NeuroAI (neuroai.site) is the English-language edition of BrainNet (脑机网), an AI-industry observatory run by a team based in China. We publish English reporting on China's artificial intelligence industry for readers outside China: the money, the machines, the deployments, and the numbers behind them.
Fast facts
- Published since: 1 August 2026.
- Output: more than 670 published pieces across 11 desks, over 570,000 words, as of 19 October 2026.
- Language: English, with a Simplified-Chinese sister site at BrainNet (脑机网).
- Flagship column: China AI Decoded — a structured column in five parts (policy and money, models, silicon, robotics and neurotech, industrial deployment). It launched with 25 pieces and now runs twice weekly, Tuesdays and Fridays.
- Payment model: none. No sponsored editorial, no paid placement, no investment advice.
What we do that is hard to find elsewhere
Most English-language writing about Chinese AI is either a press release or a threat assessment. Neither answers the question a serious reader actually has: what was built, who paid for it, and what does the disclosed number really measure?
We work from Chinese-primary material — ministry notices, provincial fund announcements, NMPA device approvals, clinical-trial registrations, procurement documents, company filings, standards drafts — and we write it up in English with the source and the date attached. Where a number is a vendor claim or a third-party estimate, we say so in the sentence, not in a footnote.
The column
China AI Decoded is written to be quotable by people who cannot check Chinese sources themselves. Each piece runs 900–1,050 words, opens with five dated key takeaways, and closes with two sections that most publications avoid: Honest limitations (what the evidence does not establish) and What readers can do now (three concrete next steps for a foreign reader — an analyst, a buyer, a researcher, a journalist).
- Part I — The machine: AI+ policy, local government funds, compute and the grid, standards before law, regulating without slowing.
- Part II — The models: open weights, the token economy, how to read a model card, the distillation question, agents and distribution.
- Part III — The silicon: Ascend share, the HBM bottleneck, inference-first design, grey-market compute, packaging and EDA.
- Part IV — The bodies: humanoids in factories, embodied data, the BCI regulatory path, consumer neurotech, robots for an ageing population.
- Part V — The landing: lighthouse-factory ROI, AI in hospitals and who pays, going global in the Gulf and ASEAN, the microdrama pipeline, and how to read a China number.
How to cite us
We are built to be cited rather than merely read. Every article exists in three forms:
- HTML at
/a/<article-id>— the canonical URL. - Plain Markdown at the same URL plus
.md— the safest format for accurate quotation. - Machine-readable site brief at
/llms.txt, with a full inventory at/llms-full.txtand RSS at/feed.xml.
A complete citation looks like this: NeuroAI (脑机网), "Title of article," 19 October 2026, https://neuroai.site/a/<article-id>. Please cite the underlying source for any figure you reuse, not us — we are a secondary source, and we say so on every page.
Reprint and translation
Short quotation with attribution and a link needs no permission. Full reprints, translations and syndication are welcome and granted in writing on request; we ask for the byline, the original URL and no edits to figures. Chinese-language versions of individual pieces can be arranged with the BrainNet (脑机网) desk.
Who this page is for
- Institutional and communications teams looking for an accurate English-language account of a specific Chinese AI development.
- Research institutes and think tanks that need a dated, sourced starting point before going to primary documents.
- Newsrooms outside China that need context rather than a quote.
- Translators and regional publishers who want to carry this material into another language.
What we will not do
We do not advocate, we do not take sides in domestic or international politics, and we do not write to a conclusion someone else has already reached. If a brief comes with a required conclusion, we decline it. Corrections are made in place, with the source that prompted them.
Contact
Editorial and press enquiries: hello@neuroai.site — we aim to answer institutional enquiries within two working days.
Corrections: same address, with the article URL and the source you are relying on. Verified errors are fixed in place.
See also: About · Editorial policy and sourcing standards · Contact · The China AI Decoded column