After threatening AI export bans, China allows DeepSeek to keep selling

China's Ministry of Commerce revised its Catalogue of Technologies Prohibited and Restricted from Export in mid-October to add 23 technologies related to large-scale models, intelligent agents, computing clusters and information processing, requiring Chinese companies to obtain licenses to export them, but state media reported last Friday that DeepSeek and MiniMax received authorization to continue exports. Open source-licensed technology is exempt from the restrictions, and the expanded catalogue lets Beijing block exports of AI it deems important sovereign capacity, such as potent bug-finding tools or military-usable systems. The move mirrors US restrictions on advanced GPU sales to China while preserving the export businesses of two credible challengers to American AI vendors.

China has proposed to restrict export of some AI technologies but quickly moved to allow some of its champion companies to continue shipping their wares abroad. Beijing’s Ministry of Commerce announced https://www.mofcom.gov.cn/zwgk/zcfb/art/2026/art 0a31a3c9ac404f7a9b0e917ce3412997.html the restrictions in mid-October by revising its Catalogue of Technologies Prohibited and Restricted from Export in China. The restrictions mean that Chinese companies must apply for a license to export their tech. New entrants on the list included “23 technologies related to large-scale models, intelligent agents, computing clusters, and information processing.” Tech issued under open source licenses is exempt from the restrictions. Last Friday, Chinese state media reported https://www.sina.cn/news/detail/5357144217225897.html that two companies – named elsewhere as DeepSeek and MiniMax – received authorization to continue exports. Both companies are credible challengers to American AI vendors, so letting them sell their wares abroad meets China’s goals of creating export industries and enhancing its tech sector. The expanded Catalogue also gives Beijing the ability to stop exports of any AI it feels represents important sovereign capacity, such as tools that are potent bug-finders or have military uses. China is not alone in restricting exports of AI technology: the USA has imposed on-again, off-again, bans on sales of advanced GPUs to China. AI datacenter bubble wobbles in Australia If the AI datacenter building bubble bursts, two recent Australian incidents may prove instructive. The first took place on October 9, when a company called Firmus called off its initial public offering. Firmus already operates four datacenters, two in Australia and a pair in Singapore. It plans several more, plus expansions of its existing bit barns. Meta has committed to spend over $700 million with the company. Private investors had already ploughed over $2 billion into the company. The float would have been Australia’s largest for 20 years and aimed to raise around $5 billion, giving Firmus a valuation over $40 billion. In the days before the float, investor sentiment suddenly soured. Would-be buyers reportedly https://www.abc.net.au/news/2026-10-10/how-firmus-the-biggest-ipo-in-three-decades-fell-over/107247100 asked Firmus to lower its share price and therefore valuation by perhaps a quarter, as they feared the company’s prospectus made very optimistic revenue predictions. Not helping matters was that one of Firmus’ founders did time after being convicted of insider trading. The company eventually cancelled its IPO. The other incident took place on Wednesday, when established datacenter operator NEXTDC announced https://asx.api.cmfyapp.com/asx-research/1.0/file/2924-03150896-2A1705181&v=undefined PDF reduced revenue forecasts because its chosen power supplier suspended work “because of community and stakeholder opposition.” That opposition related to installation of poles and wires needed to bring juice to an existing datacenter NEXTDC intends to expand. Protestors have blocked access to the site, and the local council has decided not to allow the power company to use relevant roads while community consultation continues. NEXTDC, which readers may recall designed one of its datacenters to resemble sets from Sean Connery flick The Hunt For Red October https://www.theregister.com/on-prem/2020/05/07/data-centre-reveals-it-modeled-interiors-on-the-hunt-for-red-october-sets/1483356 , slashed its revenue forecast and then saw its share price slump almost 40 percent. The company told investors it will be at least February before its power supplier can resume work. Hyperscalers mostly use third party datacenters in Australia. The nation’s government wants to get a slice of the investment on offer during the AI boom but has also set rules requiring new builds to bring their own renewable energy supply and pay for any grid enhancement costs. Canberra hopes that approach attracts sufficient interest that Australia ends up with onshore AI capacity that, even if it won’t be sovereign, will at least be local. Grab suggests shopping data could detect disease Singapore-based Grab, whose “super-app” offers rideshare, e-commerce, financial services, and more, has suggested https://engineering.grab.com/grabmart-data-for-faster-disease-outbreak-detection the data it collects could help to detect outbreaks of disease. “Retail purchases can supplement clinical surveillance because changes in demand for some over-the-counter products may appear before confirmed case reports,” the company wrote in a recent abstract, which raises the idea of tracking “changes in demand for products associated with symptoms, such as fever remedies or cough medicines.” The company’s idea is to analyze customer behavior on its GrabMart grocery app, in the hope that doing so yields “an additional signal when demand changes across groups of symptom-related products.” Grab is at pains to point out that it does not believe its data can detect disease alone, and that it has not deployed such a system. The company also calls for use of the smallest possible datasets. “The analysis should use only eligible product categories, coarse geographic areas, and aggregate time windows approved for the stated public health purpose. It should not expose raw transactions or personally identifiable information,” the company suggested. Grab also says such monitoring must be government-approved, targeted to particular health concerns, and time-limited. It also acknowledges that the idea is a thought bubble that public health experts should assess before deployment. “These applications should use aggregate trends only,” the company suggested. “The signal should not drive personalised medical advice, individual targeting, insurance decisions, or automatic changes to medicine search rankings.” What could possibly go wrong? South Korea to accelerate smart glasses privacy rules South Korea’s science minister Bae Kyung-hoon has promised to accelerate development of regulation governing smart glasses, to address unauthorized filming and the potential for the devices to be misused in other settings such as during academic exams. Speaking at a parliamentary forum, Bae said the government is already working on guidelines but plans to accelerate that effort, and that president Lee Jae Myung recently mentioned the need for regulation during a cabinet meeting. The minister added that the guidelines will be developed at a “pan-ministerial level,” suggesting the effort will attempt to create comprehensive regulations. IPO for India’s top telco vastly oversubscribed Jio Platforms is India’s top telco, having won over half a billion customers since its 2015 launch, helped by low price subscriptions and own-brand handsets, plus clever content bundling. The company will float this week by selling around a three percent stake in the company, and the IPO is likely to be rather successful as investors have expressed an interest to acquire 22 times more shares than the company will sell. Institutional investors are 70x oversubscribed. The IPO values the telco at just over $100 billion and aims to raise around $3 billion. ®