NVIDIA Loses Ground With AI Engineers as Cooling and Power Costs Push Hyperscalers Toward Custom ASICs, Evercore Warns

Evercore says AI buyers are shifting from raw GPU performance to total cost of ownership, especially power, cooling, and cost per token.

As workloads move toward inference, hyperscalers are showing more interest in custom ASICs and other accelerators instead of mainly NVIDIA GPUs.

The trend could sharpen competition for NVIDIA by boosting cheaper homegrown chips from cloud providers and alternative AI hardware vendors.