Thursday, August 27, 2026
Tech Beat
Aug 12, 2026, 7:30 AMEnterprise AI Infrastructure

Enterprise AI Compute Surges as Cost Tracking and GPU Utilization Lag

VentureBeat finds 66% of enterprises run AI in production, but 53% lack rigorous cost tracking and 69% keep GPU utilization at 50% or less across 170 firms.

A GPU-powered rocket speeds ahead with an opaque fuel gauge, symbolizing rapid AI growth without cost visibility.
Listen to this briefingAudio briefing

Summary

VentureBeat's August 12, 2026 report draws on a self-selected July 2026 Pulse survey of 170 organizations with more than 100 employees. It found 66% run AI in production, 29% at scale, 30% remain in proofs of concept and 4% have not started. The directional, single-wave sample, 57% from companies above 1,000 employees, averages three platforms: OpenAI 49%, Google Gemini 48%, Microsoft Azure 47% and Google Cloud 42%. Azure leads primary use at 26%, followed by Google Cloud at 19%, OpenAI and Gemini at 14% each, and Anthropic at 8%. Custom open-source stacks reach 13%, owned GPU clusters 9%, CoreWeave and Lambda 3.5% each, and specialized clouds only 1% as primary.

Selection prioritizes existing-stack integration at 40%, performance at 35% and GPU availability at 24%, above total cost of ownership at 22%. Success measures likewise favor uptime and reliability at 51% and developer productivity and deployment speed at 39%, over cost per million tokens at 31%. Among 155 GPU operators, 69% use 50% of capacity or less, 23% exceed 50% and 12% do not measure utilization. Only 47% rigorously track compute cost and return, rising to 56% among at-scale operators. Satisfaction averages 4.14 overall, 4.04 for implementation and 3.87 for value on a five-point scale.

Within 12 months, 62% plan to add or switch providers, including 29% next quarter; 44% will evaluate AI-specialized clouds, which have +36 net momentum, while 39% target non-Nvidia accelerators. Near-term consideration favors OpenAI and Google Cloud at 29% each and Azure at 28%; CoreWeave draws 4% and Lambda 3.5%. For inference's KV-cache memory constraint, Dell leads at 24%, Nvidia 21%, open source 12% and model efficiencies 11%, while 19% do not recognize or address it.

Positives

  • 66% of surveyed enterprises have AI workloads in production, including 29% operating at scale.
  • 23% of the 155 enterprises operating GPUs report utilization above 50%.
  • Overall infrastructure satisfaction averages 4.14 out of five, while ease of implementation scores 4.04.
  • 42% expect to increase specialized-cloud use, compared with 6% expecting to reduce it.
  • 62% intend to switch or add an infrastructure provider within 12 months, signaling active competition for enterprise workloads.

Risks & concerns

  • 53% cannot rigorously track AI compute costs and returns, including 44% of enterprises operating AI at scale.
  • 69% of GPU operators use half their capacity or less, while 12% do not measure utilization.
  • Value for money scores 3.87 out of five, below overall satisfaction of 4.14.
  • AI-specialized clouds attract 44% evaluation interest, but CoreWeave and Lambda each appear in only 3.5% of current stacks.
  • 19% of enterprises either do not recognize or have not addressed the emerging KV-cache memory constraint.
Primary sourceVentureBeathttps://venturebeat.com/resources/infrastructure-and-compute-enterprises-are-buying-ai-compute-for-speed-while-flying-blind-on-what-it-costs
Read full article
Editorial note: Tech Beat summarizes and analyzes third-party reporting. The source link is the authoritative article. This page does not reproduce the full source text.

More From The Wire

A sliding puzzle locks into a tight grid after one piece changes order, symbolizing smarter GPU scheduling. Enterprise AI InfrastructureAug 17

Dharma-AI GPU Allocator Boosts Utilization 33 Points Over FIFO

CybersecurityAug 27

Visa VVAH AI Patches Code Before Human Review

Artificial IntelligenceAug 27

OpenAI Brings ChatGPT Ads to India With 50 Brands, ₹725 Daily Floor