Chinese regulators have stepped up pressure on domestic companies to limit their reliance on U.S. artificial-intelligence chips, issuing guidance that discourages the use of Nvidia Corp.’s H20 processor and, in some cases, ordering a suspension of new purchases. Bloomberg reported that the Ministry of Industry and Information Technology recently circulated notices advising firms to avoid the H20, especially in government or national-security projects, while The Information said the Cyberspace Administration of China instructed tech giants to freeze orders until it completes a security review. Companies including Alibaba Group, ByteDance and Tencent Holdings were told to justify any need for the H20 over Chinese alternatives, according to people familiar with the directives. The notices stop short of a formal ban but add administrative hurdles that could slow deliveries and prompt buyers to switch to domestic chips made by Huawei or Cambricon. Beijing’s guidance also extends to Advanced Micro Devices’ MI308 accelerator, the reports said. The clampdown comes only weeks after Washington lifted export curbs on the H20 in exchange for Nvidia handing 15% of related revenue to the U.S. government, a deal intended to restore sales worth billions of dollars. Analysts at Bernstein now expect Nvidia’s share of China’s AI-chip market to fall to 55% this year, from 66% in 2024, as customers reassess orders amid regulatory scrutiny. Nvidia said the H20 “is not a military product or for government infrastructure” and denied allegations of backdoors that could allow remote tracking or shutdown. Nevertheless, local semiconductor suppliers gained on the news: Semiconductor Manufacturing International Corp. rose about 5% in Hong Kong trading, while Cambricon Technologies hit the 20% daily limit in Shanghai. The latest measures highlight the widening technology rift between Washington and Beijing despite the revenue-sharing arrangement and reinforce China’s campaign to accelerate adoption of home-grown chips in sensitive sectors.

Alibaba Group announced plans to spin off Banma Network Technology, its autonomous driving software unit, for an independent listing on the main board of the Hong Kong Stock Exchange. Alibaba will retain over 30% ownership in Banma following the proposed spin-off. Details of the transaction have not yet been finalized. The move reflects Alibaba's strategy to unlock value in its smart car business by allowing Banma to operate as a separate publicly traded entity.

Alibaba Group reported fiscal first-quarter revenue of RMB 247.65 billion (US$34.6 billion), missing analysts’ RMB 253.17 billion consensus, while adjusted earnings per share slipped 10% year-on-year to RMB 14.75. Net income nevertheless jumped 76% to RMB 42.38 billion, helped by investment gains, and the operating margin held at 14%. The Cloud Intelligence unit, now Alibaba’s main growth driver, posted a 26% revenue increase to RMB 33.4 billion. AI-related products recorded triple-digit growth for an eighth consecutive quarter and account for more than 20% of external cloud sales. Executives said the company has invested over RMB 100 billion in AI infrastructure and is prioritising scale over near-term margins in the segment. Separately, the company is testing a home-grown AI chip aimed at a broad range of inference tasks, a move that could reduce dependence on Nvidia processors amid tightened US export rules. The processor, fabricated by a mainland manufacturer, expands Alibaba’s push to build a full domestic AI hardware–software stack. Alibaba repurchased 667,200 ADRs for about US$10 million on 28 August and reiterated plans to channel cash into its “consumption” and “AI + cloud” pillars. The results and chip news sent the shares as much as 12% higher in US trading on Friday.

Alibaba Group Holding has developed a new artificial-intelligence chip aimed at inference tasks, the Wall Street Journal reported, citing people familiar with the matter. The move is intended to narrow the gap created by restrictions on Nvidia’s most advanced products in China. The processor is described as more versatile than Alibaba’s earlier in-house silicon and can handle a broader range of AI workloads while remaining compatible with Nvidia’s development tools, enabling engineers to reuse code written for Nvidia GPUs. It is currently undergoing testing. Unlike Alibaba’s previous AI chip, which was fabricated by Taiwan Semiconductor Manufacturing Co., the new part is being produced by a mainland Chinese foundry, underscoring Beijing’s drive to localize critical technology amid escalating U.S. export controls. China’s push for self-reliance in high-end semiconductors has accelerated since Washington tightened rules on shipments of Nvidia’s H20 and more powerful processors. Alibaba, the country’s largest cloud-computing provider and a major Nvidia customer, joins local peers such as Huawei in stepping up development of domestically sourced alternatives.

Nvidia Corp. shares fell as much as 1.7% in early U.S. trading on Friday after the Wall Street Journal reported that Alibaba Group Holding Ltd. is developing a new artificial-intelligence chip designed to fill what the paper called a "Nvidia void" in China. The report said the Chinese e-commerce and cloud-computing giant aims to provide an alternative source of advanced AI processors for domestic customers. Details on specifications or a launch timeline were not disclosed. Investors sold off Nvidia on the prospect of fresh competition in one of its most important growth markets.

Alibaba Group Holding Ltd. said it repurchased 667,200 of its American depositary shares on the New York Stock Exchange on Aug. 28 for about $9.99 million. The transaction forms part of the e-commerce giant’s ongoing share-buyback program aimed at reducing share count and returning capital to investors. Separately, an Alibaba executive reported that the company’s quick-commerce operation reached a peak daily order volume of 120 million in August, while the weekly average stood at about 80 million daily orders. The figures underscore rapid growth in the one-hour delivery service as Alibaba seeks to deepen engagement with consumers in China’s highly competitive online retail market.

Tencent Holdings Ltd. released four open-source variants of its HunYuan large language model on 4 Aug., offering parameter sizes of 0.5 billion, 1.8 billion, 4 billion and 7 billion. The lightweight models handle up to 256,000 tokens and feature separate fast and slow reasoning modes, allowing them to run on a single consumer GPU and target use cases ranging from smart-home speakers to in-car assistants and personal computers. Hours later, Alibaba Group’s Qwen team introduced Qwen-Image, a 20-billion-parameter text-to-image model based on the MMDiT architecture. Alibaba says the system achieves state-of-the-art in-pixel text rendering in both English and Chinese, broadening its open-weight portfolio beyond language to multimodal generation. The rapid succession of releases underscores intensifying competition among Chinese technology giants to court developers with freely available model weights, a strategy that contrasts with U.S. rival OpenAI’s decision to delay open-sourcing its latest systems over security concerns.

Alibaba has released Qwen-Image-Edit, a 20 billion parameter open-source AI model for image editing that ranks as the top open weights model in its category. Released under the Apache 2.0 license, Qwen-Image-Edit offers capabilities such as rotating, restyling, and adding text to images without compromising image integrity. The model's performance is comparable to leading proprietary models like OpenAI's GPT-4o and FLUX.1 Kontext [max]. This release is part of Alibaba's broader Qwen series evolution, which has rapidly advanced since 2023, encompassing multimodal and reasoning-focused iterations. The Qwen-Image-Edit model supports natural language prompts for image editing and has been integrated into updated image generation and editing applications. Meanwhile, other open-source AI models, including Microsoft's VibeVoice-1.5B text-to-speech model and Alibaba's Qwen3-Coder-480B-A35B-Instruct for coding and speech applications, continue to expand the open-source AI ecosystem. On the AI leaderboard, non-Chinese open-source models such as GPT-OSS 120B and Mistral Small 3.2 hold the 16th and 17th positions respectively, highlighting Alibaba's leading position in open-source image editing AI.

The All-China Federation of Industry and Commerce released its 2025 Top 500 Private Enterprises List, highlighting that e-commerce giant JD topped the list as the private Chinese firm with the highest revenue last year. Alibaba and oil refiner Hengli Group followed as the next largest private companies by revenue. Collectively, these top 500 private enterprises generated a combined revenue of $6 trillion and reported profits of 1.8 trillion yuan (approximately $252 billion) in 2024. Their investment in research and development reached 1.13 trillion yuan (around $158 billion). This ranking underscores the robust growth of China's private sector amid the country's evolving economic landscape. Additionally, Forbes unveiled several 2025 lists, including the Asia 100 to Watch, Best Employers by State, Global 2000, Best In-State Lawyers, America's Richest Immigrants, Most Valuable Sports Agencies, Canada's Best Employers for Company Culture, Top Creators, Highest-Paid Golfers, and Richest Female Sports Team Owners. The Forbes Asia 100 to Watch list also featured two Pakistani startups, while Foxtale Skin, a skincare and body-care brand with a loyal customer base exceeding 1.5 million users, was recognized in the same list.

Alibaba Group announced plans to spin off Banma Network Technology, its autonomous driving software subsidiary, through an independent listing on the main board of the Hong Kong Stock Exchange. The move aims to support Alibaba's smart car initiatives. Following the proposed initial public offering, Alibaba intends to retain over 30% ownership in Banma, down from its current stake of approximately 44.72%. Details of the spin-off are still being finalized. Banma, based in Shanghai, specializes in AI intelligent cockpit technology for automobiles, aligning with Alibaba's broader push into new energy vehicles and automotive technology.

Alibaba Group Holding Ltd. said it will close its last members-only Freshippo supermarket, bringing an end to the ‘Hema X’ warehouse-style format it introduced in 2020 to rival Costco and Walmart’s Sam’s Club in China. The e-commerce giant will wind down the store gradually and instead open five discount outlets as it pivots toward lower-priced formats amid fierce competition and slowing consumer spending. Alibaba did not disclose a timetable or financial impact, but the move underscores its broader retreat from capital-intensive bricks-and-mortar experiments. Foreign warehouse clubs such as Costco and Sam’s Club have continued to expand in China, while domestic rivals including JD.com are pushing into value grocery offerings, intensifying pressure on Freshippo. Alibaba’s downsizing follows its decision last year to suspend the initial public offering of Freshippo, once seen as a key growth engine for its local services division.

Alibaba has released Wan 2.2, an open-source text- and image-to-video model that packs 27 billion parameters—14 billion of them active—and is distributed under the permissive Apache 2.0 licence. Early testers say the system delivers video quality on par with closed-source rivals such as Veo 2 and Kling 2.0, marking one of the first high-performance video generators whose weights are freely available for developers to fine-tune. AI-hosting platform Replicate became the first commercial provider to deploy Wan 2.2, charging US$0.05 per 480p clip and US$0.10 for 720p output. Stock-image marketplace Freepik simultaneously integrated the model into its AI Suite, highlighting its frame-level control over lighting, colour and camera movement. Replicate also rolled out a speed-optimised version of MiniMax’s Hailuo 02 that can render six- or ten-second videos at 512p for about US$0.10 each—pricing that undercuts many proprietary services. Independent benchmarks suggest generation times have fallen to roughly 55 seconds, with further reductions expected. The twin launches signal accelerating competition in generative video, as open-source licences and rapidly falling inference costs broaden access for filmmakers, advertisers and software developers while increasing pressure on closed platforms to match price and performance.