Microsoft isn’t the only one making waves. AWS has already established itself with custom chips such as Trainium for machine learning training, Inferentia for inference workloads, and its Nitro system for advanced virtualization and security. Google brought its Tensor processing units (TPUs) to market years ago as a custom solution for machine learning tasks. Combined, these innovations are not just filling the gap left by conventional GPUs, they’re redefining how we think about workloads at scale.
Other industry players are following suit. Nvidia, best known for its GPUs, has introduced Bluefield chips, and AMD is leaning into its Pensando portfolio. The result is an ecosystem increasingly reliant on custom accelerators tuned to specific tasks—a far cry from the days when a handful of brands dominated the chip market.
Fixing GPU shortages and more
The overarching motivation for investing in custom silicon is clear: Traditional GPUs, while powerful, are often too power-hungry, expensive, and general-purpose to handle the nuanced demands of modern cloud computing. With more demand than supply of Nvidia’s GPUs, for instance, alternative solutions like custom-designed chips offer more control over price-performance ratios, energy efficiency, and cooling requirements.
