Among Respondents Reporting GPU Use, 79% Say It Was 50% or Less

Enterprise respondents who reported using graphics processing units (GPUs) said the processors were busy half the time or less in 79% of cases, while platform plans were similar among respondents reporting higher use. The comparison does not establish that lower use prompted those plans.

The pooled results combine three monthly surveys conducted from July through September 2026. They included 385 qualified responses, with 218 respondents providing GPU-use figures. Of those 218, 172 (79%) reported use at 50% or less. Monthly shares ranged from 77% to 81%, with no statistically significant difference among the months.

In the lower-use group, 120 of 172 respondents (70%) planned to adopt, add or replace an AI compute platform within a year. Among the 46 respondents reporting use above 50%, 34 (74%) planned one of those actions. The difference between the groups was statistically insignificant.

The platform question grouped plans to adopt a new platform, add one or replace one. It did not measure whether those plans would increase total computing capacity, so the responses do not show that respondents expected to buy more GPUs.

Cloud or model providers appeared among the options selected by 108 of those 120 respondents (90%). Nine (8%) named an on-premises or colocated GPU cluster or a self-managed open-source stack. Respondents could select multiple options.

Participants came from organizations with at least 100 employees or were consultants answering for their main clients at qualifying organizations. The survey drew self-selected participants, and each monthly wave included a different group. Its results describe the respondents, not all enterprises.

In September, 58 of 64 respondents (91%) who planned to adopt, add or replace a platform had no finance-reviewed return-on-investment figure for their AI infrastructure. Across all 102 qualified respondents that month, 10% had such a figure.

That survey also found that 55 of 68 production-AI respondents who provided a GPU-use figure (81%) reported use at 50% or less. This is a separate measure from the pooled result and matches the September-only 81% finding.