Huawei's newest AI chip, the Ascend 950PR, might not deliver strong compute performance relative to NVIDIA for domestic hyperscalers, but it offers a major upgrade with CUDA compatibility. Huawei's Ascend 950PR Sees Massive Interest, Mainly With CUDA-Like Programming Brought In With CANN Next The Chinese computing industry has been trying to challenge NVIDIA's market dominance, and while the focus has been on upgrading offerings in terms of architecture and onboard features, it hasn't worked out to a large extent. Reports suggest that Chinese hyperscalers remain strongly inclined toward NVIDIA's hardware, and a key reason isn't just the compute gap; CUDA […]

Read full article at https://wccftech.com/huawei-ascend-950pr-just-won-over-chinese-customers-by-mimicking-cuda/

The Trump administration is exploring options to address AI chip exports, and initial reports suggest the proposed regulations are far more aggressive than the industry anticipated. The US Is Planning New AI Chip Export Regulations, By Looking at Compute Power Being Shipped Out The debate around AI chip exports has emerged several times since chip manufacturers like NVIDIA and AMD achieved significant compute breakthroughs. This matter was also under intense focus by the Biden administration, which introduced the "AI Diffusion" act that addresses AI chip exports by categorizing countries into different levels, each with its own caveats. The Diffusion Act […]

Read full article at https://wccftech.com/the-us-could-soon-turn-nvidia-and-amds-ai-chips-into-a-foreign-policy-tool/

The debate over the future of Taiwan's chip industry has resurfaced, but for those who haven't noticed, fabless manufacturers have already begun preparing for the worst. Taiwan's Chip Production Shift Is Driven By the Worries of Big Tech CEOs Amid Their Dependency On the Region We have extensively discussed the topic of US-Taiwan and the grand shift of supply chains from the East to the West, but the NYT's latest report has revealed aspects that indicate the geopolitical constraints on Taiwan's chip industry are being taken much more seriously. But one of the more important events to spot in the […]

Read full article at https://wccftech.com/big-tech-now-sees-relying-solely-on-taiwan-for-their-chip-needs-as-a-death-trap/

ASML has reportedly achieved a critical breakthrough in its EUV lithography technology, boosting the power of its light source to 1000W. This leap from the current 600W standard is projected to increase per-machine chip output by 50% by the end of the decade, enabling foundries to process approximately 330 wafers per hour, up from the 220-wafer limit of today's systems.

Generating EUV light remains one of the most complex engineering feats in modern manufacturing. In ASML's laser-produced plasma (LPP) system, microscopic droplets of molten tin are fired through a vacuum and struck by a CO₂ laser. According to Reuters, to achieve the 1,000W threshold, ASML implemented two significant architectural changes: droplet acceleration, which enables the system to fire roughly 100,000 droplets per second (twice the current rate), and two-pulse laser shaping, which uses two smaller laser bursts instead of the single pulse currently in use.

ASML's roadmap suggests this is just the beginning, with internal targets already set for 1500W and, eventually, 2000W. Assuming the same ratio, that would triple the current production rate.

By scaling the power of the 13.5 nm light beam, ASML aims to directly lower the cost-per-chip for advanced AI and logic processors while extending the economic viability of the sub-2 nm era.

KitGuru says: The “photon bottleneck” has long been the primary limiter for EUV economics. By hitting the 1000W mark, ASML is effectively telling the world that it can keep Moore's Law alive through raw power scaling. For companies like TSMC, a 50% increase in wafer production without expanding their footprint is certainly welcome news.

Well, it appears that the chip startup Taalas has found a solution to LLM response latency and performance by creating dedicated hardware that 'hardwires' AI models. Taalas Manages to Achieve 10x Higher TPS With Meta's Llama 8B LLM, That Too With 20x Lower Production Costs When you look at today's world of AI compute, latency is emerging as a massive constraint for modern-day compute providers, mainly because, in an agentic environment, the primary moat lies in token-per-second (TPS) figures and how quickly you can get a task done. One solution the industry sees is integrating SRAM into their offerings, and […]

Read full article at https://wccftech.com/this-new-ai-chipmaker-taalas-hard-wires-ai-models-into-silicon-to-make-them-faster/