Enterprise Buyers Put Non-Nvidia Chip Evaluations Ahead of Nvidia’s Next GPUs

Enterprise AI buyers are widening their chip searches without signaling a fast break from Nvidia. In a July survey, alternatives to Nvidia accelerators drew more near-term evaluation interest than Nvidia’s next-generation GPUs, while the share expecting an immediate platform change declined. The result is a market focused on keeping options open while using existing infrastructure more intensely.

An evaluation list, not a deployment count

Among 170 AI-infrastructure respondents, 39.4% said they were likely to evaluate non-Nvidia accelerators over the next 12 months. That compared with 25.3% for Nvidia Blackwell GB300 or other next-generation Nvidia GPUs, a 14-point gap. The alternatives included AWS Trainium, Google TPU, AMD Instinct, Intel Gaudi and in-house ASICs.

The figures measure intended evaluation, not purchased or deployed chips. VentureBeat characterized Nvidia as the default in most production environments, while describing the interest in alternatives as an effort to build options around availability, software stacks and workload placement.

Likely accelerator evaluations over the next 12 months

0139.4%

Non-Nvidia accelerators

The July survey included 170 AI-infrastructure respondents.

0225.3%

Next-generation Nvidia GPUs

The comparison covers likely evaluations, not installed production use.

The hardware question is tied to the surrounding stack

Buyers’ selection criteria suggest they are judging infrastructure against individual deployment needs. Integration with an existing cloud and data stack led in July at 40.0%, followed by performance at 35.3%. GPU access reached 23.5%, cost per million tokens 15.9%, and broad total cost of ownership 21.8%.

Those priorities suggest an alternative-chip evaluation reaches beyond processor specifications. Teams are also weighing fit with cloud and data systems, production performance and operating reliability. The share selecting uptime and reliability as an important effectiveness measure rose from 42.1% in June to 51.2% in July.

More infrastructure activity, less immediate change

The near-term platform-change clock moved in the opposite direction from the hardware search. The share expecting a change within zero to three months fell from 38.3% in June to 28.8% in July, while roughly 40% reported no planned change. Among organizations operating their own GPUs, the share running above 50% utilization rose from 13% to 23%.

Optionality extends beyond chips

• The share expecting to do more with specialized AI cloud providers, or neoclouds, rose from 33% to 38%; named-provider production use rose from 1.9% to 5.9%.
• Custom, self-managed open-source infrastructure production use rose from 3.7% to 12.9%. The survey definition included PyTorch, Triton, vLLM, Ray and Kubernetes.
• In a separate survey on systems connecting models to enterprise data and tools, 79.2% favored approaches retaining some control outside a single model provider’s stack.

A directional result, with a conversion test

Interest in non-Nvidia accelerators rose from 31.8% in June to 39.4% in July. But the survey waves were independent samples rather than a panel following the same organizations, July had a larger share of bigger companies, and no statistical-significance testing was applied. VentureBeat says the monthly movement should be treated as directional rather than proof of causation.

The unresolved question is whether evaluation interest becomes sustained production deployment, or chiefly gives buyers more leverage while Nvidia remains their operating standard.