Broadcom's custom silicon business for AI—including custom ASICs and networking—has quietly grown into a major competitive power alongside Nvidia and AMD.
Nvidia and Advanced Micro Devices are among the leading designers of chips used in artificial intelligence data centers, which explains their robust financial growth. The graphics processing units and server central processing units designed by both companies power AI workloads including model training, inference, and AI agent deployment. Both semiconductor stocks have delivered strong returns, with the potential for continued gains as AI infrastructure investments accelerate.
However, Broadcom is quietly emerging as serious competition for Nvidia and AMD in the AI chip market. Unlike its rivals, Broadcom designs application-specific integrated circuits (ASICs)—custom processors that address a distinct niche from the general-purpose server CPUs and GPUs that Nvidia and AMD produce.
The demand for custom AI chips has surged due to their superior cost efficiency, particularly for inference workloads. Custom AI processors can reduce inference costs by 40% to 60% compared to GPUs, prompting major hyperscalers and AI companies to develop in-house chips to lower operating costs. Broadcom serves six core customers for its custom AI silicon: Alphabet's Google, Meta Platforms, OpenAI, and Anthropic.
Broadcom's customers are deploying these custom processors at scale. The company's custom AI chip shipments increased 3.5x year-over-year in the third quarter of fiscal 2026 (ended August 2), driving a 221% year-over-year increase in AI semiconductor revenue to $16.7 billion. Custom silicon accounted for nearly three-fourths of Broadcom's AI revenue.
Looking ahead, Broadcom expects its key customers to accelerate custom AI processor deployments. The company projects AI revenue of $21.7 billion in the current quarter—a 236% year-over-year increase—putting it on track for $58 billion in fiscal 2026 AI revenue, surpassing its original $56 billion forecast. Broadcom anticipates exponential growth in custom chip demand beyond fiscal 2026. For Anthropic alone, the company will deploy 1 gigawatt of custom AI chips in 2026, scaling to 5 gigawatts in 2027 and 10 gigawatts in 2028. Google and OpenAI are expected to accelerate their custom silicon deployments similarly. This confidence is reflected in Broadcom's fiscal 2027 AI revenue target of $115 billion, followed by $230 billion in fiscal 2028. Notably, the company has secured the supply chain needed to support this growth.
Broadcom's AI business already dwarfs AMD's. The company's $58 billion fiscal 2026 AI revenue forecast far exceeds AMD's $25 billion data center revenue run rate for 2026 (AMD generated $12.5 billion in data center revenue in the first half of 2026). Broadcom's fiscal 2028 projection of $230 billion positions it as the second-largest company in AI chips.
Nvidia, by contrast, generated $164 billion in data center revenue in the first six months of fiscal 2027—a near $330 billion annualized run rate—with expectations for 70% top-line growth in fiscal 2028. Since over 90% of Nvidia's revenue derives from data center chips, the company could see data center revenue exceed $800 billion in fiscal 2028. Nvidia will remain the AI market leader, but Broadcom's steeper projected growth rate could eventually narrow the gap.
From a valuation standpoint, all three companies are projected to deliver impressive long-term earnings per share growth, with AMD expected to lead, followed by Broadcom and Nvidia. However, for investors seeking a blend of value and growth, Broadcom offers compelling terms: it trades at 19 times forward earnings, compared to Nvidia's 25x multiple and AMD's 39x multiple. This combination of accelerating growth and lower valuation positions Broadcom as a compelling AI chip stock for long-term gains relative to its peers.