英伟达GTC大会前瞻：三大看点！

Core Viewpoint - The upcoming NVIDIA GTC conference is expected to signal a significant shift in the AI industry, particularly focusing on the transition from training to inference and adjustments in supply chain strategies [3][4][5]. Group 1: Key Signals from GTC - NVIDIA may leverage the integration of Groq technology to make a substantial entry into the AI inference market [5][6]. - The chip manufacturing process may shift from TSMC to Samsung, marking a potential break from TSMC's long-standing monopoly [5][7]. - The ecosystem for physical AI and open-source models is anticipated to expand further [5][10]. Group 2: Inference Market Focus - The AI industry is transitioning from a "training-first" approach to a "inference-driven" model, with NVIDIA's strategy being closely monitored [6]. - NVIDIA is expected to announce a new chip system that integrates Groq technology, which was acquired for approximately $20 billion [6]. - Groq's chips, known as Language Processing Units (LPU), are optimized for inference workloads, representing NVIDIA's first integration of another company's AI processor into its server architecture [6]. Group 3: Supply Chain and Client Developments - The Groq LPU is projected to be manufactured by Samsung in the latter half of the year, which could signify a shift in NVIDIA's reliance on a single supplier [7][8]. - OpenAI is expected to be one of the first customers for the new chip system, potentially utilizing it for AI tasks such as coding [8]. Group 4: Architectural Changes and Future Technology - The new system architecture will differ significantly from existing setups, featuring 256 Groq chips per rack, with Intel processors managing communication [9]. - NVIDIA is exploring deeper integration of LPU into future product roadmaps, including a potential single-chip solution combining Groq processors with next-generation Feynman GPUs [9]. Group 5: AI Application Ecosystem Expansion - NVIDIA's advancements in robotics and physical AI are gaining attention, especially in the context of the rapidly developing humanoid robot industry in China [10]. - The company is also progressing in the open-source model space, having released a 120 billion parameter model and planning to launch a new model with four times the parameters, which could lower AI inference costs and improve ROI [10]. Group 6: Long-term Industry Impact - The signals released at this GTC conference are likely to significantly influence the AI industry landscape by 2026 [11].