光联芯科CEO 陈超：光互连是通往AGI的必由之路｜WISE2025 商业之王

Group 1: Event Overview - The 36Kr WISE 2025 Business King Conference, recognized as an annual technology and business trendsetter, took place in Beijing on November 27-28 [1] - This year's WISE transformed from a traditional industry summit into an immersive experience using "tech short dramas" to convey insights on trends and business practices [1] Group 2: AI Development and Challenges - AI development has significantly accelerated in the past decade, driven by a more than 1 billion times increase in computing power, with year-on-year growth approaching 10 times [4] - The evolution of AI is heavily reliant on computing power, with predictions indicating that achieving Artificial General Intelligence (AGI) will require approximately 10^41 FLOPs, while current capabilities are around 10^25 FLOPs, indicating a gap of about 10^16 [4][5] - Two major challenges facing the computing power industry are bandwidth limitations and energy consumption, with interconnect bandwidth growth lagging significantly behind computing power growth [5][7] Group 3: Interconnect Bandwidth and Energy Consumption - The performance of computing clusters is severely limited by interconnect bandwidth, which has only increased 30 times over the past 20 years, compared to a 60,000 times increase in chip computing power [5] - Current data centers, such as those being developed by OpenAI and Musk's xAI, require energy comparable to nuclear power plants, highlighting the unsustainable nature of current energy demands for computing [7] Group 4: Optical Interconnect Solutions - Optical interconnects are proposed as the optimal solution to overcome computing power bottlenecks, allowing for high bandwidth and low energy consumption [8] - The company aims to provide next-generation optical interconnect solutions that enable direct optical connections between chips, significantly enhancing bandwidth efficiency [8][9] - The vision includes creating a domestic semiconductor ecosystem that can potentially surpass foreign competitors in overall computing cluster performance and energy efficiency [9]