Samsung Rises 3.3% as CXMT Launches Fifth-Generation DRAM

ASML Holding won broader customer commitments for its roughly $400 million High-NA EUV machines, even as its U.S. shares fell about 1.8% to $1,713.41 in Thursday morning trading. Intel is already using High-NA in volume production on selected chip layers, Samsung plans to bring the technology into DRAM production in 2028, and TSMC targets advanced chip production from 2030. ASML says its EXE:5200B can print features 1.7 times smaller than its current NXE systems in a single exposure, potentially sparing customers costly manufacturing steps. The catch is price: chipmakers need enough output and yield gains to earn back the cost of each machine, and ASML needs those gains to turn early commitments into lasting demand. At $1,713.41, the shares sit 35.5% above the roughly $1,260 GF Value estimate, a premium that puts pressure on High-NA to deliver on the factory floor.

CLSA's Bhavtosh Vajpayee said the AI semiconductor cycle will not peak in 2028, calling that expectation "utterly wrong" and projecting the current chip shortage will persist until roughly 2030. Speaking to CNBC on Wednesday, Vajpayee said demand for GPUs and custom AI chips currently exceeds supply by about 73%, and he expects upstream semiconductor and hardware companies to benefit as the shortage worsens alongside expanding AI infrastructure spending. He also noted that large semiconductor companies trade at lower valuation multiples than many other parts of the AI ecosystem. Vajpayee highlighted Meta's Muse as a notable recent development for making AI agent creation easier through simple language commands, and said similar advances should emerge in China, where companies including Alibaba are developing smaller AI models. He pointed to Apple as particularly important because smartphones already sit at the center of consumers' daily lives and could become the primary gateway for accessing AI services, while CLSA sees clearer opportunities in AI hardware suppliers than among model developers, where he said it remains too early to determine the eventual winners.

Super Micro Computer has begun commercial shipments of server systems built around NVIDIA's next-generation Vera Rubin platform, sending its shares down more than 3% Thursday afternoon. The company is shipping Vera Rubin NVL72 racks through its Data Center Building Block Solutions, combining computing hardware with liquid cooling, networking, storage and power components; each rack design uses 72 Rubin GPUs and 36 Vera CPUs. Super Micro said the platform can support deployments ranging from five-megawatt projects to data centers spanning gigawatt-scale capacity, and that a scalable unit can include 16 compute racks with 1,152 Rubin GPUs and 331 terabytes of HBM4 memory. At a Sept. 10 Goldman Sachs conference, Senior Vice President Michael Staiger said the company had about $60 billion in orders backing its fiscal 2027 revenue forecast of $65 billion to $72 billion. The company sees demand from neocloud operators, enterprises and sovereign customers.