Enterprises Rate Non-Nvidia AI Chips Higher Than Nvidia

Among enterprise AI infrastructure leaders, there is growing momentum to evaluate chips other than Nvidia as next-generation candidates. According to a VentureBeat survey conducted in July 2025 (170 AI infrastructure leaders), 39.4% reported plans to evaluate non-Nvidia AI accelerators within the next 12 months. Meanwhile, the percentage planning to assess next-generation Nvidia GPUs such as Nvidia's Blackwell (GB300) reached only 25.3%, a 14-percentage-point gap.

Non-Nvidia products cited as evaluation candidates included AWS Trainium, Google TPU, AMD Instinct, Intel Gaudi, and self-developed ASICs (application-specific semiconductors), among others. While Nvidia continues to be the standard choice in production environments for many enterprises, the survey results indicate that companies are shifting strategy away from viewing Nvidia as the sole evaluation target, instead moving toward comparative assessment of multiple chips. Although Nvidia has maintained overwhelming market share in the AI accelerator market, cloud providers and semiconductor manufacturers have been steadily strengthening the performance and supply chains of their proprietary chips, supporting this trend.

Meanwhile, fewer enterprises are considering switching platforms in the short term. The share of respondents expecting to change platforms within three months declined from 38.3% in June to 28.8% in July. This indicates that enterprises are prioritizing increased utilization of existing infrastructure over large-scale platform switches. Among enterprises operating GPUs in-house, the percentage with utilization rates of 50% or below dropped from 83% in June to 69% in July, while the percentage exceeding 50% rose from 13% to 23%.

Changes also appeared in production adoption of AI platforms. The percentage using Microsoft Azure in production environments increased from 29% in June to 47.1% in July, a gain of 18.1 percentage points. However, it should be noted that 57% of July respondents were large enterprises with over 1,000 employees (compared to 37% in June), indicating that differences in sample composition may have influenced the figures. Google Gemini maintained the highest usage rate in both months, rising from 41.1% in June to 47.6% in July, slightly exceeding Azure. OpenAI production usage increased from 40.2% to 49.4%, while Anthropic rose from 12.1% to 24.7%.

Changes are also evident in the metrics used to measure infrastructure effectiveness. The percentage selecting "utilization rate and reliability" as important metrics increased from 42.1% to 51.2%, and the percentage choosing "throughput (processing capability)" also rose from 21.5% to 24.7%. Average ratings for ease of deployment improved on a 5-point scale from 3.84 to 4.04, and overall satisfaction increased slightly from 4.07 to 4.14. Cost-performance ratings, however, remained relatively flat at approximately 3.9.

What this survey reveals is that enterprises are beginning to adopt a more practical and business-oriented perspective toward AI infrastructure. The growing emphasis on utilization rates and reliability reflects a shift in viewing AI—from "something used experimentally" to "something operated stably as part of business operations." Regarding chip selection, enterprises are in a phase where multi-vendor strategy is becoming established, reducing dependence on specific vendors while securing multiple options. Going forward, actual operating costs and reliability assessments of platforms like Azure and Gemini may become decisive factors determining adoption choices.