Two Paths to the AI Boom: Broadcom’s Custom Chips Face Off Against Nvidia’s Platform Play

The race to supply artificial intelligence compute has created two distinct winners with very different strategies. Broadcom has staked its AI growth on designing bespoke silicon for a handful of very large customers, while Nvidia sells a general-purpose accelerated computing platform that spans a broad base of cloud providers, enterprises, and researchers.

Broadcom’s custom-silicon bet is producing striking results. The company said AI semiconductor revenue reached $16.7 billion in its fiscal third quarter, which ended Aug. 2 — a 221% increase from the year-ago period and now 56% of total company revenue. That figure underscores how central AI silicon has become to Broadcom’s business model, which also spans networking connectivity and infrastructure software.

Management’s forward outlook is even more aggressive. Broadcom projected AI chip revenue of $115 billion for fiscal 2027, scaling to $230 billion in fiscal 2028. Such targets imply continued rapid expansion of custom accelerator programs with hyperscale customers, who commission bespoke chips to handle specific workloads at lower cost and higher efficiency than off-the-shelf parts.

Nvidia, by contrast, has built its franchise on a broader platform. Rather than designing chips for individual customers, the company sells data center scale AI infrastructure — GPUs, networking, and the software ecosystem around them — serving a far more diverse set of buyers and use cases. That breadth has made Nvidia the dominant supplier for general-purpose AI training and inference, though it also means the company faces emerging competition from the very custom programs Broadcom serves.

The market reflected no clear verdict on the debate in recent trading. Broadcom shares closed at $361.99, up 0.71% on the day, valuing the company at roughly $1.99 trillion. Nvidia shares slipped 0.73% to $223.67, leaving its market capitalization at about $5.56 trillion — still more than double Broadcom’s.

The divergence in strategy frames the larger question for the AI hardware market: whether hyperscalers will increasingly internalize chip design through partners like Broadcom, or continue concentrating purchases on Nvidia’s platform, which remains the default for most AI development. The two models are not mutually exclusive — many large cloud operators pursue both custom accelerators and merchant GPUs simultaneously.

Source: original release

What to watch

• Broadcom’s progress toward its stated $115 billion fiscal 2027 AI revenue target as quarterly results are reported.
• New custom accelerator customer announcements or expansions of existing hyperscale chip programs.
• Nvidia’s upcoming earnings and any updates on its data center platform roadmap and product cadence.
• Hyperscaler capital expenditure guidance, which shapes demand for both custom and merchant AI silicon.