Nvidia pushes deeper into custom AI systems, raising pressure on Broadcom

Nvidia is expanding beyond GPUs into memory, interconnects, racks and licensing so it can remain inside custom accelerator programs it does not design—an approach that analysts say puts new competitive pressure on Broadcom even as hyperscalers build their own chips. The strategy, described by industry investors as an intent to own more of the system, is now showing up in products such as NVHBM and NVLink Fusion rather than in a rival GPU alone.