Broadcom's AI revenue increased by 63% year over year, outpacing the 56% growth in Nvidia's data center revenue. The latter's growth is quite impressive considering its huge size, but there is a good chance that Broadcom will be able to significantly accelerate its healthy growth due to a shift in AI workload needs.
Earlier this year, Advanced Micro Devices CEO Lisa Su said that the demand for AI inference applications is outpacing the demand for AI model training. Su said that the number of purpose-built AI models for handling specific kinds of workloads in areas such as automaking, healthcare, and finance is also booming. So the need for custom AI processors designed by Broadcom is set to increase.
The company's custom processors are designed to tackle specific workloads as compared to the general-purpose function of Nvidia's GPUs. Custom AI processors not only consume less power, but they are also considered to be more powerful while performing a specific task, when compared to GPUs.
This explains why customers have been lining up to buy custom AI processors and networking chips from Broadcom. OpenAI, which relied on Nvidia to train ChatGPT three years ago, has been the latest company to partner with Broadcom.
The company will design and deploy 10 gigawatts of custom AI processors for OpenAI from 2026 to 2029. Though there has been no disclosure about the financial aspects of this deal, it is estimated that $35 billion worth of AI accelerators are typically deployed in a 1 gigawatt data center. One analyst estimates that this deal could add a total of $100 billion to Broadcom's top line during the contract period.
As such, Broadcom could see a significant expansion in its addressable market following the OpenAI deal. The chip designer reported a strong revenue backlog of $110 billion at the end of its previous fiscal quarter, a number that's well above the $60 billion in revenue that it has generated in the past year. That backlog is likely to have grown much bigger in the wake of the OpenAI deal, as well as the $10 billion contract that it landed from an unnamed client last month.
In all, Broadcom's growing backlog, the fact that it counts the likes of OpenAI, Alphabet's Google, Meta Platforms, and ByteDance as customers -- along with other customers in the company's pipeline -- clearly suggest that its status in the AI chip industry is rising. Moreover, 80% of the chips performing AI inference tasks in 2030 are expected to be ASICs, up from just 15% last year*.
Broadcom is the leading player in custom AI processors, with an estimated share of 70%, so it is one of the best ways to capitalize on the huge potential of this market. Management expects to ride the custom AI processor market's growth and corner a 24% share of the overall AI chip market by 2027. That would be more than double its estimated share of 11% in 2025, suggesting that it is likely to eat into Nvidia's dominance.
Could This Semiconductor Leader Become the New Face of Artificial Intelligence (AI)?
*Rolling MoAPS disruption JUST started.
SVG/Cymer/Brion/HMI/Mapper/ASML/et al stuff oh soooo enabling.
With Village all in. |
