Nvidia may still set the pace in AI accelerators, but Broadcom is finding plenty of room beside it. Demand for custom chips built around the needs of individual AI companies and hyperscalers is turning into a much larger business for the semiconductor vendor.
Broadcom’s AI semiconductor revenue reached $16.7 billion in its fiscal third quarter of 2026, up 221% year over year, with custom accelerators accounting for 73% of that business. Google, Meta, OpenAI, and Anthropic are all tied to Broadcom custom silicon programs, putting some of the world’s biggest AI spenders behind its growth.
Big tech isn’t replacing Nvidia hardware. Instead, custom silicon is taking over predictable, high-volume workloads, leaving general-purpose GPUs to handle tasks that require maximum flexibility.
Broadcom is winning customers with chips built for specific workloads
Crypto Briefing reported that Broadcom’s custom accelerators drove most of its third-quarter AI revenue. Unlike Nvidia’s general-purpose GPUs, Broadcom develops application-specific integrated circuits, or ASICs, around the requirements of individual customers.
Google has worked with Broadcom across multiple generations of its Tensor Processing Units. Meta is moving toward production shipments of its MTIA accelerator, while CNBC said that OpenAI is preparing a second-generation custom chip and planning a third with Broadcom.
Broadcom CEO Hock Tan also said the company expects to accelerate shipments of Google Ironwood TPUs to Anthropic and deliver tens of billions of dollars of processors to Google annually over the next several years.
These programs give Broadcom more than one-time chip sales. Custom silicon typically requires close collaboration across hardware, software, and infrastructure, making established relationships harder to replace once systems are deployed at scale.
Nvidia still leads when customers need flexibility
Broadcom’s growth does not mean Nvidia is being pushed out of AI infrastructure.
Nvidia remains the dominant choice for large-model training and other workloads where customers want flexible computing backed by its mature CUDA software ecosystem. According to Yahoo Finance, Nvidia generated $89 billion in Data Center revenue in its fiscal second quarter of 2027, showing the scale of its lead.
Custom accelerators become more attractive when hyperscalers know precisely which workloads they need to run repeatedly. A chip optimized for a specific inference or recommendation workload can make more sense than deploying general-purpose hardware everywhere.
Nvidia still has the edge for flexible AI workloads, while Broadcom is gaining ground with customers that want chips tailored to specific tasks.
Broadcom’s advantage extends beyond custom accelerators
Broadcom also competes in the infrastructure surrounding AI processors. Its portfolio includes Ethernet switching, SerDes, interconnect, and other networking technologies needed to move data across increasingly large GPU and custom XPU clusters.
Yahoo Finance noted that Broadcom’s networking position strengthens its role as customers scale both types of infrastructure. That reach gives Broadcom opportunities even when Nvidia GPUs remain part of the deployment.
Broadcom expects the AI business to expand rapidly from here. CNBC reported that Tan is targeting $115 billion in AI revenue for fiscal 2027 and $230 billion in fiscal 2028.
Channel partners may need to support both architectures
For channel partners, Broadcom’s momentum makes AI infrastructure less about choosing one chip vendor.
Customers may increasingly combine Nvidia GPUs for flexible training and computing with specialized accelerators designed for high-volume inference or other predictable workloads. Those environments create requirements around networking, storage, power, cooling, software, and integration regardless of which processor sits at the center.
Broadcom’s growth therefore adds another infrastructure path rather than replacing Nvidia outright. Partners that can connect the two worlds may be in a stronger position as AI environments become more specialized and complex.
Explore how Broadcom’s hardware strategy is creating an alternative to Nvidia’s flagship GPUs in enterprise data centers.