Juro Osawa

DeepSeek has open-sourced DeepInfer, its proprietary inference framework, making public the technical infrastructure behind its cost-efficient AI model deployment. The move exposes the methods DeepSeek uses to achieve high-margin inference operations at scale. The release signals a shift toward transparency in how the Chinese AI developer optimizes computational efficiency.

Apple plans to launch Apple Intelligence, its on-device AI service, in mainland China in December after regulatory delays. The rollout to one of Apple's largest markets marks a significant milestone for the company's AI strategy. The release has been held up by Chinese regulatory requirements, but approval now appears imminent.

Manus, an AI agent startup that recently separated from Meta Platforms, has raised more than $500 million in new funding led by Boyu Capital and other new investors. The funding round was announced by Butterfly Effect, Manus's parent company, via a WeChat post. The capital raise follows the startup's unwinding of its relationship with Meta, signaling continued investor confidence in the AI agent space despite the separation.

DeepSeek announced Wednesday that it will open-source software tools customized for Huawei's AI chips, marking an escalation in the Chinese AI firm's push to rely on domestic semiconductors for model training and deployment. The move includes a Huawei-compatible version of TileLang, a software framework. This effort reflects broader Chinese industry strategy to reduce dependence on foreign chip technology amid U.S. export restrictions.

Samsung Electronics and five affiliates are investing $1 billion in Helix Digital Infrastructure, an AI infrastructure venture launched by private equity firm KKR in June. The investment reflects intensifying competition among tech companies to secure AI computing capacity as demand surges. Samsung joins a growing list of corporations betting on dedicated infrastructure plays to support AI workloads.

Manus, an AI agent startup that recently separated from Meta, launched Cue, a standalone app providing users with multiple personal AI agents. Each agent has its own email and phone number, positioning the product as a competitor to Meta's Muse. The launch marks Manus's first major product release since its independence from Meta.

Chinese leader Xi Jinping called for U.S.-China cooperation on AI safety during a White House summit with President Trump on Thursday. Xi specifically proposed that the two nations work together to prevent the misuse and abuse of artificial intelligence. AI safety and governance were identified as major topics in the bilateral talks, signaling potential alignment on a contentious global issue.

Alibaba unveiled the Zhenwu V900, a new AI chip for model training and inference, at its annual Apsara conference on Tuesday. The chip delivers three times the performance of its predecessor, demonstrating progress in China's domestic semiconductor capabilities. The announcement was accompanied by a data center expansion plan, though specific details on scale and investment were not fully disclosed.

Naive AI, a Beijing-based startup founded by a Tsinghua University professor in February, has reached a $1.4 billion valuation after raising $400 million across three funding rounds from investors including Tencent. The company plans to release its first large language model this month as an open-weight model, positioning itself as a new entrant in China's competitive LLM market alongside DeepSeek, Moonshot, and Alibaba.

Chinese AI startups focused on data evaluation and model benchmarking are attracting venture capital attention as a critical layer in the country's AI development. Silicon Valley investors visiting China identified a growing ecosystem of local firms comparable to U.S. counterparts like Surge, Mercor, and Scale. These companies provide high-end data access and sophisticated evaluation tools that help developers refine cutting-edge AI models for complex, expert-level tasks. The trend reflects how access to quality training data and rigorous benchmarking has become essential infrastructure for advancing AI capabilities.

Z.ai, a Hong Kong-listed AI company, is raising approximately $5 billion through equity and convertible bond offerings to fund the escalating costs of developing frontier AI models. The company plans to raise $2 billion through a share offering, according to a regulatory filing released Sunday. The fundraising reflects the capital-intensive nature of competing in advanced AI development.

Humain, Saudi Arabia's state-owned AI company, announced the humain-m3 model, an Arabic language model built on Chinese firm MiniMax's open-source M3 foundation. The model was pre-trained on more than 1 trillion tokens of Arabic content. The development represents a collaboration between Saudi and Chinese AI capabilities focused on Arabic language processing.

Moonshot AI, the Chinese startup behind the popular Kimi K3 open-source model, has filed confidentially for a Hong Kong initial public offering, according to Chinese media outlet LatePost. The filing marks a significant milestone for the company, which has gained global traction with its model. The company previously underwent restructuring to unwind certain operations before pursuing the public markets.

Chinese AI developer MiniMax reported its annualized revenue run rate surged to $800 million in August, up from $150 million in February, driven by enterprise demand for its AI video model and large language models. The Shanghai-based company's growth reflects a broader shift toward commercializing AI capabilities for business customers. The expansion demonstrates significant market traction in China's competitive AI sector.

DeepSeek generated $70.7 million in revenue during the first seven months of 2026, a tenfold increase from its full-year 2025 revenue. The Chinese AI lab posted a net loss of 715 million yuan for the same period, down from 935 million yuan for all of 2025. The company is currently fundraising for a second round targeting 50 billion yuan at a 500 billion yuan valuation.

Alibaba CEO Eddie Wu announced on an earnings call that the company expects AI-related annualized revenue run rate to reach $10 billion in the quarter ending September, up from $7.3 billion in the prior quarter. This represents a 37% quarter-over-quarter increase in AI revenue. The projection reflects Alibaba's growing focus on monetizing AI products and services across its cloud and enterprise offerings.

Chinese AI developer Z.ai released GLM-5.3, an open-source model it claims matches Anthropic's Mythos 5 in cybersecurity capabilities. The Beijing-based company, also known as Zhipu, positioned the model as a significant improvement over its predecessor GLM-5.2. The release marks another step in China's competitive push in generative AI development.

Tencent nearly tripled its capital expenditures to 52.8 billion yuan ($7.8 billion) in Q2 2026, driven by investments in computing infrastructure for AI model training and its coding and productivity tools. The spending surge reflects the Chinese tech giant's commitment to scaling AI capabilities amid intensifying competition in the sector. The capex increase signals Tencent's pivot toward infrastructure-heavy AI development.

Junyang Lin, a former Alibaba researcher who departed in March, has launched Pragmatik Labs, a Shanghai-based startup focused on next-generation AI agents for both digital and physical applications. The announcement confirms earlier reporting and marks Lin's entry into the competitive AI agent space. The startup's focus on agents spanning digital and physical domains positions it within a growing segment of AI development.

Sony Group and Taiwan Semiconductor Manufacturing Co. plan to invest approximately 1 trillion yen ($6.3 billion) in a joint venture to mass produce next-generation image sensors in Japan starting as early as 2029. The partnership represents a significant capital commitment to semiconductor manufacturing in Japan and signals both companies' focus on advanced imaging technology production.

Chinese AI developer DeepSeek has resumed its second funding round after a week-long pause triggered by a leaked transcript of a confidential call between CEO Liang Wenfeng and investors. The company also plans to increase prices for its AI models. The funding resumption signals investor confidence despite the operational disruption from the leak.

Chinese AI firm MiniMax announced H3, a new open-source video generation model, on Friday. The move marks MiniMax's first time open-sourcing its flagship video model, intensifying competition in the AI video space against rivals like ByteDance and Google. Open-sourcing the model represents a strategic shift to gain market share in a crowded sector.

Sierra, the AI customer support startup led by Bret Taylor, is acquiring Takeoff, a three-person startup founded by Aakash Thumaty that builds agents capable of handling extended tasks. Takeoff was generating nearly $10 million in annualized revenue one year after its founding and had raised $15 million from Matrix and venture capitalist Lachy Groom. The acquisition allows Sierra to expand beyond customer support into broader agent applications, leveraging Thumaty's expertise in building long-running autonomous systems.

Ant International, the overseas arm of Chinese fintech giant Ant Group, raised approximately $1.2 billion in a funding round to expand its global payment network. The round was backed by Ant Group, affiliate Alibaba Group, and other global institutional investors. The funding underscores Ant Group's continued push to grow its international presence despite regulatory scrutiny in China.

Moonshot AI, a Beijing-based startup whose open-source model has gained attention in Silicon Valley, is seeking formal investor approval to begin IPO preparations in Hong Kong. The move signals the company's confidence in its market position and represents a significant step toward public markets for a Chinese AI firm. The timing reflects broader momentum in the AI sector and Moonshot's emergence as a competitive player in generative AI development.

Chinese AI developer MiniMax is developing a 2.7-trillion parameter language model, which would be larger than any currently available Chinese AI model. The model could launch as early as Q3 2026, according to sources with knowledge of the plan. This represents a significant scaling effort by a Chinese AI company in a competitive market dominated by larger players.

Kuaishou Technology announced that its Kling AI video unit has secured nearly $3 billion in funding at a $15 billion pre-money valuation. The Chinese social media company is bringing in outside investors to support the unit's expansion. After the fundraising closes, Kuaishou's ownership stake in Kling will be diluted, though the article does not specify the final ownership percentage.

ByteDance unveiled Seedance 2.5, an upgraded AI video generation model, at a Beijing conference on Tuesday. The new model improves upon Seedance 2.0, which was previously recognized as a significant breakthrough in AI video generation.

Early Chinese investors in AI firm Manus are planning to repurchase the company from Meta Platforms at the $2 billion acquisition price Meta paid in December, following a Chinese government order to reverse the deal. Since Meta's acquisition, Manus has seen significant revenue growth. The buyback represents a reversal of Meta's entry into the Chinese AI market and reflects government intervention in cross-border tech acquisitions.

Microsoft is exploring the integration of DeepSeek's V4 model as a cost-effective option for its Copilot Cowork AI assistant, according to reporting from Axios. The company is evaluating either a Microsoft-hosted version of DeepSeek V4 or another open-source alternative to reduce expenses associated with powering the assistant. This move reflects Microsoft's effort to balance capability with cost efficiency in its AI product offerings.

Chinese AI developer MiniMax launched M3, a new large language model designed for coding and complex multi-step tasks. The company claims M3's coding capabilities match those of Anthropic's Opus 4.7, released in April. The launch reflects intensifying competition in open-source AI coding models among Chinese and Western developers.

Kuaishou Technology's Kling AI video generation unit reached approximately $500 million in annualized revenue as of March 2026, based on first-quarter performance. The Chinese social media company disclosed that Kling generated over $96 million in revenue during the first quarter alone. This milestone reflects growing commercial traction in the AI video generation market, where demand for synthetic video tools continues to expand.

Qualcomm has secured a deal to supply AI data center chips to ByteDance, according to Bloomberg. The agreement marks Qualcomm's push to expand beyond smartphone processors into the competitive AI computing chip market. The deal reflects growing demand from major tech companies for specialized AI infrastructure.

CATL, China's leading EV battery manufacturer, plans to invest in DeepSeek's first funding round, which targets 50 billion yuan ($7.35 billion) and could close next month. The investment reflects CATL's strategic pivot toward supplying power equipment to AI data centers. DeepSeek's post-money valuation could exceed 350 billion yuan ($51.4 billion) following the round.

Alibaba's semiconductor design unit unveiled a new AI chip designed for training and running AI models, marking a significant step in China's broader push to reduce dependence on Nvidia through domestic chip development. The move reflects government pressure to accelerate adoption of homegrown AI semiconductors and build indigenous capability in a critical technology area. The chip represents Alibaba's contribution to China's strategy of developing competitive alternatives to foreign AI hardware.

Lenovo's internal memo to staff reveals that no new export licenses or policies for Nvidia's H200 chips have been issued since March, when the U.S. approved sales to select Chinese companies. The memo signals uncertainty around the pace and scope of H200 adoption in China despite the regulatory green light. This suggests potential friction between U.S. policy approval and actual market execution, with implications for both Nvidia's China revenue and Chinese server makers' AI infrastructure plans.

Anthropic published a research paper Friday calling for stricter U.S. chip export restrictions to China, timed to coincide with President Trump's state visit to Beijing. The paper, titled '2028: Two Scenarios for Global AI Leadership,' argues that limiting China's access to advanced semiconductors is critical for maintaining U.S. AI competitiveness. The move puts Anthropic at the center of ongoing geopolitical tensions over AI development and chip supply chains.

The U.S. Commerce Department has approved Nvidia's H200 chip sales to approximately 10 Chinese companies, including major tech firms Alibaba, Tencent, ByteDance, and JD.com. The clearance signals a shift in Washington's approach to AI chip export restrictions, which have been a central point of U.S.-China tech competition. The decision affects some of China's largest technology players and suggests the Biden administration is recalibrating its stance on semiconductor access for Chinese enterprises.

Junyang Lin, the former lead researcher behind Alibaba's Qwen language models, has launched a new AI lab and is raising several hundred million dollars in funding. The lab is expected to reach a $2 billion valuation after this funding round. Lin's departure from Alibaba and immediate move to start a competing venture signals both the competitive talent dynamics in large language model development and the capital appetite for new AI research initiatives outside established tech giants.

Kuaishou Technology confirmed via regulatory filing that its board is evaluating a restructuring of Kling, its AI video generation unit, that could involve external funding. The move follows earlier reporting and signals the Chinese social media company's intent to potentially spin out or partially divest the video AI business. The filing does not specify funding amounts, timelines, or the structure of any capital raise.

Kuaishou Technology, a Chinese social media platform, is planning to spin off its Kling AI video generation unit ahead of an IPO next year. The company is in talks with investors for a pre-IPO funding round that would value Kling at $20 billion. Kling competes directly with AI video models from ByteDance, Google, and Alibaba, positioning the spinoff as a way for Kuaishou to capitalize on investor appetite for standalone AI businesses.

Chinese tech startups including Moonshot AI and DeepRoute.ai are reconsidering their corporate structures, moving from overseas incorporation to domestic Chinese registration. The shift follows signals from China's securities regulator that it will be less likely to approve IPOs for companies incorporated outside the country. This regulatory stance emerges in the wake of China's forced unwinding of Meta's acquisition of AI agent startup Manus, signaling tighter government control over foreign investment in Chinese AI firms.

Meta is preparing to unwind its $2 billion acquisition of Manus, an AI agent application company, after Chinese regulators ordered the parties to revoke the deal. The reversal could prove operationally complex given the integration work likely underway since the acquisition closed. This marks a significant setback for Meta's AI agent strategy and reflects growing regulatory scrutiny of cross-border tech acquisitions involving AI capabilities.

Tencent's latest AI model, Hy3, has received positive developer feedback, but the company appears to have leveraged Anthropic's Claude to help evaluate and fine-tune the system, according to internal memos and sources with direct knowledge. This is notable because Anthropic officially restricts access to its models and services in countries deemed U.S. adversaries, including China. The disclosure raises questions about how AI capabilities are being shared across geopolitical boundaries and the practical limits of export controls in the AI sector.

Tencent launched an international version of QClaw, an AI agent built on its OpenClaw framework, on Tuesday. The move marks the company's expansion beyond China after releasing multiple OpenClaw-based products domestically over recent months. QClaw had generated significant interest in its initial rollout, prompting Tencent to pursue global distribution. The launch signals Tencent's intent to compete in the global AI agent market alongside other major tech companies.

Anthropic has begun requiring government-issued photo ID and facial verification from some customers, escalating its enforcement of restrictions on access from countries deemed U.S. adversaries, including China, Russia, and North Korea. The policy shift follows a year of incremental steps to block users in these regions. At least one Beijing-based AI startup founder lost access to Claude Code following the announcement, prompting him to switch to OpenAI's Codex as an alternative, though OpenAI maintains similar geographic restrictions without requiring ID verification.

Moonshot AI, a Beijing-based startup, released its Kimi K2.6 model with claimed advances in coding capabilities, timing the launch ahead of DeepSeek's anticipated V4 release, which also emphasizes coding performance. The move reflects intensifying competition among Chinese AI labs to establish dominance in code generation and developer-focused applications. Both releases signal a strategic focus on coding as a key differentiator in the broader AI model race.

Victory Giant Technology, a Chinese manufacturer of printed circuit boards for AI servers and a key Nvidia supplier, surged 60% on its Hong Kong Stock Exchange debut. The jump reflects investor appetite for companies positioned in the AI infrastructure supply chain as demand for AI server hardware accelerates globally. The company's strong market reception signals confidence in sustained demand for the components that underpin large-scale AI deployments.

ByteDance's cloud unit BytePlus has launched Seedance 2.0, its AI video generation model, to enterprise customers across more than 100 countries and regions globally. The rollout notably excludes the United States, and it remains unclear whether BytePlus intends to offer the model to U.S. customers. The move reflects ByteDance's strategy to expand its AI capabilities internationally while navigating geopolitical constraints that have limited its operations in the American market.