Four engagements. No percentage of savings, ever.
Contingency pricing gives the person measuring your infrastructure a financial stake in the size of the number, which is the one incentive you do not want in the room. It also turns every invoice into an argument about what your spend would have done anyway. Every number below is fixed and agreed in the SOW before anyone touches anything.
Spend Snapshot
No charge. Forty-five minutes. One call.
kubectl against one clusterThree numbers, produced live on the call:
- Amortized GPU spend for the last full month, per resource.
- What share of it carries no owner tag.
- How many provisioned GPUs have zero pods requesting
nvidia.com/gpu.
The third number is usually the one that lands, and it needs no new instrumentation at all. If kubectl describe node shows 0 of 8 GPUs allocated, utilization is zero by definition and no telemetry can argue with it.
The read-only Kubernetes RBAC you would need to grant ships with the tooling, in a minimal and a standard version, alongside a plain shell script your own engineer can run instead if you would rather not grant anything at all.
Hard scope limit: three numbers and a conversation. No document, no recommendations, no remediation. If you want any of that, it is Tier 2. This is not a trial that expands; it is forty-five minutes that tells us both whether the rest is worth doing.
Cost Attribution Report
$18,000 to $25,000 fixed. Two weeks.
- Amortized cost per GPU resource, per team, per namespace, per model.
- Ranked recoverable spend, each line carrying a dollar figure and a confidence rating.
- Evidence gaps stated explicitly: what cannot be claimed yet and exactly which data would close each one.
- A remediation sequence with effort and risk, ordered by dollars per unit of disruption.
If you have no GPU telemetry today, the report includes the instrumentation plan to get it, specified to the exporter version and the scrape config.
The confidence framework
Every finding is marked, and the marking is enforced in the tooling rather than promised in a footnote.
Unproven dollars are excluded from the recoverable total. That exclusion lives in code because under client pressure discipline is the thing that bends. The competing document quotes one confident figure that does not survive its first contact with a skeptical staff engineer. Ours is lower, and it holds up in the room.
Excluded: any change to your infrastructure. This tier produces a document. Nothing is touched.
Read one before you buy one
A complete specimen report, generated by the same tooling from a synthetic Cost and Usage Report. Nine findings across all three confidence ratings, the cluster allocation section, the order of work, and the method note. The fictional client is labelled as such on the first line.
Recovery Sprint
$45,000 to $70,000 fixed. Six to eight weeks.
Typical work:
- GPU telemetry deployed: DCGM exporter, Prometheus, per-workload attribution that survives a pod restart.
- Requests right-sized against measured p95 plus headroom, not against what someone guessed at deployment.
- Time-slicing or MIG partitioning where workloads are genuinely small enough to share a card.
- Bin-packing and autoscaler tuning so scale-in actually happens instead of being blocked by one unevictable pod.
Plus spot migration for interruption-tolerant training, and idle reclamation policies with owner notification so nothing disappears without a human being told.
Excluded unless separately scoped: model or training-code changes, application refactoring, migration between clouds, and anything touching production traffic routing.
Without a named owner, change-window access and merge authority, this becomes a consulting report with extra steps. We will say so before starting rather than after.
Cost Watch
$12,000 to $15,000 per quarter. Two-quarter minimum, 30-day termination.
Right-sizing that runs once at deployment is not right-sizing. Workloads change, traffic shifts, and a configuration that was accurate six months ago almost certainly is not now.
- Quarterly attribution report and drift analysis against the same baseline.
- Alerting on new unowned GPU spend and on allocation without utilization.
- Commitment review: Savings Plans, reservations, capacity blocks.
- A standing engineering channel and a quarterly review with your leadership.
Quarterly rather than monthly is deliberate. It matches how fast the underlying problem actually drifts, and at roughly 8 to 10 percent of a qualified client's annual GPU spend it stays defensible against savings already delivered.
Excluded: on-call, incident response, anything carrying a response-time commitment. This is advisory with artifacts, not managed services.
Where you land in a range is decided before work starts, not after.
Four inputs, all of which we establish on the qualifying call:
- Linked accounts. One payer with one account is not six accounts with inconsistent tagging.
- Billing readiness. Whether a CUR 2.0 export at hourly granularity with resource IDs already exists, or has to be stood up and backfilled.
- Cluster count. Each additional cluster is another allocation join and another set of provider IDs to reconcile.
- Telemetry. Whether any GPU profiling metrics are running today, or whether the instrument has to be built before it can be read.
A single account with billing already exported and telemetry running sits at the bottom of the band. Six accounts, three clusters and no telemetry sits at the top. You will know which before you sign anything.
Terms
- Fixed fee, invoiced 50 percent at kickoff and 50 percent on delivery.
- Mutual NDA before any access is granted.
- Read-only access, scoped and time-limited, revoked at the end of the engagement.
- Your data stays yours. Nothing is retained after close-out except the deliverable itself.