Flash News

TREE NEWS reports: JPMorgan CEO Jamie Dimon said the AI spending boom shows little sign of slowing, with hyperscaler ecosystem investment potentially reaching $1 trillion next year. Spending has more than doubled from about $300 billion last year to roughly $700 billion this year, he said. Dimon added this is adding about 1 percentage point to annual GDP growth and “could slightly push up inflation.”

TREE NEWS reports: TypeSafe AI’s new JEV model, a low-latency AI that returns only buy, sell or hold calls in 70 to 500 milliseconds, has triggered a wave of crypto projects since going viral over the past week. Developers have built open-source trading bots on Monad and Hyperliquid, sentiment and token-safety tools, and prediction-market pricers, while JEV-themed tokens and vaults have appeared on Robinhood Chain, BSC and Solana. None are official TypeSafe releases.

TREE NEWS reports: Unreleased Gemini 4.0 test results circulating on Arena and social media have revived talk of a Google AI comeback, though Google has not officially launched the model. Alphabet’s Q2 Google Search & Other revenue rose 17% year on year, Google Cloud revenue jumped 82% to $24.8 billion, and the Gemini App passed 1 billion monthly active users in August.

TREE NEWS update: OpenAI said on September 21 that an internal model, which began training on August 28, has solved more than 100 long-standing open problems across multiple fields of mathematics, in addition to the previously disclosed Navier–Stokes Millennium Prize problem. The company is bringing in an independent advisory group to address academic scrutiny of the results.

TREE NEWS update: Alibaba Cloud said at the 2026 Yunqi Conference that it will expand its cloud-based super node computing service, adding Lingjun Zhenwu M890 super node service in the Ningxia Zhongwei region in the fourth quarter of this year, following the first batch that went live in Ulanqab. The company did not disclose pricing, capacity or investment figures for the new deployment.

TREE NEWS update: Qianwen Office announced at the 2026 Apsara Conference that it will integrate Salesforce on Alibaba Cloud. The tie-up combines Qianwen Office’s enterprise context understanding and task planning and execution with Salesforce’s customer management capabilities, letting employees query sales opportunities, maintain customer information and advance business processes from within Qianwen Office.

TREE NEWS update: Xingyun Technology announced it has signed three computing power service contracts to purchase 43 sets of GPU computing power services, with a total tax-inclusive value of RMB 453 million. The contracts, each running for five years, follow a broader framework procurement agreement with a state-owned supplier for 128 sets of high-end computing servers and supporting services.

TREE NEWS update: Volcano Engine has launched a Draft mode for its Seedance 2.5 API, letting enterprises and creators generate 480P sample clips first to validate ideas at low cost before rendering selected shots at 1080P. The company said final output closely matches the draft while retaining native 1080P quality and maximizing cost savings.

TREE NEWS update: Chen Long, former head of Xiaomi Auto’s smart-driving foundation model and XLA large model, has left to co-found a startup in Silicon Valley developing embodied large models with a former Google DeepMind scientist. The project has already secured funding and begun commercial partnerships.

TREE NEWS reports: UBS strategists said the popularity of Meta’s new consumer-facing AI agent Muse points to rising demand across the AI supply chain. They said Muse’s successful launch shows AI firms have revenue paths beyond selling technology to large enterprises. Growing competition in AI products will further lift demand for computing power and AI infrastructure, benefiting leading semiconductor and hardware companies.

TREE NEWS update: At Intel’s Technology Innovation and Industry Ecosystem Conference on Sept 22, CEO Lip-Bu Tan said x86 remains the foundation of modern computing and will keep driving computing toward intelligent applications. He cited Intel’s advantages in CPU, GPU, ASIC, advanced packaging and wafer manufacturing, plus an open ecosystem spanning cloud, edge and device, saying Intel is confident of capturing the opportunities ahead with customers and partners.

TREE NEWS update: NetEase Youdao’s open-source Confucius4-R2T2 streaming speech recognition model and Confucius4-T3PO streaming translation model each topped the Hugging Face Trending lists in their respective categories, the company said. The two models address low-latency, stable real-time speech recognition and fast, accurate streaming translation, and can be combined into a base technology stack for real-time AI voice agents and simultaneous interpretation.

TREE NEWS update: Qwen announced at the 2026 Yunqi Conference on September 22 that it will accelerate development of a Personal Agent, building on the Agentic capabilities of its Qwen 3.8 series models. Product lead Zheng Sishou said the goal of giving every person their own AI assistant has been unchanged since the service launched last year. The company said the new Personal Agent format will serve 300 million users with continuous, personalized intelligence.

TREE NEWS update: Aoni Electronics said its wholly owned subsidiary Aoni Intelligent signed a purchase contract with Company A to buy GPU compute card products for a total of 1.67 billion yuan, tax included. The company said the contract is a major routine operating contract that requires no board or shareholder approval and does not constitute a related-party transaction or major asset restructuring. Fulfillment is expected to positively affect results in the performance year.

TREE NEWS update: Amap’s mapping unit Amap Cloudmap unveiled its Spatial Intelligence AI Platform during the 2026 Yunqi Conference, the company said. The platform has three parts: Yunjing, which reconstructs and generates the real world as a digital base for AI agents; Yunrui, offering AI agents across seven industries including commerce, transport, logistics, finance, tourism, industry and new energy, opened externally via an MCP/Skill/Agent three-tier architecture; and Yunce, which packages the platform’s capabilities into deliverable, replicable industry solutions.

TREE NEWS update: Qwen released Qwen Intelligence, a full-stack AI phone solution it says is built for co-creation with smartphone makers so handsets can complete complex tasks. The offering comes from the Qwen large-model team and is positioned as an end-to-end package for device manufacturers rather than a standalone app.

TREE NEWS reports: Zhipu AI’s GLM-5.3-Flash recorded 4.36 trillion tokens in daily API calls, up 73% from the previous counting period, according to the latest daily ranking from API aggregation platform OpenRouter. The model ranked first among all models on the platform by call volume.

TREE NEWS update: Biwin Storage said on its investor interaction platform that its OEC (optoelectronic co-packaging) products are being developed with customers for 6.4T NPO optical engine packaging, with R&D targeted for completion in the fourth quarter of 2026. The company said its advanced packaging ecosystem spans the three core areas of AI hardware infrastructure: storage, computing and transmission, with OEC aimed at ultra-high-speed interconnect demand.

TREE NEWS update: Chinese Foreign Ministry spokesperson Guo Jiakun confirmed that the US and China discussed establishing an artificial intelligence dialogue and possible notification mechanism, following talks between Vice Premier He Lifeng and US Treasury Secretary Scott Bessent. Asked whether specific AI measures would be announced during this week’s leaders’ summit or whether Beijing believes Washington aims to curb China’s AI development, Guo said the specific question should be directed to the competent Chinese authorities.

TREE NEWS update: Qwen Office released a new enterprise-facing product called Enterprise Context, aimed at helping corporate AI agents understand business operations, collaborate across teams and earn trust. The product treats data as the key prerequisite for agents entering enterprise workflows, compressing and structuring years of accumulated corporate data so employees see only what is relevant to the task at hand.