How we measure · Last updated Q1 2026
Every number on this site has a source.
We state what we measured, how we measured it, and where the data came from. If a stat requires a hedge, we include it. If it requires attribution, we give it.
GPU utilization
"Up to 92% GPU utilization"
Source
R3D Studios production environment. Measured as the average GPU utilization within active time-sliced sessions across their AWS g5.2xlarge fleet. This is a production average, not a peak reading.
How to use it
"Up to 92% GPU utilization" when citing as a general Orion capability. "92% production average" when attributed directly to R3D Studios with context. Never cite bare without a qualifier.
Industry baseline
"Enterprise on-premises GPU utilization sits at 10-15%." Source: Goldman Sachs and Cast AI 2025 data. Never cite as "20-50%", "less than 35%", or "below 40%" — those figures are not approved.
Last verified
Q1 2026. R3D production environment. Updated quarterly. † To cite: Juno Innovations Measurement Methodology, Q1 2026. juno-innovations.com/methodology#gpu-utilization
Compute cost reduction
~40% compute cost reduction
Source
R3D Studios. AWS billing comparison, April-September 2025 vs. October 2024-March 2025 (pre-Orion baseline). Compute-only line items: EC2, EBS, and GPU instances. No credits applied.
What changed
GPU time-slicing replaced 1:1 exclusive GPU assignment. Suspend-on-idle eliminated forgotten-GPU billing. Infrastructure headcount reduced from 5 to 2 FTE.
How to use it
"~40% compute cost reduction" when attributed to R3D Studios. This measures compute-only spend, not total infrastructure. Do not generalize as total cloud savings without attribution and context. † To cite: Juno Innovations Measurement Methodology, Q1 2026. juno-innovations.com/methodology#compute-cost
Related stat
"Up to 78% total cloud spend reduction" includes team reduction, eliminated idle instances, and overhead, not EC2 alone. This figure requires full context and R3D attribution. Not for standalone use in headlines.
Provisioning time
60-90 second provisioning
Source
R3D Studios production environment and Orion product spec. Measured from workload request submission to container ready state with a pre-pulled image. Cold-start (image pull) is 3-8 minutes depending on image size.
Comparison
Legacy manual provisioning at R3D: 24-48 hours per environment request. Orion cached image start: 60-90 seconds. This is the basis for the "was 24-48 hours" comparison used on the site. † To cite: Juno Innovations Measurement Methodology, Q1 2026. juno-innovations.com/methodology#provisioning-time
Workload capacity
Typically 2-4x more workload capacity
Source
R3D Studios production deployment: 5 g5.2xlarge instances supporting 10 concurrent artist workstations via GPU time-slicing, a 2:1 density ratio. Customer survey confirmed up to 4x concurrent users per GPU in specific workload profiles.
How to use it
Always include "typically." The range reflects workload variation. GPU-bound render jobs land closer to 2x, CPU-light interactive workloads can reach 4x. Never cite bare "4x" without the hedge. † To cite: Juno Innovations Measurement Methodology, Q1 2026. juno-innovations.com/methodology#workload-capacity
VMware pricing
150% to 10x+ post-Broadcom
Source
CloudBolt Survey, January 2026. Forrester analysis. Public AT&T and enterprise reporting. The range reflects real variance: smaller deployments see 150-300% increases; larger enterprise agreements with bundled ELAs have seen 5-10x+ increases.
How to use it
"150% to 10x+" is the only approved range. Never use "300-400%", "300-1,200%", or any other variation. vSphere 7 EOL: October 2, 2025 (past). vSphere 8 EOL: October 11, 2027. † To cite: Juno Innovations Measurement Methodology, Q1 2026. juno-innovations.com/methodology#vmware-pricing
See a number that needs updating?
We update these stats when customer data changes. If you have questions about methodology or want to verify a specific claim, reach out directly.
No long-term contract required · Deploy in your environment · Running in under two minutes

