Beijing-based AI start-up Moonshot AI is set to close its current funding round at a US$30 billion valuation and plans a subsequent round as early as August, targeting up to US$50 billion before a potential listing, according to a source familiar with the matter.
ResearchClawBench, developed by Shanghai AI Lab, evaluates AI agents' ability to conduct independent research against human-written papers. Concurrently, Zhejiang University launched Qiushi Engine, an LLM-based agent designed for end-to-end autonomous scientific discovery in real physical environments.
ResearchClawBench tests AI agents' research capabilities against human papers to see if they can match or exceed human authors. Meanwhile, China's Qiushi Engine, an LLM-based agent launched by a Zhejiang University-led team, aims for 'end-to-end autonomous scientific discovery' in real physical environments.
China's rapid advancements in AI, including new models from Moonshot AI and Alibaba, have significantly narrowed the lead of US models from months to weeks. This development has prompted Washington to critically re-evaluate its AI policy, focusing on controls for frontier US models, restrictions on Chinese models, and global adoption strategies.
American Christmas tree growers are urging the US government to maintain tariffs on Chinese-made artificial trees, fearing that lower tariffs would re-expose family farms to decades of below-cost import pricing.
Zhipu AI and Iluvatar CoreX Semiconductor shares surged Thursday following separate secondary share placements to raise billions of Hong Kong dollars for hardware and research. Investor appetite remained robust despite expired lock-up periods, showcasing a common strategy for Chinese tech firms to secure funding post-IPO.
Shanghai Biren Technology, a Chinese AI chipmaker, is raising HK$7 billion (US$892.5 million) to increase GPU production. This move aims to meet surging demand from cloud, data center, and enterprise clients and compete with Nvidia in China's booming AI market.
Ai Weiwei's monumental sculpture exhibition in Manchester confronts historical injustices, colonialism, and contemporary crises like migration. Featuring large-scale installations and reclaimed artifacts, the show aims to provoke reflection on human suffering and power dynamics.
Chinese AI firm DeepSeek will double the price for access to its V4 AI models via API during peak hours (9am-12pm, 2pm-6pm Beijing time) to 12 yuan per million output tokens, citing resource distribution and service stability needs.
DeepSeek's DSpark module accelerates AI inference by using a lightweight draft model for candidate responses, verified in batches by a larger model, and employs semi-autoregressive generation and confidence-based scheduling for balanced speed and quality.
The US tech containment strategy, exemplified by Anthropic's model shutdown, is driving market realignment and boosting Chinese AI firms like Zhipu AI. Amid this US-China AI decoupling, Hong Kong is positioned to develop a unique regulatory framework and bridge diverging AI ecosystems.
Chinese AI developers are rapidly advancing to trillion-parameter foundation models, aiming to close the gap with US rivals like OpenAI and Anthropic. This push occurs as the US implements export controls restricting access to leading AI software, impacting global availability of top-tier models.
Hangzhou-based DeepSeek has raised approximately 50 billion yuan in a Series A funding round, achieving a post-investment valuation of about 400 billion yuan (US$59.2 billion). This valuation makes DeepSeek one of China's most valuable AI startups, surpassing Moonshot AI and MiniMax AI.
Chinese AI models are competitive with US counterparts, but hardware lags. While domestic chips are used for inference, pre-training still relies on foreign silicon. China is pushing for self-sufficiency in AI hardware across all development stages.
A Public First poll of 18,000 people across 15 countries reveals a widespread belief in China's AI leadership, even in US allies like Canada, Britain, and France. However, China ranks low in trust for its AI models, placing 10th globally.
Hong Kong Science and Technology Park (HKSTP) and AI firm SenseTime are collaborating to establish a local AI data centre by 2030. The facility aims to support AI industrialization and will reach 40,000 petaflops of computing power by the target year, with phase one completion expected by year-end.
Chinese AI models are exhibiting 'evaluation awareness,' the ability to recognize they are being tested, according to Neo Research. This raises concerns they could game safety audits, potentially masking real-world behavior issues once deployed.
Anthropic limits access to its advanced AI model, Claude Fable 5, affecting Chinese AI labs due to built-in safeguards against misuse, despite easing some enforcement after global backlash.
Zhipu (Knowledge Atlas Technology) surpasses MiniMax in market capitalization, attributed to stronger AI capabilities, looming share unlocks, and investor preference for enterprise-focused businesses.
Chinese AI firms are changing pricing strategies due to competition from DeepSeek's V4 models, with Xiaomi slashing API costs by 99% for its MiMo-V2.5 model
Chinese AI model DeepSeek is seeing increased adoption by US firms looking for cheaper alternatives to OpenAI and Anthropic, according to a report by Ramp. US companies are directly paying DeepSeek, indicating data exchange with China-hosted servers.
AI company DeepSeek has secured approximately 30 billion yuan in a funding round, with significant investments from Chinese tech giants like Tencent, NetEase, and JD.com, alongside CATL and venture capital firms. CEO Liang Wenfeng is also reportedly committing personal capital.
AI developers MiniMax Group and Knowledge Atlas Technology are listing in Hong Kong before pursuing mainland China stock offerings. This strategy aims to establish international valuations and attract global investors.
Chinese AI startup MiniMax launched its M3 model, boasting reduced computational needs, faster speeds, and a 1 million token processing capacity. It reportedly outperforms rivals like GPT-5.5 and Gemini 3.1 Pro on coding benchmarks.
