When most discussions around Chinese brands still focus on how to expand overseas, Zhang Xue is taking a more radical approach: redefining the stage itself
Embodied AI is moving out of the lab and into real-world environments at increasing speed. The humanoid robot half-marathon scheduled for April 19 in Beijing Economic-Technological Development Area (E-Town) is framed as a public race, but functions more as an industry stress test, compressing a full stack of technologies into a 21-kilometer trial
The global AI boom is shifting infrastructure bottlenecks from GPUs to CPUs, as inference-heavy and agentic AI workloads push compute demands beyond accelerator capacity into system-level constraints
In an era where AI systems are rapidly scaling beyond the limits of traditional silicon, new experimental companies are beginning to question what "compute" itself should look like. One of the most unusual entrants is The Biological Computing Company (TBC), which proposes a hybrid model integrating living neurons with modern machine learning systems to enhance performance, efficiency, and adaptability. The idea sits at the intersection of neuroscience and computing
ASML delivered first-quarter 2026 results that exceeded expectations, prompting management to raise its full-year guidance despite a cautious outlook for the second quarter. The company announced plans to expand its extreme ultraviolet (EUV) manufacturing capacity to meet strong demand expected in 2027. The updated outlook suggests stronger momentum in the second half of 2026
Spain-based AI chip startup Semidynamics has recently seen a wave of positive developments. Its first chip, fabricated on TSMC's 3nm process, has successfully completed tape-out, and just weeks ago, the company secured an investment commitment from SK Hynix. Both significantly enhance its operational credibility and global visibility. CEO Roger Espasa stated that the company's proprietary Gazzillion technology is a specialized memory subsystem designed to address one of AI's most pressing challenges today: inefficient memory utilization, which has led to persistent supply constraints
Agentic AI is pulling CPUs back to the center of the AI stack, turning them into a renewed battleground for chipmakers. After Arm moved into AGI-focused CPU design, Nvidia has followed with a stake in SiFive, a RISC-V IP provider, signaling a broader shift in how control of AI infrastructure is being contested
Flipping an egg takes less than two seconds, but every step involves continuous sensing, judgment, and force control; spreading jam and arranging plates are the same. The difficulty of these atomic skills does not lie in executing fixed trajectories, but in performing correctly in complex environments
As generative AI demand shifts from centralized cloud training to edge inference, Nokia and AI chip startup Blaize have expanded their partnership in Singapore, unveiling a full-stack solution for hybrid heterogeneous computing at GITEX AI Asia
As the global generative AI boom shifts from scale to localization, Southeast Asia is emerging as a critical testing ground for the next phase of development. The region's linguistic diversity and complex cultural landscape are pushing developers to rethink model design, fueling the rise of localized large language models and smaller, more efficient systems tailored to specific use cases
As demand for AI computing continues to surge, terrestrial data centers are facing mounting constraints in power supply, thermal management, and land availability. As a result, the feasibility of space-based computing infrastructure is drawing increasing attention across the industry
AI agents are moving beyond chat into execution, handling tasks, workflows, and decisions. That shift is giving rise to a new hardware category: the agent computer. Built for persistent, local AI operation with direct control over tools and data, it is emerging as a contender for the next core of personal computing
As HPC and AI processors push computing performance to unprecedented levels, transistor density has reached a point where thermal behavior is no longer uniform. Instead of gradual, evenly distributed heating, modern chips exhibit sharp, localized hotspots that concentrate extreme thermal loads within small regions
Large language models (LLMs) have demonstrated three surprising capabilities in recent years: generalization—providing reasonable answers to unseen questions; multitasking—handling programming, translation, financial analysis, and legal interpretation within a single model; and reasoning—breaking down complex problems into stepwise solutions. Together, these abilities have triggered a paradigm shift in the digital world
