Broadcom’s AI semiconductor business is forecast to reach $230 billion by 2028, a trajectory that positions the chipmaker as the most credible long-term challenger to Nvidia’s dominance.
That figure represents a doubling from the projected $115 billion Broadcom expects to generate in 2027, itself nearly double the roughly $58 billion the company estimates for its AI segment in 2026.
The explosive growth curve is drawing attention to a corner of the semiconductor market that has largely operated in Nvidia’s shadow: custom AI chips designed for specialized, repetitive workloads.
Unlike general-purpose graphics processing units, custom silicon can be optimized for specific tasks, giving hyperscalers a compelling reason to commission their own chips rather than buying off the shelf.
Broadcom has assembled a client roster that reflects just how seriously the industry’s biggest players are taking this shift, with partners including Alphabet, Meta Platforms, Anthropic, and OpenAI.
Rival custom chip designer Marvell has also secured meaningful contracts, working with Microsoft and Amazon, but its scale remains far smaller, with AI data center revenue of $2.1 billion in the most recent quarter.
Broadcom’s comparable figure stood at $16.7 billion in AI semiconductor revenue, underscoring the gap between the two custom chip firms and why analysts view Broadcom as the primary competitive threat to Nvidia.
Nvidia’s own numbers remain formidable: its data center division posted $89 billion in revenue during the second quarter ended July, expanding at a 117% annual pace that outstripped even AMD’s growth rate.
Advanced Micro Devices, the most frequently cited GPU rival to Nvidia, generated just $6.7 billion from its data center segment in the same period, growing at 107% despite operating from a much smaller base where faster percentage gains are typically easier to achieve.
The reason AMD has struggled to close the gap traces back to software as much as hardware, with Nvidia’s CUDA platform giving developers a deeply entrenched tool for controlling how GPUs handle AI calculations that AMD has never fully matched.
That software lock-in means the vast majority of existing AI infrastructure is already configured for Nvidia hardware, and the switching costs involved make a large-scale migration to AMD products highly unlikely in the near term.
Broadcom sidesteps this dynamic entirely by targeting workloads where customers are building new infrastructure rather than replacing existing Nvidia setups, which makes switching costs largely irrelevant to its growth story.
Broadcom is expected to grow at a faster pace than Nvidia over the next few years, a projection that shifts the competitive narrative away from the GPU-versus-GPU battle and toward a broader contest over how AI computing gets done at scale.
