Blog ▸ From Cost Centre to Revenue Line: How Enterprises Are Rethinking Owned GPU Infrastructure
GPU Infrastructure
A GB200 NVL72 rack can generate $18,000 to $28,000 a day in compute revenue, most of which enterprises never collect. Here is how leading organizations are rethinking owned GPU infrastructure.
From Cost Centre to Revenue Line: How Enterprises Are Rethinking Owned GPU Infrastructure
GPUaaS.com Team
GPU Infrastructure
July 28, 2026
The world's most wanted GPU, NVIDIA B200 bare metal DC in US West - live on packet.ai →→(Access it from Bare metal CTA on top after login)
A single GB200 NVL72 rack can generate $18,000 to $28,000 a day in on-demand compute revenue. Most enterprises that own one collect none of it. The rack sits, depreciating, drawing power, idle most of the day.
Standard accounting treats a GPU cluster as pure cost. Capex on the balance sheet. Depreciation running against it. Power and cooling on top.
Key takeaways
A GB200 NVL72 rack can generate $18,000 to $28,000/day in on-demand compute revenue, revenue most owners never collect
Average enterprise GPU utilization sits at 5% across 23,000 measured production clusters (Cast AI, 2026)
Two-thirds of enterprises report peak GPU utilization below 70%, meaning even at the busiest moment most fleets run well under provisioned capacity
Enterprises shifting from cost-centre to revenue-line thinking aren't buying fewer GPUs, they're offering unbooked windows to external buyers instead of leaving them idle
Procurement and runtime are the same problem from two ends. Overcommitment upfront creates the idle capacity downstream
◆ NEO-REAL ESTATE, NOT NEO-CLOUD
5% average utilization across 23,000 clusters
Cast AI's 2026 measurement across 23,000 production clusters found average enterprise GPU utilization at 5%. Cast AI's co-founder named what many fleets have become. Not cloud infrastructure. Neo-real estate.
◆ EVEN AT PEAK, MOST FLEETS ARE UNDERUSED
Two-thirds sit below 70% at their busiest moment
A separate survey found two-thirds of enterprises report peak GPU utilization below 70%. Not average. Peak. Even at the busiest moment of the day, most fleets run well under provisioned capacity.
◆ A LEASABLE ASSET, NOT A SUNK COST
What the shift actually looks like
The enterprises shifting this aren't buying fewer GPUs. They're treating owned hardware as a leasable asset. A rack not fully booked for internal work during a window gets offered to a vetted external buyer for that window. Capacity returns to internal use the moment it's needed.
$18K-$28K
the daily on-demand compute revenue a single GB200 NVL72 rack can generate, revenue most owners never collect
GPU rental rates on the exact hardware most enterprises already own run a few dollars an hour to well over ten, depending on chip generation and demand. Multiply that by the hours a rack sits idle in a typical month. The uncollected revenue on a rack at 5% utilization is not a rounding error.
The same scarcity pressure that drove overbuying is why nobody releases capacity back, even capacity sitting unused. Procurement and runtime get managed as two separate problems. They're one problem from two ends.
A rack stays available for internal use the moment it's needed. Still owned. Still on the balance sheet. What changes is whether the idle hours sit unused or get offered against real external demand.
See what your idle hours are worth right now.
Submit cluster availability, get matched with vetted buyers. For single-GPU or month-to-month demand, packet.ai handles self-serve access with 24/7 human support.
A single GB200 NVL72 rack can generate $18,000 to $28,000 a day in on-demand compute revenue. At 5% average utilization, the large majority of that daily figure goes uncollected on a rack that's fully owned and paid for, sitting idle rather than earning against its own capacity.
The same scarcity pressure that drives overbuying also discourages releasing unused capacity back to the market. Procurement decisions and runtime utilization tend to get managed as two separate problems by different teams, when they're actually one problem viewed from two different ends.
It means offering unbooked capacity windows to external buyers instead of leaving them idle, while keeping the hardware fully available for internal use whenever it's actually needed. The asset stays owned and on the balance sheet the same way it always was.
No. Availability windows are set by the owner, and capacity returns to internal use the moment it's needed. External use only applies to hours the owner has designated as unbooked for internal work.
Submit cluster availability, GPU tier, count, available hours, region, through GPUaaS.com, and get matched with vetted buyers looking for exactly this kind of capacity.
Last reviewed: 29 July 2026. Utilization data from Cast AI's 2026 State of Kubernetes Optimization Report, 23,000 clusters. Rack-level revenue data from ProphetStor's GPU Infrastructure Yield Layer analysis, 2026. Peak utilization survey data from the State of AI Infrastructure at Scale 2026 report. List your idle GPU cluster on GPUaaS.com.
