untitled URL shares with no extracted claims, plus five data center articles about DeepSeek (29808–29812) whose titles are consistent with the existing infrastruct…

MiniMax M3 (428B MoE, day-0 open weights, CUDA/ROCm support, 9x faster prefill vs. M2.7) released June 13, adding a concrete recent model release to the story [^28455][^28459]. DeepSeek's disclosure of MW-to-GW scale in…

New items are primarily social media amplification (X, Facebook, Instagram, LinkedIn) of the existing OpenRouter story, adding no new primary data. Arnaud Bertrand introduces a specific causal attribution — a particular…

New items add a global AI model usage ranking showing China holds 6 of the top 9 spots [^27498], a second data source consistent with the OpenRouter share figures. Secondary coverage has amplified the OpenRouter story w…

Chinese AI models reached 61% of weekly token consumption on OpenRouter by mid-2026, up from a minority share through most of 2025, with American AI startups now routing significantly more application traffic to Chinese…