Enterprise Buyers Favor Non-Nvidia AI Chips
VentureBeat’s latest AI infrastructure survey shows enterprises are more likely to evaluate non-Nvidia accelerators than Nvidia’s next-generation GPUs, even as production use and utilization rise.
Enterprise AI buyers are widening their accelerator shortlist, and non-Nvidia chips are now more likely to appear on it than Nvidia’s next-generation GPUs. That is the clearest signal from VentureBeat’s July VB Pulse survey of 170 AI infrastructure respondents, which suggests enterprises are treating accelerator selection as a broader strategic decision rather than a default Nvidia-only process.
The survey found that 39.4% of respondents said they are likely to evaluate non-Nvidia accelerators over the next 12 months. Those options include AWS Trainium, Google TPU, AMD Instinct, Intel Gaudi, and in-house ASICs. By comparison, 25.3% said they are likely to evaluate Nvidia Blackwell (GB300) or other next-generation Nvidia GPUs. That 14-point gap matters because it shows evaluation interest is spreading beyond the market leader, even though Nvidia remains the dominant choice in production environments.
Non-Nvidia Chips Gain Ground In Evaluation Lists
The key takeaway is not that enterprises are abandoning Nvidia. It is that more buyers are building optionality into their plans. For AI infrastructure teams, that means future procurement cycles may involve formal comparisons across several accelerator families instead of a simple upgrade path within Nvidia’s lineup.
This shift matters for several reasons. First, it gives cloud providers, chip designers, and internal silicon teams a larger opening to compete for enterprise workloads. Second, it suggests buyers are increasingly willing to match accelerator choice to workload needs, supply constraints, or infrastructure strategy. Third, it could gradually reduce the assumption that each new AI deployment will automatically center on Nvidia hardware.
The survey also points to a more cautious pace of platform change. The share of respondents expecting a platform change within three months fell from 38.3% in June to 28.8% in July, even as other infrastructure activity increased. In other words, enterprises are becoming more active in AI infrastructure without necessarily becoming more eager to replace the platforms they already use.
Production Use Continues To Broaden
July’s data shows that enterprises are not standing still. They are putting more platforms into production and running existing infrastructure more intensively. Microsoft Azure posted the largest production adoption gain among the measured platforms, with reported production use rising from 29% in June to 47.1% in July.
That increase should be read with some caution, since July’s respondent base skewed larger than June’s. Still, the broader pattern is clear: Azure adoption rises with company size, and the survey’s more up-market mix likely amplified the change.
Other platforms also posted gains. Google’s Gemini remained the most-used platform in both survey waves, rising from 41.1% in June to 47.6% in July. OpenAI rose from 40.2% to 49.4%, while Anthropic climbed from 12.1% to 24.7%. The direction of travel is toward wider production use across multiple AI providers, not consolidation around a single stack.
For enterprises, that may mean more complex infrastructure planning. Teams choosing models, cloud services, and accelerators may now need to optimize across cost, performance, reliability, and vendor diversity at the same time.
GPU Utilization Is Improving, But Capacity Still Matters
The survey also suggests enterprises are getting more out of the GPUs they already own. Among respondents operating their own GPUs, the share running at half capacity or less fell from 83% in June to 69% in July. At the same time, the share running above 50% utilization increased from 13% to 23%.
That trend is important because it shows infrastructure maturity is not just about buying more hardware. It is also about improving how efficiently current systems are used. Better utilization can delay new purchases, change procurement timing, and influence which accelerator architectures look attractive for the next round of investment.
What readers should watch next is whether the growing interest in non-Nvidia chips turns into more formal enterprise trials and purchasing decisions. The survey suggests buyers are already broadening their evaluation lists, but Nvidia still has the advantage in production. The next indicator will be whether optionality in evaluation starts to show up as meaningful share shifts in live deployments.
For now, the message is measured but significant: enterprises are not racing to replace existing AI infrastructure, but they are becoming more selective about what they test next. And in that testing phase, non-Nvidia chips are gaining visibility fast.