Core Insights - The article discusses the escalating competition for core computing resources triggered by the suspension of tariffs, leading to significant price fluctuations in server prices, which have increased by 15%-20% recently [2][4] - The emergence of new high-end products from NVIDIA, such as the Hooper and Blackwell series, is reshaping the supply landscape, with limited suppliers controlling the market [3][6] - The article highlights the complexities of the supply chain and the hidden trading networks that have developed in response to the demand for high-performance computing [8][10] Group 1 - The NVIDIA Hooper series, particularly the H200, is in high demand, with suppliers capable of providing 100 units weekly, as the market shifts from H100 due to its discontinuation [6][10] - The supply chain for computing resources is characterized by a lack of transparency, with contracts often abstracting the specific hardware used, focusing instead on computing power units [7][8] - The rise of speculative trading in high-end GPUs has led to inflated prices, with reports of individual suppliers marking up NVIDIA A100 GPUs to 128,000 RMB, significantly above the official price [10][11] Group 2 - The rapid construction of intelligent computing centers has resulted in over 458 projects initiated in 2024 alone, but many remain in the planning or construction phases, indicating a potential bubble in the sector [11][13] - The article notes that many of these centers are underutilized, with less than 50% activation rates, primarily due to the performance limitations of domestic chips and outdated server technology [15][19] - Major companies like ByteDance and Alibaba are making substantial investments in AI infrastructure, with ByteDance planning to invest over $12.3 billion in AI by 2025, highlighting a stark contrast to the struggling smaller suppliers [17][18][20] Group 3 - The article discusses the shift in focus from pre-training to inference in AI applications, indicating a growing demand for computing resources in various sectors, including automotive [30][31] - Despite the increasing demand for inference, the article points out a mismatch in supply, with many domestic chips unable to meet the performance standards required for advanced AI tasks [32][33] - The lack of a cohesive ecosystem and the need for a "blood-producing" nurturing environment for the intelligent computing industry are emphasized as critical challenges that need to be addressed [40]