(Reuters) - Broadcom said on Thursday its custom chip unit, which makes AI processors for cloud providers, has developed new technology to enhance semiconductor speeds amid escalating demand for generative artificial intelligence infrastructure.
Palo Alto, California-based Broadcom is among the biggest beneficiaries of hefty demand for AI-supporting hardware, as so-called hyperscalers turn to its custom chips to diversify their supply chains beyond Nvidia's costly processors.
The technology, called 3.5D XDSiP, will allow Broadcom's custom-chip customers to boost the amount of memory inside each packaged chip and speed up its performance by directly connecting critical components.
To do this, it uses manufacturing processes from TSMC, the world's largest contract-chip producer, including chip-on-wafer-on-substrate techniques - also known as advanced packaging - capacity for which is limited, making it a key bottleneck in the supply chain for AI chips.
There are five products in development that use the new tech, with production shipments starting in February 2026, Broadcom said.
While Broadcom does not identify the cloud companies it is developing custom chips for, analysts widely believe tech giant Alphabet's Google and Facebook-owner Meta Platforms are among its customers.
"Our hyperscale customers continue to scale up and scale out their AI clusters," Broadcom CEO Hock Tan said in September, when the company raised its forecast for AI revenue to $12 billion for fiscal year 2024, up from an earlier forecast of $11 billion.
The chipmaker, which also provides networking equipment for data centers, has three main customers for its custom processor unit, Tan said in September.
Broadcom's central competitor in this space is Marvell. The total market for custom chips could grow to about $45 billion by 2028 and be split between the two companies, Marvell COO Chris Koopmans said on Tuesday.
The market for custom chips will expand further, Summit Insights senior analyst Kinngai Chan said, adding that Marvell and Broadcom would benefit from this trend.
(Reporting by Arsheeya Bajwa in Bengaluru and Max Cherney in San Francisco; Editing by Pooja Desai)
