Latest

AI Drives Shift to Universal Entertainment Apps Across Streaming PlatformsGoogle Launches New Gemini AI Models Amid Ongoing Absence of 3.5 ProGoogle Ships Three New Gemini Models — But Skips 3.5 ProUS Threatens Sanctions on Chinese AI Models Over Alleged IP TheftOpenAI Says Hugging Face Breach Traces Back to Its Own Pre-Release ModelsAI Data Centers Set to Consume 4× More Electricity by 2035Inference Startup Infinity Lands $15M From OpenAI and Anthropic ResearchersEnterprises Are Buying AI Compute Faster Than They Can Measure It54% of Enterprises Have Already Had an AI Agent Security IncidentOpenAI Is Scared of Open-Weight Models. Should the US Be?Anthropic's Landmark $1.5B Copyright Settlement Wins Final ApprovalEU AI Act Enters Second Wave of Enforcement ActionsWhat to Watch After Jensen Huang's Japan VisitCan an Apple Lawsuit Derail OpenAI's Hardware Plans?Meta's Llama 5 Release Reportedly ImminentAI Drives Shift to Universal Entertainment Apps Across Streaming PlatformsGoogle Launches New Gemini AI Models Amid Ongoing Absence of 3.5 ProGoogle Ships Three New Gemini Models — But Skips 3.5 ProUS Threatens Sanctions on Chinese AI Models Over Alleged IP TheftOpenAI Says Hugging Face Breach Traces Back to Its Own Pre-Release ModelsAI Data Centers Set to Consume 4× More Electricity by 2035Inference Startup Infinity Lands $15M From OpenAI and Anthropic ResearchersEnterprises Are Buying AI Compute Faster Than They Can Measure It54% of Enterprises Have Already Had an AI Agent Security IncidentOpenAI Is Scared of Open-Weight Models. Should the US Be?Anthropic's Landmark $1.5B Copyright Settlement Wins Final ApprovalEU AI Act Enters Second Wave of Enforcement ActionsWhat to Watch After Jensen Huang's Japan VisitCan an Apple Lawsuit Derail OpenAI's Hardware Plans?Meta's Llama 5 Release Reportedly Imminent

Models·July 19, 2026·1 min read

Enterprises Are Buying AI Compute Faster Than They Can Measure It

New research finds 83% of enterprises run GPUs at under 50% utilization, and fewer than half can accurately track their AI compute spend.

W

We Read AI

We Read AI

A new enterprise AI infrastructure report finds a widening gap between AI spend velocity and financial oversight. 83% of surveyed organizations operate their GPU fleets at 50% utilization or below, and fewer than half can rigorously track what those clusters actually cost per workload.

The scale of the waste

At current H100 and B200 pricing, a fleet running at 40% utilization is effectively burning six-figure monthly line items on idle silicon. Multiply that across the Fortune 500 and the aggregate waste dwarfs most enterprise SaaS categories.

Why the meters are broken

Cost visibility is hard because production AI workloads span reserved capacity, spot markets, provider credits, and increasingly, colocated on-prem gear. Existing FinOps tools weren't built for that mix, and the vendor-side "unit economics" dashboards remain shallow.

What's shipping

A wave of AI-specific FinOps startups (Robusta, MosaicML's new spinoff, and Anthropic's own admin console updates) are trying to close the gap. Adoption is early but growing fast — for the same reason cloud FinOps eventually became table stakes.

Source: VentureBeat

#gpu#enterprise#cost#infrastructure

ModelsJuly 21, 2026

AI Drives Shift to Universal Entertainment Apps Across Streaming Platforms

AI is blurring the lines between content formats, pushing streaming giants to evolve into comprehensive entertainment hubs.

ModelsJuly 21, 2026

Google Launches New Gemini AI Models Amid Ongoing Absence of 3.5 Pro

Google's latest Gemini models raise strategic questions as the absence of the anticipated 3.5 Pro model continues to unsettle the AI landscape.

ModelsJuly 21, 2026

Google Ships Three New Gemini Models — But Skips 3.5 Pro

Google's latest release brings Gemini 3.6 Flash, 3.5 Flash-Lite, and a Flash Cyber variant, but the absence of a mid-tier Pro model raises questions about the company's tier strategy.

ModelsJuly 21, 2026

US Threatens Sanctions on Chinese AI Models Over Alleged IP Theft

Treasury Secretary Scott Bessent signaled potential sanctions targeting Chinese open-source models, escalating the Trump administration's tech-decoupling push.