All Updates
Page 711 of 1669
May 7, 2026
China Q1 Smart Devices Output Surges
China's smart consumer device manufacturing accelerated in Q1, driven by trade-in policies and AI fusion in products. Smartphone production hit 298 million units (+6.9% YoY), service robots over 4.4 million sets (+2.6%). IC output reached 127.2 billion units (+24.3%), bolstering upstream components.
ByteDance Doubao AI subscription faces China skepticism
ByteDance is testing a paid subscription model for its Doubao AI chatbot in China, but users like Shenzhen fund employee Chloe Wang refuse to pay the proposed price. Despite finding Doubao user-friendly among domestic AI tools, Wang views it as not worth the cost for occasional work queries. She would prefer other productivity aids if forced to pay.
Kimi Lacks DeepSeek Despite Ample Funds
Kimi, backed by strong funding, faces a critical shortfall in matching DeepSeek's capabilities. The company is simultaneously fighting two urgent battles in the competitive AI landscape.
TikTok Approved $25B Thailand Data Centers
Thailand BOI approved six projects worth $29B on May 6, led by TikTok's ~$25B investment. TikTok will expand servers, data storage, and processing in Bangkok, Nonthaburi, and Chonburi. It supports surging digital services and elevates Thailand's regional infra role.
Anthropic Rents 220K GPUs, Commits $200B to TPUs Same Day
Anthropic leased 220,000 NVIDIA GPUs on the same day it committed $200 billion to Google TPUs. This dual move underscores the AI industry's shift from parameter arms race to compute efficiency competition.
Tujian Tech Tactile Gloves Mass Produce Soon
Stanford postdoc-founded Tujian Tech raises over 100M RMB in Pre-A rounds. Focuses on stretchable multimodal e-skins for robots. Low-cost tactile gloves for data collection to mass produce in 2026.
Anthropic Secures 220K GPUs from SpaceX
Anthropic partners with SpaceX to access over 220,000 Nvidia GPUs in the Colossus 1 data center starting May, addressing Claude's compute shortages. The deal serves Claude Pro/Max users and includes interest in orbital AI compute. Anthropic also raised usage limits for paid users on the same day.
Korea Tops Canada as 7th Largest Market on AI Chips
South Korea’s equity market has surpassed Canada’s to become the world’s seventh-largest stock market. This milestone is driven by strong demand for chips powering artificial intelligence. The growth underscores the semiconductor sector's expansion amid AI boom.
Musk Sells 220K GPUs to Claude, Doubles Limits
Elon Musk sold 220,000 GPUs to Anthropic for Claude AI. Paid Claude users woke up to doubled 5-hour rate limits and faster speeds. The companies are partnering to build space-based computing infrastructure.
150k Stool Images: Scaling CV Annotation Best Practices
A user shares a 150k+ stool images dataset for training computer vision models, detailing a manual verification process for annotations like Bristol type, consistency, and blood indicators. They train iteratively on verified data and seek advice on scalable workflows. The post asks for ML best practices in high-stakes annotation domains.
Cambricon Tops A-Shares as New Stock King
Cambricon overtook Moutai to become A-share's highest-priced stock at 1699.96 CNY after blockbuster Q1 earnings with 159% revenue growth and first sustained profitability. Driven by AI chip demand from Nvidia bans and DeepSeek V4 ecosystem, it signals market pivot to tech over consumer goods.
HK Medtech IPOs: Post-Listing Stock Review
Analysis of 2024-2025 HK IPOs in medical/health shows mixed results: XtalPi shines at 385B HKD mcap (from 180B issue), iFlytek Medical holds above issue price post-unlock. Most flop post-unlock, dropping to 30-40% of peak, with liquidity tied to HK Connect status.
Montage Tops CATL Valuation on AI Chip Rally
Montage Technology Co. has overtaken CATL as the priciest dual-listed stock in Hong Kong relative to mainland shares. This follows a rally in chip stocks driven by surging AI demand. The shift highlights investor enthusiasm for AI-related semiconductors.
ILP Networks Learn Rules from Pure Images
A Reddit discussion on a new paper about Visual Perceptual to Conceptual First-Order Rule Learning Networks using ILP on image datasets for predicate induction. It notes recent ILP advances claiming strong performance where previously difficult. Questions if ILP can stably compete with deep learning in machine vision.
T-Head Launches Panmai 920 AI Smart NIC
Alibaba's T-Head released Panmai 920, China's first 400G smart NIC with built-in PCIe Switch, addressing AI cluster networking bottlenecks. It boosts GPU utilization in large-scale training and inference via multi-path RDMA and smart congestion control. Already mass-produced for Aliyun data centers.
PixelBloom Funds AI Agents Post C-Round
PixelBloom completed C-round financing led by GuoKe Investment and ShangTang GuoXiang Capital to advance AI office Agents. The new '小方同学' (Fangan.com) delivers full commercial marketing schemes, shifting from tools to digital employees. Plans include multi-Agent matrix and global expansion with 30M users.
AMD 26.5.1 Driver Supports Pragmata, Fixes God of War
AMD released Adrenalin Edition 26.5.1 driver after two months, adding support for five games like Pragmata and Honor of Kings: World. It fixes stuttering in Resident Evil on RX 9000 series and graphics corruption in God of War. Known issues persist with Battlefield 6 on Ryzen AI 9 HX 370.
Anthropic CEO Predicts 80x Growth This Year
Anthropic's CEO Dario Amodei states the company could grow 80 times this year. This rapid expansion has exponentially boosted the startup's need for computing power.
Qwen-Code v0.15.6 Nightly Release with Core Fixes
Qwen-Code released v0.15.6-nightly, introducing FileReadCache for efficient reads, shared permission flows for tools, and background agent resumption. CLI enhancements include model switching, task monitoring, and export improvements, alongside telemetry fixes and DeepSeek compatibility. Multiple performance tweaks and unescaping fixes enhance reliability.
RTX 5090 vs M5 Max for Agentic Coding
Debate on RTX 5090 (3x faster than M5 Max 128GB for Qwen3.6 27B coding) vs M5 Max (4x memory for higher quants/bigger context). 5090 suits speed with MTP/KV offload; M5 Max for massive repos/contexts. User seeks experiences for local agentic dev.