Quarterly · deliverable
Token Economics & Topology Audit
Cross-border optimization study with token-leak identification across every gateway, model, and modality in your stack.
- Routing topology map
- Provider arbitrage windows
- Leak & retry forensics
Token Fiscal Governance · 2026
Compass Advisors:
Navigating the Financial Architecture of AI.
Establish auditable cost controls, dynamic token routing, and fiscal governance across your multi-model AI infrastructure.
37%
avg. token spend reduction
<48h
to first audit signal
$1.2B
AI spend under governance
Architected for the modern AI stack
Kong
Shakudo
Portkey
Bifrost
LiteLLM
/ methodology
The hidden P&L of multi-model consumption.
Problem · Value erosion
Unoptimized consumption is silently compounding into your COGS.
Frontier-model spend is sprawling across 3–7 providers. Without a fiscal operating policy, gateway routing rules default to the most expensive path, KV caches go unreused, retries blow past budgets, and finance has zero visibility until the invoice arrives.
- Token leak across redundant retrieval & retry loops
- Cross-border egress + provider arbitrage left on the table
- Premium-tier model usage for commodity workloads
- No unit-economics tie-back to product or customer
Solution · Policy engine
A financial operating policy that runs at the gateway.
We codify your unit economics into your gateway — Kong, Portkey, Bifrost, LiteLLM, Shakudo — so every token is routed to the model, region, and tier with the best dollar-per-quality outcome. Auditable, reversible, finance-grade.
- Dynamic routing tied to live unit-economics signals
- Cost-per-customer attribution to your data warehouse
- Cross-region failover that respects egress economics
- Quarterly-reviewed governance with finance + platform
/ token economics calculator
Model the real cost of a token —
API providers vs. your own silicon.
Most TCO models miss the unsexy 40%: facility PUE, fabric capex, KV-cache HBM headroom, on-call SREs, depreciation, WACC, and gateway routing efficiency. Model your workload below.
01
users × per-user activityWorkload
Application preset
50 users × 20 req/day = 1,000 req/day│1.13B tokens/yr
02
public pricingAPI provider
03
open-weights frontier classOn-prem model
Recommended silicon
NVIDIA B300 288GB · 8× per replica
Weights ≈ 355GB · KV/activation headroom 80GB → 435GB HBM
04
time value of moneyCapEx & silicon depreciation
05
PUE, kWh, sqftFacilities & cooling
06
people, software, riskOpEx & ecosystem
07
the idle penaltyWorkload dynamics
Effective throughput ≈ 744 tok·s⁻¹ / GPU│726.5B output tok/yr cluster-wide
Annual TCO comparison
On-prem wins · Δ $4,796
Anthropic API
$6,023
$5.32 / 1M tokens
On-prem (allocated)
$1,226
$5.60 / 1M tokens (fully loaded)
API$6,023
On-prem$1,226
On-prem cost composition
Total $4,067,607 / yrSilicon depreciation
56.6%
$2,300,416
Cost of capital (WACC)
6.7%
$273,174
Power + cooling
3.3%
$132,875
Real estate
1.8%
$72,000
Ops engineering
21.6%
$880,000
Software + orchestration
5.5%
$224,000
Network + transit
2.4%
$96,000
Taxes + insurance
2.2%
$89,141
Total capex outlay
$5,751,040
compute $4,992,000 · fabric $544,000 · cooling $215,040
IT power draw
89.6 kW
1,060 MWh/yr @ PUE 1.35
Footprint
160 sqft
$72,000 / yr
Capacity utilized
0.0%
of 726.5B tok/yr
Break-even volume
764.22B tok/yr
at chosen API price
5-yr cumulative Δ
$23,982
savings on-prem
/ the offering
The Token Fiscal Governance Suite.
A quarterly subscription engagement. Three integrated workstreams, one fiscal policy engine, governed by your finance and platform leaders.
Engagement model
Quarterly · retained
Continuous · monitoring
Production Telemetry Review
We instrument your gateways and continuously optimize routing rules against your fiscal policy as model pricing and quality move.
- Cost-per-customer dashboards
- Routing-rule pull requests
- Anomaly & burn-rate alerting
Steering · committee
Dedicated Financial Engineering Advisory
A quarterly steering committee with your CFO, CTO, and our principal advisors to align AI unit economics with the P&L.
- Quarterly business review
- Capital-allocation framework
- Direct Slack + on-demand access
Proprietary metric
The Dynamic Token Efficiency Index (TEI).
The global standard for AI unit economics — a single auditable score that benchmarks your dollar-per-quality token against the market and your own historical baseline.
Sample score
87.4
TEI · Q2 2026
/ engage
Begin with a Token Economics Audit.
A two-week diagnostic. We instrument one gateway, baseline your TEI, and surface the three highest-leverage routing changes — before you sign anything else.
Request an Audit hello@compass-advisors.ai