The AI industry is shifting towards customized chips, and OpenAI’s self-developed chips will be mass-produced
OpenAI and Broadcom collaborate to develop self-developed AI chips: customized solutions to meet computing power needs. Industry background and motivation for cooperation. In recent years, with the rapid development of artificial intelligence technology, the demand for computing power by major technology companies has continued to rise. As a globally leading AI enterprise, OpenAI’s reliance on high-performance computing resources is particularly significant. However, Nvidia, the mainstream GPU supplier in the current market, is struggling to fully meet the growing demand, especially when facing bottlenecks in training complex models. To address this issue, OpenAI has chosen to collaborate deeply with American semiconductor giant Broadcom to jointly develop customized AI chips. The author of the article points out that OpenAI’s move aims to reduce its dependence on Nvidia products while enhancing its flexibility and efficiency in computing power. Through cooperation with Broadcom, OpenAI plans to launch a chip specifically designed for AI model training, which is expected to be delivered next year. This collaboration marks the beginning of the industry’s exploration of alternative solutions to Nvidia chips to meet the larger computing power demands of the future. Cooperation details and market response: Chen Fuyang, CEO of Broadcom, revealed last week that an undisclosed client placed a large order worth $10 billion with the company. Subsequently, informed sources confirmed that the client was OpenAI. Affected by this news, the stock price of Broadcom rose sharply by nearly 11% last Friday, demonstrating the market’s high recognition of the cooperation. Broadcom has significant advantages in the field of customized chips, with its “XPU” series of semiconductors designed specifically for specific application scenarios, which can effectively optimize performance and reduce energy consumption. The collaboration between OpenAI and Broadcom has been ongoing for over a year, with both parties dedicated to developing high-performance chips suitable for model training. HSBC analysts predict that by 2026, Broadcom’s custom chip business will grow much faster than Nvidia’s. It is worth noting that although Broadcom’s chips are seen as a supplementary solution, they do not directly challenge Nvidia’s position. On the contrary, it is aimed at filling OpenAI’s hardware requirements gap and helping it better support complex AI model training tasks. The trend and prospects of customized chips. The author of the article mentioned that OpenAI’s strategy is similar to the practices of tech giants such as Google, Amazon, and Meta. These companies have invested resources in developing proprietary chips to optimize the operational efficiency of AI models. According to analyst Burn from market research firm eMarketer, although Nvidia’s GPU is still the mainstream choice in the industry, customized chips, with their optimization capabilities for specific tasks, can effectively break through existing bottlenecks and provide stronger support for the development of AI technology. In addition, OpenAI also faces enormous cost pressure. According to The Information, OpenAI has significantly raised its cash flow consumption forecast for the next few years, expecting to consume $115 billion in cash by 2029, which is $80 billion higher than the previous forecast. This year’s cash flow consumption will also exceed $8 billion, which is $1.5 billion higher than expected. To cope with the surge in costs, OpenAI plans to further develop and deploy its own data center server chips, in order to achieve higher cost-effectiveness and sustainable development. The collaboration between OpenAI and Broadcom is not only an innovative attempt in the field of technology, but also a microcosm of the industry’s development trend. Through the development of customized chips, OpenAI is expected to alleviate the problem of computing power shortage and promote AI technology to new heights. At the same time, this cooperation also indicates that technology companies are actively exploring diversified solutions to ensure their competitiveness and long-term development in the context of sustained growth in computing power demand.