Big tech moves and startups
ByteDance reorganizes Seed AI division into four departments as it plans a 5-trillion-parameter model that would dwarf current industry leaders, signaling aggressive expansion in the AI arms race....
Anthropic preparing to give CEO Dario Amodei and co-founders supervoting shares ahead of IPO. Amodei owns just 2% but will control board elections. Company maintains public benefit corporation structure with Long-Term Benefit Trust....
Stanley Druckenmiller returns to Chinese tech with $10M Baidu bet after 2-year hiatus. David Tepper nearly doubles Baidu stake to $148M while cutting Alibaba 42%. Smart money betting on AI pivot over legacy internet....
Alibaba launches HappyShrimp 1.0 AI music model, generating complete songs from text prompts. End-to-end creation with emotional intelligence. Developed by Token Hub, now live globally on PC web....
Chinese hedge funds like Perseverance slashed Nvidia by 72% in Q2, rotating into Micron and SanDisk. Greenwoods and Oriental Harbor made similar moves. The shift signals maturing AI investment thesis: from chasing hyperscalers to buying supply chain bottlenecks....
Zhipu AI released GLM-5.3, claiming the strongest open-weights coding model, while Alibaba released Qwen 3.8-27B and a 2.4T MoE model under Apache 2.0 — another coordinated Chinese push in open-weight AI....
Tencent revealed a $53B quarterly AI capex splurge, 68-fold Hy3 API usage growth, and a strategic stake in Lovable’s $13.3B Series C, signaling a full-stack AI push....
Japan has labour shortages and the G7’s worst productivity, yet just 8.4% of workers use AI, far behind the US, UK, and Singapore. Risk aversion, legacy systems, and a jobs-first culture are holding back adoption....
OpenAI has released a preview of its ChatGPT desktop app for Linux, bringing ChatGPT, ChatGPT Work, and Codex to Ubuntu, Debian, and Fedora users....
Indian startups like Sarvam AI and Krutrim are building affordable, multilingual AI models that could serve as a blueprint for the Global South — proving sovereignty and accessibility are achievable....
Tencent Hy3 goes global with free WorkBuddy access....
Alibaba launches Qwen 3.8-Max at 2.4T parameters while DeepSeek V4 Flash undercuts OpenAI by 40% on cost per task. Chinese open-weight models are closing the performance gap with American proprietary models while offering dramatically lower prices and unrestricted access....
ByteDance launches Seedance 2.5 — a video generation model that produces 30-second clips with integrated audio in a single pass, outlasting Google Gemini and enabling full production pipelines....
Snap will stop recommending wholly AI-generated Spotlight videos, while LinkedIn adds a dedicated AI-slop reporting button and reveals billions of blocked automated comment attempts....
Chinese platforms ActID and New Claw are paying people $15 to $700 to license their faces for AI-generated microdramas, as the industry tries to replace face theft with a legal marketplace....
DeepSeek CEO Liang Wenfeng revealed a radical strategy prioritizing AGI over profit through open-source models, ten-month break-even pricing, staying out of the spotlight, and focusing relentlessly on continuous learning and agent capabilities while rejecting distractions like video generation....
China’s daily AI token calls reached 140 trillion by March 2026, a 1,000-fold increase driven by AI agent adoption, signaling a massive shift in scale and infrastructure demand....
Alibaba unveils Qwen 3.8, a 2.4-trillion-parameter open-weight multimodal model, challenging Moonshot’s Kimi K3 and intensifying China’s AI competition as Moonshot prepares for a possible IPO....
Netflix confirms AI is used in 300 productions, with tools like Interpositive and Eyeline reshaping post-production as global rivals from China race to own the generative video stack....
Moonshot is reportedly launching Kimi K3, a 2-3 trillion parameter open-weight model expected to outperform Anthropic's Claude Opus 4.8 on benchmarks while costing significantly less....
