Daily Highlights
- Meituan unveils 1.6T-parameter multimodal model LongCat-2.5
- Oracle files for another $20B AI data center funding round
- OpenAI merges ChatGPT chat and Work tabs into one entry
- OpenAI seeks $30B at over $1.4T valuation ahead of Anthropic IPO
- Apple confirms Qwen integration for Apple Intelligence in China
Today's Headlines
[Meituan launches LongCat-2.5, a 1.6T-parameter multimodal model built for long-horizon coding]On November 2, Meituan released its multimodal foundation model LongCat-2.5, with 1.6 trillion total parameters and roughly 48 billion activated per token, targeting long-horizon coding and complex reasoning. Building on its MoE architecture, the model improves long-context performance while keeping inference costs in check - the largest parameter jump in the LongCat open-source series and a sign that platform-level foundation model competition is escalating. (Source: AIFOD)
[Oracle files for another $20B raise as it doubles down on AI data centers]On November 2, Oracle filed with the U.S. SEC to sell $10B in common stock and $10B in senior notes - $20B total for AI data center construction, its second raise of that size this year. The stock fell over 5% on the filing day and is down about 33% year-to-date; FY2026 data center capex hit $55.7B, about 11% above the company's own guidance, and leverage pressure on AI infrastructure is starting to show. (Source: Winzheng/SEC filing)
[OpenAI eyes $30B raise at over $1.4T valuation, racing Anthropic's IPO]According to Reuters, OpenAI is racing to lock in investor commitments at a valuation above $1.4 trillion, seeking around $30 billion with backers including SoftBank and MGX in talks to participate. The move is widely read as an effort to secure funds before Anthropic prices its mid-November IPO, as the two AI giants compete for the same investor pool; OpenAI had previously trimmed its target amid an AI stock selloff. (Source: Reuters)
New Products & Tech
[OpenAI merges ChatGPT chat and Work tabs, unifying usage pools]On November 2, OpenAI announced it is merging ChatGPT's chat entry with the Work (workspace) tab, unifying usage pools and the product entry point. The change removes the split between personal conversations and team workflows, paving the way for agentic workflow distribution and pulling enterprise features deeper into the consumer entry. (Source: Winzheng)
[Researcher debuts QwenPU: a nested LLM running inside Qwen's weights]Researcher divagr1925 released QwenPU on Hugging Face, demonstrating a functional language model embedded entirely within the weight space of a Qwen model - a model inside a model. The experimental project explores weight-level model reuse and has sparked community discussion on parameter compression and nested architectures. (Source: Hugging Face)
IPO & Fundings
[Yitai Intelligence raises over 200M yuan in Series A for high-speed interconnect chips]On November 2, Chinese high-speed interconnect chip maker Yitai Intelligence announced a Series A of over 200 million yuan, with the A3 tranche led by Wuxi Venture Capital and joined by Jintou Zhiyuan and Hongshan Capital. Funds go to mass production of edge switching chips and talent; its automotive-grade chip ETLX610 is in volume production with over 100,000 vehicles on the road, making it China's top automotive Ethernet switching chip supplier by shipments. (Source: Yiou)
[Omen AI acquires Aquatic Labs for $16M, doubling down on data center liquid cooling]On November 3, Omen AI announced the $16 million acquisition of Aquatic Labs to strengthen its data center liquid cooling technology. As AI compute density climbs, cooling has become a core operating cost driver, and M&A in the liquid cooling space continues to heat up. (Source: Winzheng)
Open Source & Research
[QwenPU and LongCat-2.5: weight-nesting experiments and a trillion-scale sparse reference]Researcher divagr1925 open-sourced QwenPU on Hugging Face, demonstrating a nested language model running inside Qwen's weight space - a fresh take on model reuse and parameter compression. In parallel, Meituan's LongCat-2.5 continues the series' open-source approach with a 1.6T-parameter, 48B-activated MoE design, giving the community a large-scale engineering reference for sparse models. (Source: Hugging Face)
Company & People
[OpenAI lobbies Congress for "AI speed limits", vying to define frontier model rules]Reported on November 2, OpenAI is lobbying the U.S. Congress to introduce a "speed limit"-style regulatory framework for the most capable AI models, seeking to define frontier capability boundaries at the federal level. The move is seen as an assertive play on safety narrative and regulatory influence, consistent with earlier safety-coordination talks among OpenAI, Anthropic and Google DeepMind. (Source: Winzheng)
Policies & Data
[BofA: AI infrastructure debt financing tops $1.37 trillion cumulative]Bank of America's Digital Infrastructure Market Monitor shows cumulative AI-related corporate debt, securitization and project financing has reached $1.374 trillion since ChatGPT launched, with $635.1 billion issued in 2026 alone; total U.S. data center capacity in operation, under construction and planned stands at roughly 103GW, and credit risk on AI infrastructure is being repriced. (Source: Cailianshe/BofA Securities)
Disclaimer: For reference only; exercise caution for investment or decisions.
Powered by AOYII.com
