Industry Focus - Pharmaceutical Development - The pharmaceutical industry faces high costs (billions of dollars) and lengthy timelines (years) for drug development, including FDA approval [1] - Traditional drug development involves sequential or parallel exploration of multiple avenues, which can be inefficient [1] Technological Solution - Digital Twins and Omniverse - The company proposes using digital twins in the Omniverse to simulate drug development processes [1] - The Omniverse is described as a "phenomenal thing" that can revolutionize how things are done [1] Potential Benefits - Combining attributes from different approaches (e.g., "pass number one" and "pass number four") in the digital environment can maximize the likelihood of success [1] - This approach can compress the time to market for new drugs and maximize their benefits [1]

AI Infrastructure and Architecture - Infinia was conceived due to the need for a different architecture for AI, one that scales efficiently for training, has low latency, is distributed on-premise and multi-cloud, and minimizes data movement [1] - The industry is shifting towards Data Intelligence, reframing storage of raw data into informational form, which is a new opportunity for DDN to provide data intelligence for enterprises running AI [1] - Metadata and tagging are essential for multimodal AI, enabling the movement of metadata and making the economics viable due to the compression ratio [1] AI Application and Adoption - Enterprises need to adopt AI at an accelerated pace, requiring the application layer to be supercharged and the infrastructure to be efficient [1] - The industry is moving from high-performance computing to Enterprise, and then to digital twins of Enterprise, enabled by technologies like Omniverse [2] - AI is enabling companies to create digital twins, allowing them to run thousands of experiments simultaneously and optimize outcomes, applicable to enterprises, governments, and individuals [2] AI Model and Ecosystem - Post-training, which involves problem-solving and reasoning, is a crucial and compute-intensive part of intelligence, following pre-training [3] - The release of open-source reasoning models like DeepSeek's R1 is accelerating AI adoption by highlighting opportunities for more efficient models [3] - The CUDA ecosystem is enabling the application of AI in specific industries like Life Sciences, Financial Services, and autonomous driving [3] Strategic Partnership and Future Vision - The partnership between Nvidia and DDN is expanding from supercomputing to Enterprise and Omniverse, with Infinia playing a key role [4] - Companies should both use public cloud AI and build their own specialized AI, curating AI agents from various sources to solve large problems [3] - Differentiation for organizations comes from specialized application of AI, enabled by technologies like Nvidia's Nims and DDN's Infinia [4]

中国银河证券：推理算力重要性提升 光模块等算力 细分赛道发展再加速 智通财经APP获悉，中国银河证券发布研报称，近日，英伟达GPU技术大会2025(GTC 2025)结束，相较 于此前GTC大会，硬件性能提升符合预期，英伟达对软件的布局、推理算力以及Agent的愿景更强，硬 件发展带动软件及大模型持续高景气。在算力需求总体增长的情况下，当前算力相关板块仍然具备较大 投资价值，建议优选空间较大且动能充足的细分子板块：运营商、光模块、光芯片以及AIDC相关等。 当前算力板块在推理应用的不断发展中并非呈现出算力需求下降的态势，而是算力需求得到进一步的刺 激。 中国银河证券主要观点如下： 推理算力预计将持续增长，Scaling Law仍将持续 硬件方面：英伟达发布基于Blackwell架构的升级版Blackwell Ultra，并着重强调了其在推理端的重大进 展，可以为数据中心提供 50倍增收的机会；英伟达也明确了26-27年乃至更远期的发展规划，硬件性能 进一步加速已成定局；CPO交换机方面，英伟达展示了基于 1.6T硅光引擎的CPO(共封装光学)交换机系 列，硬件端性能未来仍将持续高质量全方位提升。 软件方面：英 ...