ByteDance’s profit plunges 70% on aggressive AI spending
ByteDance Ltd.’s net profit plunged more than 70% in 2025 as the company aggressively scaled up its artificial intelligence investments, even as the booming TikTok e-commerce business drove a nearly 50% surge in overseas revenue. The Chinese technology giant disclosed the operational data for 2025 on Monday, revealing that domestic revenue grew nearly 20% in 2025 year-on-year. Douyin Vice President Li Liang said that the profit drop stemmed mainly from changes in preferred stock and option costs, which did not reflect actual operations. Excluding these non-operational factors, ByteDance’s full-year revenue and profit actually grew, though the operating profit margin dipped slightly in the second half of the year due to slowing domestic e-commerce growth and increasing investment in emerging businesses, Li said.
Chinese GPU maker Sunrise raises over 1 billion yuan
Chinese chipmaker Sunrise said Monday it raised more than 1 billion yuan ($140 million) in a new funding round to support mass production of its artificial intelligence inference graphics processing units. The latest financing brings the startup’s total funding to about 4 billion yuan across seven rounds since it was spun off from SenseTime Group Inc. in late 2024. State-backed investors and venture capital firms including IDG Capital, Gaorong Ventures and CICC Capital have invested nearly 3 billion yuan in the company over the past year. The new funds will be used to scale production of the company’s flagship Qiwang S3 inference GPU, expand its software ecosystem and develop future S4 and S5 chips.
Moonshot AI launches new model with improved coding and agent capabilities
Chinese artificial intelligence startup Moonshot AI has launched an open-source large model that it claims rivals top foreign competitors like OpenAI’s GPT-5.4 in coding and agent-based tasks. Released on Monday, the Kimi K2.6 model is designed to execute extended programming assignments and coordinate a large cluster of AI agents for complex output. Moonshot AI said the K2.6 matched or outperformed closed-source models such as GPT-5.4, Anthropic’s Claude Opus 4.6 and Google’s Gemini 3.1 Pro on industry benchmark tests like SWE-Bench Pro for evaluating models’ software engineering capabilities and DeepSearchQA for assessing AI agents’ deep web research capabilities.
Alibaba launches flagship AI model preview
Alibaba Group Holding Ltd. released an early preview of its next-generation flagship artificial intelligence model, Qwen3.6-Max-Preview, on Monday. According to the third-party evaluation platform Artificial Analysis, the new model outperformed domestic competitors such as GLM5.1 and MiniMax-M2.7. The preview version features enhanced world knowledge and instruction-following capabilities, along with significant performance improvements in agent programming tasks.
