In-Depth Articles

Anthropic elevated Claude's data analysis accuracy from 21% to 95% through a four-layer architecture and continuous maintenance—the secret lies not in model strength.

Step 3.7 Flash model by JumpScale tops the global output speed leaderboard, marking a shift in model competition toward efficiency and cost balance.

Xingcan Intelligence secures support from three major capital players including Lockheed Science & Technology, building a robust barrier in spatial intelligence and data closed-loop, leading the home embodied intelligence sector.

NVIDIA CEO Huang Renxun announces full-scale production of the Rubin architecture, built by 40,000 engineers, alongside the launch of the most powerful Vera CPU in history.

Hackers George Hotz receives AMD chip, aims to break NVIDIA's CUDA moat with minimal software stack tinygrad, challenging its compute dominance.

Yushu achieves IPO in just 73 days, accelerates offline expansion and mass production, confronting commercialization challenges and competition with industry giants.

Tencent and CATL have invested in DeepSeek, with the former aiming to enrich its ecosystem and the latter securing orders for computing power and energy consumption—each gaining what they need.

AI music company Suno secures $400 million in Series D funding, achieving a valuation of $5.4 billion, becoming the highest-valued company in the industry.

ChatGPT will merge Codex, shifting from conversation to execution—though the name remains, its core has evolved.

Silicon Valley giants once aggressively pushed AI adoption across the entire workforce, but now they are tightening policies amid soaring Token bills and ineffective resource consumption, shifting toward more rational usage.

Stardust Intelligence secures over 1 billion in funding within 3 months, valuation surpasses 10 billion, backed by state-owned capital and industrial investors, and has already secured orders at the thousand-unit scale.

Microsoft's Build conference unveiled several in-house developed models, aiming to reduce reliance on OpenAI and avoid becoming merely an infrastructure conduit in the AI era.

StarSource AI, incubated by Zhiyuan, raises 1 billion CNY in just 10 months, focusing on robotics world models to empower embodied intelligence.

NVIDIA teams up with Microsoft to launch RTX Spark, dismantling the Wintel alliance and aiming to reshape PC architecture through the CUDA ecosystem.

VolcEngine MaaS revenue target raised to 15 billion, Seedance 2.0 achieves monthly revenue exceeding 1 billion, video models become growth engine.

The AI patriarch LeCun establishes AMI, pioneering a novel approach to develop world models, aiming to enable AI to truly understand the world and explore the next-generation AGI paradigm.

Qianxun Intelligence secures over 4.5 billion in funding within three months, with its self-developed model ranking atop global leaderboards—demonstrating formidable technical prowess and successful commercialization, attracting strong capital interest.

DeepSeek achieves significant improvements in hardware efficiency through software-hardware co-optimization, potentially saving trillions of dollars in China's AI infrastructure costs.

ByteDance's AI chief Qu Quan has left the company, previously leading SeedFold to surpass AlphaFold 3, then shifting focus to LLM pre-training—current whereabouts remain undisclosed.

Amid the impact of generative AI, over 220 U.S. unicorns have experienced significant valuation declines, dried-up fundraising, and face discounted acquisitions.