GPU Utilization

Policy Regula

2026-10-26 06:06:11

Survey finds 79% of companies leave GPUs idle at least half the time, while 70% still plan new AI platforms

A new set of enterprise AI infrastructure survey results compiled by VentureBeat Intelligence points to a striking mismatch inside corporate AI spending. Across three VB Pulse surveys conducted in July, August, and September 2026, 79% of respondents said their GPU capacity sits completely idle at least half the time. That figure stayed elevated throughout the quarter, ranging from 77% to 81%, suggesting the problem is persistent rather than temporary. Even so, 70% of companies whose average GPU utilization does not exceed 50% said they still plan to introduce, expand, or replace AI computing platforms over the next 12 months. The survey pool included 385 qualified respondents from organizations with at least 100 employees, including consultants answering on behalf of major clients. The data also showed weak financial discipline around new AI spending. In VentureBeat’s September survey, 58 of 64 respondents planning to adopt a new AI platform lacked ROI figures reviewed by finance departments, or 91%. The report argues that many enterprises are now paying two separate AI bills at once: underused in-house GPU clusters purchased during earlier supply fears, and rising API costs tied to frontier closed-source models from providers such as OpenAI, Anthropic, and Google. As AI agents spread, the report says inference costs, context handling, and multi-agent orchestration may become the next pressure point.

20

Delphi Digita

2026-09-03 19:12:02

Delphi Digital: China's AI Labs Drive Competition on Cost of Effective Intelligence

Delphi Digital posted on X that Chinese AI labs are forcing the market to compete on the cost of effective intelligence. Due to export controls limiting access to advanced Nvidia chips, labs must squeeze more performance from existing hardware, as exemplified by DeepSeek maintaining high GPU utilization during training.

1080