OpenAI不满英伟达芯片？两高管隔空反驳

Core Viewpoint - OpenAI has expressed dissatisfaction with certain latest AI chips from Nvidia and has been seeking alternatives since last year, indicating a potential shift in the relationship between these two leading companies in the AI sector [1] Group 1: Nvidia's Chip Limitations - OpenAI's strategic shift emphasizes the importance of chips used for AI inference, which is the process of AI models responding to user queries [2] - Nvidia's GPUs are well-suited for large-scale data processing necessary for training AI models, but the focus is now shifting towards inference, which may represent a new phase in AI development [2] - Inference requires more memory than training, as chips spend more time retrieving data from memory rather than performing calculations [2] Group 2: OpenAI's Product Performance Issues - The limitations of Nvidia's chips are particularly evident in OpenAI's "Codex" product, which has faced criticism from OpenAI employees regarding performance issues attributed to Nvidia's hardware [3] - OpenAI's CEO Sam Altman noted that clients using OpenAI's coding models prioritize speed, prompting the company to explore partnerships with Cerebras to address these needs [3] Group 3: Investment Negotiations and Delays - Nvidia and OpenAI are currently in investment negotiations, with Nvidia previously planning to invest up to $100 billion in exchange for equity in OpenAI [4] - These negotiations have been prolonged for several months, during which OpenAI has engaged with other companies like AMD to produce competing GPUs [4] - OpenAI's evolving product roadmap has altered its computational resource requirements, complicating negotiations with Nvidia [4] Group 4: Alternative Partnerships - OpenAI has discussed collaborations with startups like Cerebras and Groq to provide faster inference chips, but negotiations with Groq fell through after Nvidia secured a $20 billion licensing deal with them [5] - Nvidia's acquisition of Groq is seen as a strategic move to enhance its technology portfolio in the rapidly changing AI and chip industries [5] Group 5: Rebuttals from Nvidia and OpenAI - Nvidia's CEO Jensen Huang dismissed reports of a strained relationship with OpenAI, asserting that the company plans to make significant investments in OpenAI [6] - Nvidia claims to provide the best performance and total cost of ownership for inference operations, while OpenAI confirmed its reliance on Nvidia for the majority of its inference devices [6] - Sam Altman expressed a desire for a long-term partnership with Nvidia, stating that they produce the best AI chips in the world [6]