GPU Platform Hub · As needed report

Investment Proposal

Generated July 26, 2026

Sample fleet

All figures derive from an illustrative fleet + the real GPU catalog + illustrative pricing. Connect an inventory to see live numbers. Catalog →

GPU capacity

1,048

9 clusters · 3 regions

Utilization

65%

30-day weighted

Idle capacity

35%

reclaimable

Est. monthly

$1.67M

cloud-equivalent

Est. annual

$20.06M

run rate

Refresh share

18%

prior-gen / legacy

Cost by clusterEstimate
ClusterGPUCountUtilEst./mo
prod-train-usNVIDIA25682%$710K
prod-train-euNVIDIA12877%$355K
prod-train-us2AMD6468%$164K
prod-inf-usNVIDIA19264%$154K
rag-usNVIDIA6471%$84K
prod-inf-euNVIDIA9658%$77K
legacy-usNVIDIA8037%$76K
vision-usNVIDIA12054%$31K
dev-usNVIDIA4823%$21K
Total (cloud-equivalent)$1.67M/mo
Capacity forecast+35% / yr
1,048
Now
1,130
Q+1
1,218
Q+2
1,313
Q+3
1,415
Q+4

At the assumed growth rate, the estate reaches ~1,415 GPUs in a year (+367). Start procurement ahead of lead times.

Technology lifecycle
Current472 GPUs · 45%
Mature384 GPUs · 37%
Prior gen192 GPUs · 18%

18% of the estate is prior-gen or legacy — the refresh planning horizon.

Investment recommendations
  • Now

    Reclaim idle capacity

    Consolidate or downsize legacy-us, dev-us; apply partitioning/chargeback to lift utilization.

  • Next

    Fund a hardware refresh

    Migrate the 18% prior-gen/legacy share to current parts for better perf/W and support.

  • Later

    Diversify supply

    Qualify a second GPU vendor to reduce concentration and improve negotiating leverage.

Risk register
  • medium

    Underutilized capacity

    legacy-us, dev-us run below 40% — ~$64K/mo of reclaimable spend.

  • high

    Aging hardware

    18% of the estate is prior-gen or legacy — plan a refresh to avoid support and efficiency gaps.

  • medium

    Vendor concentration

    94% of GPUs are single-vendor — a supply or pricing shock has no cushion.

Figures are illustrative — derived from a sample fleet, the GPU catalog, and illustrative pricing.