Report: Broadcom to launch new network chip, challenging Nvidia's dominance
According to media reports on Monday, Broadcom released the Thor Ultra network chip, aimed at helping enterprises build large-scale AI computing systems. This move further intensifies competition with Nvidia in the field of AI infrastructure. The chip can support the connection of hundreds of thousands of chips, providing the underlying network support for AI applications such as ChatGPT.
Broadcom CEO Hock Tan previously stated the company expects the AI chip market opportunity to reach $60-90 billion by 2027, covering both network chips and data center processors. The latest product launch comes as Broadcom accelerates its presence in the AI market.
This chip release closely follows Broadcom’s major cooperation agreement with OpenAI. Reportedly, Broadcom will begin supplying OpenAI with custom chips valued at $10 billion starting from the second half of 2026, directly challenging Nvidia’s dominance in the AI accelerator market.
The launch of the new chip is expected to reinforce Broadcom’s control in the field of AI data center network communications, and will directly compete with Nvidia’s network interface chips.
Broad Prospects for the AI Market, Rapid Revenue Growth for Broadcom
Broadcom has already shown strong growth momentum in the AI sector, with AI-related revenue reaching $12.2 billion in its 2024 fiscal year. In September, it announced a new customer order worth $10 billion for customizing data center AI chips.
Broadcom Senior Vice President Ram Velaga said in a media interview:
In distributed computing systems, the network plays an extremely important role in building large clusters. Any company engaged in the GPU business wants to ensure it can participate in the network business, which is not surprising.
The company expects its target market size for 2027 to be split between network chips and data center processors, with an overall scale reaching $60-90 billion.
Technical Upgrades Support Large-Scale AI Deployment
As a key component of Broadcom’s network chip product line, the Thor Ultra chip plays a crucial role in connecting AI systems with other parts of the data center. The chip enables infrastructure operators to deploy chip quantities far beyond previous levels, thus building and operating large models to support AI applications.
Broadcom engineers achieved double the bandwidth on Thor Ultra compared to the previous version. From the early production stages, the chip has gone through rigorous testing and evaluation processes.
When manufacturing Thor Ultra or flagship products such as the Tomahawk network switch series, engineers need to build a complete system around the chip. Velaga said engineers discuss key parameters with the hardware systems team, such as chip package type, power requirements, and heat dissipation.
Design-Oriented Strategy, Partnering with Ecosystem Players
Although Broadcom itself does not sell servers, it provides components and system design solutions to customers as a reference for building network infrastructure.
Velaga revealed the company’s investment strategy:
For every dollar we invest in chips, ecosystem partners invest at least six to ten dollars. Therefore, our focus is on design, as if we are almost ready for mass production.
Broadcom’s cooperation with major cloud computing companies has already produced results. The company has participated in the multi-generation design of Google’s Tensor processors, a project that began more than a decade ago. Analysts estimate Tensor chips have brought Broadcom billions of dollars in revenue.
Risk Disclosure and DisclaimerThe market involves risks, and investments should be made with caution. This article does not constitute personal investment advice, nor does it take into account any individual user's special investment objectives, financial situation, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their particular circumstances. Investing based on this information is at your own risk.