Group 1 - The semiconductor industry index rose by 0.53%, with notable increases in stocks such as Zhongke Feimeng (up 4.69%) and Youyan Silicon (up 4.49%) [3] - The semiconductor industry ETF (159582) increased by 0.14%, reaching a latest price of 2.21 yuan, with a cumulative rise of 4.79% over the past two weeks [3] - The ETF had a turnover rate of 8.28% and a trading volume of 37.06 million yuan, with an average daily trading volume of 72.87 million yuan over the past week [3] Group 2 - NVIDIA's CEO Jensen Huang is visiting TSMC's 3nm factory to discuss increased orders for AI chips, leading to a 45% to 50% expansion in TSMC's 3nm capacity [3] - TSMC is responding to NVIDIA's additional orders for AI chips, including the Blackwell Ultra and the upcoming Vera Rubin [3] Group 3 - Google Cloud announced the release of its seventh-generation TPU, "Ironwood," which boasts a peak performance increase of up to 10 times compared to TPU v5p and over 4 times compared to TPU v6e [4] - The semiconductor industry ETF reached a new high in scale at 450 million yuan [4] Group 4 - The semiconductor industry ETF experienced a net outflow of 3.27 million yuan, but had a total inflow of 41.48 million yuan over the past five trading days [5] - The index includes up to 40 companies involved in semiconductor materials, equipment, and applications, with the top ten stocks accounting for 78.04% of the index [5]

Investment Rating - The industry investment rating is "Overweight" [1] Core Insights - Broadcom has launched the first 102.4Tbps super chip, which doubles the bandwidth of current Ethernet switches. This chip, named Tomahawk 6, is designed for the AI era and can support ultra-large GPU clusters, driving up to 100,000 GPUs simultaneously. The theoretical peak bandwidth of Tomahawk 6 is 102Tbps, equivalent to processing 25,000 4K movies per second, achieving a sixfold throughput improvement over the previous Tomahawk 5 chip [4] - The acceleration of AI infrastructure construction is highlighted, with major internet companies like Amazon and Google investing in self-developed ASIC chips. Google's seventh-generation TPU (Ironwood) can scale to a cluster of 9,216 chips, with performance twice that of the sixth-generation TPU (Trillium) and energy efficiency nearly 30 times higher than Google's first Cloud TPU from 2018. Amazon's Trainium2 chip, set to be deployed in data centers by the end of 2024, boasts a fourfold performance increase and a threefold memory capacity increase compared to its predecessor [5] - The demand for server PCBs continues to grow under AI infrastructure, as server PCB products need to keep pace with the generational updates of server chips. The product lifecycle is generally 3-5 years, with a maturity period of 2-3 years. As the requirements for signal transmission rates, data transmission loss, and wiring density increase with each generation of chip platforms, server PCB products must also be upgraded accordingly. Companies involved in AI infrastructure construction include Jingwang Electronics, Shengyi Technology, Huitian Technology, and Shengyi Electronics [6]