--- title: "NVIDIA plans to launch a new chip, OpenAI is a major customer" type: "News" locale: "zh-CN" url: "https://longbridge.com/zh-CN/news/277270597.md" description: "NVIDIA plans to launch a new processor customized for OpenAI and other clients, aimed at enhancing the speed and efficiency of AI tools. This significant business adjustment could redefine the AI competition landscape. The new platform will be announced at next month's NVIDIA GTC developer conference and will integrate chips designed by Groq. OpenAI has agreed to become one of the main customers for this processor, marking an important victory for NVIDIA" datetime: "2026-02-28T03:16:04.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/277270597.md) - [en](https://longbridge.com/en/news/277270597.md) - [zh-HK](https://longbridge.com/zh-HK/news/277270597.md) --- > 支持的语言: [English](https://longbridge.com/en/news/277270597.md) | [繁體中文](https://longbridge.com/zh-HK/news/277270597.md) # NVIDIA plans to launch a new chip, OpenAI is a major customer NVIDIA plans to release a brand new processor customized for OpenAI and other clients, aimed at creating faster and more efficient tools. This marks a significant shift in its business, which could redefine the AI competition landscape. According to informed sources, the company is designing a new system for AI inference computing—this type of computing is responsible for enabling AI models to respond to user requests. This new platform will be officially announced next month at the NVIDIA GTC developer conference in San Jose and will integrate chips designed by the startup Groq. Inference computing has become the focal point of intense industry competition. Competitors Google and Amazon have both launched chips to rival NVIDIA's flagship products. At the same time, the explosive growth of self-coding technology in the tech industry has created a demand for new types of chips that can handle complex AI tasks more efficiently. Some insiders have stated that OpenAI has agreed to become one of the largest customers for this new processor, which is a significant victory for NVIDIA. The developer of ChatGPT has already been a core customer of NVIDIA and has been seeking more efficient alternatives to NVIDIA chips over the past few months, signing a contract with a chip startup last month to add supply options. Earlier last Friday, OpenAI indirectly mentioned this new processor when announcing plans to purchase dedicated inference computing power from NVIDIA on a large scale, while NVIDIA also made a $30 billion investment in it. OpenAI has also signed a significant new agreement with Amazon to use its Trainium chips. NVIDIA dominates the design and sale of GPUs (graphics processing units), which can perform billions of simple tasks simultaneously. However, since the rise of the AI boom, NVIDIA has faced performance bottlenecks with its flagship products for the first time. As the market shifts towards inference, some customers are pressuring NVIDIA to release chips that can more efficiently support AI applications. NVIDIA's high-performance Hopper, Blackwell, and Rubin series GPUs are recognized in the industry as top products for training large-scale AI models, and they are priced high. Most analysts estimate that NVIDIA controls over 90% of the GPU market share. NVIDIA CEO Jensen Huang has long claimed that NVIDIA GPUs lead the market in both training and inference scenarios, and this versatility is the core appeal of the products. However, over the past year, as companies have begun deploying AI agents and other tools in an attempt to disrupt hundreds of industries and generate huge profits through subscription fees, high-end computing demand has shifted from training to inference. AI agents are AI systems that can relatively autonomously complete tasks for users. Many companies developing and operating AI agents have found that GPU costs are too high, power consumption is too great, and they do not fully meet the actual operational needs of the models. With the rapid rise of AI agents, NVIDIA is under immense pressure to develop inference chips that are lower in cost and higher in energy efficiency. Last month, OpenAI reached a multi-billion dollar computing partnership with Cerebras. Cerebras focuses on inference chips, and its CEO Andrew Feldman stated that their chip is faster than NVIDIA GPUs According to previous reports, as early as last autumn, OpenAI engineers proposed the need for faster inference chips for agent coding applications, and the company subsequently began negotiations with Cerebras. Additionally, it was reported that NVIDIA agreed at the end of last year to acquire key technology licensing from Groq for $20 billion and to bring in its core management team, including founder Jonathan Ross, marking one of the largest "talent acquisitions" in Silicon Valley history. The chips designed by Groq use a completely different architecture from NVIDIA, called a language processing unit, which is highly efficient in inference capabilities. However, as of now, NVIDIA has remained tight-lipped about how it plans to utilize Groq's technology. AI inference computing is mainly divided into two major stages: \- Pre-filling: The process by which the model understands user prompts. \- Decoding: The process by which the model generates responses word by word. For large AI models, pre-filling is usually faster, while decoding is often particularly slow. Coding applications have become one of the most important and profitable application scenarios for enterprise AI, with Anthropic's Claude Code recognized as a leader in this field. However, Anthropic primarily relies on chips designed by teams under Amazon Web Services and Google Cloud to support its models, rather than NVIDIA. Nevertheless, one of Claude's main competitors is OpenAI's rapidly growing Codex tool. Insiders have indicated that OpenAI plans to upgrade Codex using NVIDIA's new system. In the past, NVIDIA would pair its own Vera CPU (central processing unit) with Rubin GPU in high-performance data center servers, but some large clients have found that certain AI agent tasks run more efficiently using only the CPU. This month, NVIDIA announced an expansion of its collaboration with Meta, which includes the first large-scale pure CPU deployment in history to support Meta's advertising-targeting AI agents. This deal has revealed NVIDIA's strategy: to step outside the GPU realm and capture the AI niche market ### 相关股票 - [NVIDIA (NVDA.US)](https://longbridge.com/zh-CN/quote/NVDA.US.md) - [OpenAI (OpenAI.NA)](https://longbridge.com/zh-CN/quote/OpenAI.NA.md) - [Direxion Daily NVDA Bull 2X Shares (NVDU.US)](https://longbridge.com/zh-CN/quote/NVDU.US.md) - [T-Rex 2X Long NVIDIA Daily Target ETF (NVDX.US)](https://longbridge.com/zh-CN/quote/NVDX.US.md) - [VanEck Semiconductor ETF (SMH.US)](https://longbridge.com/zh-CN/quote/SMH.US.md) - [GraniteShares 2x Long NVDA Daily ETF (NVDL.US)](https://longbridge.com/zh-CN/quote/NVDL.US.md) - [XL2CSOPNVDA (07788.HK)](https://longbridge.com/zh-CN/quote/07788.HK.md) - [XI2CSOPNVDA (07388.HK)](https://longbridge.com/zh-CN/quote/07388.HK.md) - [YieldMax NVDA Option Income Strategy ETF (NVDY.US)](https://longbridge.com/zh-CN/quote/NVDY.US.md) - [Direxion Daily NVDA Bear 1X ETF (NVDD.US)](https://longbridge.com/zh-CN/quote/NVDD.US.md) - [T-Rex 2X Inverse NVIDIA Daily Target ETF (NVDQ.US)](https://longbridge.com/zh-CN/quote/NVDQ.US.md) - [Direxion Semicon Bull 3X (SOXL.US)](https://longbridge.com/zh-CN/quote/SOXL.US.md) ## 相关资讯与研究 - [Nvidia Asked by US Senators to Provide Details on Groq Deal](https://longbridge.com/zh-CN/news/279931631.md) - [OpenAI expected to notify staff about strategy changes in coming weeks - WSJ](https://longbridge.com/zh-CN/news/279339086.md) - [World launches tool to verify humans behind AI shopping agents](https://longbridge.com/zh-CN/news/279456181.md) - [LPU: Nvidia popularizes a new AI acronym](https://longbridge.com/zh-CN/news/279662346.md) - [Nvidia to focus on competition-beating AI advances at megaconference](https://longbridge.com/zh-CN/news/279187247.md)