Enterprise customers are actively evaluating alternatives to Nvidia's next-generation GPUs at a rate 14 percentage points higher than interest in Nvidia's Blackwell architecture, according to new survey data. This shift signals a potential crack in Nvidia's stranglehold on the AI accelerator market, even as the company maintains dominance in production deployments.
VentureBeat's July survey of 170 AI infrastructure decision-makers found that 39.4% plan to evaluate non-Nvidia accelerators over the next 12 months. These alternatives include AWS Trainium chips, Google TPUs, AMD Instinct processors, Intel Gaudi accelerators, and custom in-house ASICs. By contrast, only 25.3% expressed plans to evaluate Nvidia's Blackwell (GB300) or other next-generation Nvidia GPUs during the same period.
The gap reveals a decisive preference among enterprise buyers to diversify their compute strategies. This doesn't mean Nvidia loses its market position tomorrow. Nvidia's CUDA ecosystem, software maturity, and installed base remain formidable moats. Production environments still run predominantly on Nvidia hardware. But the evaluation phase is where architectural decisions get made. When enterprises compare options for their next major infrastructure refresh, they're increasingly including non-Nvidia options in the conversation.
Several factors drive this shift. Cost concerns rank high. Nvidia's premium pricing for Blackwell and H100 GPUs has prompted buyers to scrutinize total cost of ownership. Custom silicon from cloud providers like AWS and Google delivers performance tailored to their own workloads at lower per-unit costs. AMD's Instinct lineup offers comparable performance at undercut prices. Intel's Gaudi chips target inference workloads where Nvidia's dominance is less complete. Some large enterprises now build internal ASICs when volumes justify the engineering investment.
Supply constraints matter too. Nvidia's production bottlenecks during the past two years left customers waiting months for orders. Organizations dependent on consistent capacity can't afford to be locked into a single supplier. Evaluating alternatives becomes strategic hedge against future scarcity.
Software improvements for competing platforms reduce switching costs. AMD has strengthened ROCm, its GPU computing framework. Google and AWS continue optimizing their silicon stacks. Intel invests in OneAPI and Gaudi software. These ecosystems remain behind CUDA in maturity, but gaps narrow each quarter. Enterprise teams increasingly ask whether the CUDA premium justifies the cost differential when alternatives deliver 80-90% of the performance at half the price.
The survey also hints at organizational dynamics. Procurement teams gain leverage when they can pit vendors against each other. Engineering leaders gain credibility by demonstrating due diligence across multiple options. In practice, many enterprises will still deploy Nvidia hardware after evaluation. But the fact that 39% plan to seriously look elsewhere changes negotiating positions and forces Nvidia to defend its market share actively.
This doesn't suggest imminent collapse of Nvidia's market dominance. The company's lead in production workloads and software ecosystem remains substantial. But the evaluation data points to a market maturing beyond single-vendor dependency. Enterprises now possess genuine leverage. Nvidia must justify premium pricing through real performance advantages or risk losing share to cheaper alternatives that deliver adequate results.