Broadcom's AI Edge: ASICs Outgrow GPUs, Says Analyst

Broadcom Inc. (AVGO) is making waves in the artificial intelligence (AI) market, with its custom ASICs (Application-Specific Integrated Circuits) gaining traction over traditional GPUs (Graphics Processing Units). Rosenblatt analyst Hans Mosesmann expects Broadcom's AI-related silicon revenue to hit $3.5 billion this quarter, growing 10% sequentially. He predicts that non-AI semiconductor revenue should continue its recovery in CY25, while Infrastructure Software is likely to be driven by VMware strength.

Mosesmann calls the comparison between Broadcom's ASICs and Nvidia's GPUs "an apples-to-oranges comparison." He believes that AI ASICs will outgrow GPU compute over the next several years, highlighting Broadcom's growth potential in AI acceleration, Gen AI, and networking across Cloud and Edge.

ASICs offer several advantages over GPUs in AI applications. They are designed specifically for the task at hand, providing better performance and energy efficiency. ASICs can also be optimized for specific AI workloads, such as image recognition or natural language processing, leading to further improvements in performance and power consumption.

However, ASICs also have some disadvantages. They are more expensive to design and manufacture than GPUs, which are mass-produced for the gaming market. Additionally, ASICs are less flexible than GPUs, as they are designed for specific tasks and cannot be repurposed for other applications.

Broadcom's ASIC strategy caters to the unique demands of AI workloads by focusing on high-speed, low-cost connections between processors, unlike Nvidia's GPUs. Broadcom's ASICs excel in internal connectivity for high-performance AI workloads, while Nvidia's GPUs are better suited for graphics and gaming. This 'apples-to-oranges comparison' highlights Broadcom's growth potential in AI acceleration, Gen AI, and networking across Cloud and Edge, as per Rosenblatt analyst Hans Mosesmann.

ASICs' specialized design and architecture contribute to their efficiency in AI workloads compared to GPUs. Their tailored design enables them to outperform GPUs in AI workloads, as seen in Broadcom's custom ASICs, which are expected to outgrow GPUs in the coming years.

In what ways do ASICs' low power consumption and high-speed connectivity provide an edge over GPUs in AI applications? ASICs' low power consumption and high-speed connectivity give them an edge over GPUs in AI applications. ASICs are designed for specific tasks, like AI inference, leading to 10x-100x better performance per watt compared to GPUs (Source: [1]). This efficiency is crucial for data centers and edge devices, where power consumption is a significant concern. Additionally, ASICs offer high-speed, low-latency connectivity, enabling faster data processing and reduced communication overhead. This is particularly important in AI workloads, where data transfer rates can bottleneck performance.



In conclusion, Broadcom's ASICs offer significant advantages over GPUs in AI applications, but they also come with some drawbacks. As AI continues to grow in importance, the competition between ASICs and GPUs will be an interesting dynamic to watch. Investors should keep an eye on Broadcom's progress in the AI market and consider the potential impact of ASICs on the company's future growth.