Engine 26 · PTU ROI Planner · world-first

Provisioned-Throughput ROI Planner

PTUs are the reserved instances of AI — and teams buy them blind, or stay pay-as-you-go paying 2× more. Nobody does the commitment math. We do, deterministically, with every assumption disclosed.

Should you commit? Run your numbers

ModelInput tokens / mo (millions)Output tokens / mo (millions)Peak-to-avg ratio

STAY PAYG

Stay pay-as-you-go: your 3.0× peak-to-average ratio means the reservation would sit ~67.3% idle. PTUs win on high, FLAT sustained traffic — smooth your peaks (batch/queue off-peak work) or grow volume, then re-run this.

PAYG / mo

$2,250

17 PTUs / mo

$4,420

Monthly delta

−$2,170

Utilization

32.7%

Every assumption, disclosed

usd_per_ptu_month = $260

tpm_per_ptu = 2,500

min_ptu = 15

input $2.5/1M · output $10/1M

List-price assumptions (mid-2026). Enterprise discounts of 20-40% are common — override any field.

Your live PTU recommendation (last 30 days)

Why this is the 2026 version of the RI optimizer

A decade ago, companies like ProsperOps built businesses on one insight: cloud commitments are mispriced by inattention. AI capacity is repeating the pattern — Azure OpenAI PTUs, Bedrock provisioned throughput, OpenAI scale tiers — except this time the swing is bigger, because inference demand is spikier. Under-commit and you pay 2× on-demand rates. Over-commit and you burn reservations at 30% utilization. The math above is deterministic, assumption-disclosed, and re-runs on your real traffic daily.

First-to-market: no FinOps or LLM-observability tool does AI capacity commitment math (Jun 2026).

Provisioned-Throughput ROI Planner

Should you commit? Run your numbers

Your live PTU recommendation (last 30 days)

Why this is the 2026 version of the RI optimizer

Install CARTIEAI