Loading market data...

Broadcom and Marvell Post Record Revenues as Custom AI Chips Reshape Data Centers

Broadcom and Marvell Post Record Revenues as Custom AI Chips Reshape Data Centers

Broadcom and Marvell have posted record revenues for fiscal 2026, cementing their position as the dominant players in the AI chip market. The surge is driven by a shift toward custom-designed chips tailored for specific data center workloads — a trend that's reshaping how tech companies build and invest in infrastructure.

Record Revenues for AI Chip Leaders

Both companies reported unprecedented sales figures in the fiscal year 2026, though exact numbers were not disclosed. Broadcom's AI-related revenue, which includes custom accelerators and networking chips, accounted for a growing share of its semiconductor business. Marvell similarly saw a jump in demand for its custom ASICs and data center connectivity products. The records reflect an insatiable appetite from cloud providers and large enterprises for specialized hardware that can handle the massive compute loads of training and running AI models.

Custom Chips Reshape Data Center Strategies

The rise of custom AI chips marks a strategic departure from relying solely on off-the-shelf GPUs from companies like Nvidia. Broadcom and Marvell have long supplied custom silicon to hyperscale clients, but fiscal 2026 saw that business accelerate sharply. Data center operators are increasingly designing their own chip architectures or commissioning bespoke designs to optimize performance per watt and reduce costs. Broadcom's Tomahawk and Jericho families switch chips, for example, are tailored for high-bandwidth AI clusters. Marvell's custom ASICs, meanwhile, are used in everything from inference accelerators to storage controllers. This shift means data center buildouts are becoming more specialized, with chip choices tied directly to the workloads they run.

Impact on Tech Investments and Innovation

The dominance of Broadcom and Marvell in custom AI chips is redirecting capital flows in the semiconductor industry. Venture funding for AI hardware startups has slowed, as established players lock in long-term supply deals with major cloud providers. At the same time, the push for custom chips is driving innovation in packaging, memory integration, and optical interconnects — areas where Broadcom and Marvell both have deep expertise. Investors are watching closely: if custom designs continue to displace general-purpose processors, the competitive landscape for AI hardware could narrow further, concentrating power among the few firms that can deliver both volume and customization.

What remains unclear is how other chipmakers — particularly Nvidia and AMD — will respond to the rise of custom alternatives. Broadcom and Marvell have shown that the market is big enough for both general-purpose and purpose-built chips, but the revenue records suggest the pendulum is swinging toward the latter. For now, the two companies are riding a wave that shows no signs of cresting.