excel.aws.monce.ai  /  economics

Economics

Charles Dana · Monce SAS · May 2026

Token equation, AWS pass-through, scaling math.

1. The token equation

One token equals $0.00001 of underlying AWS cost (compute + storage + requests). Sale price: $0.00005 per token. Flat 5× markup. Bulk pack: $50 buys 1,000,000 tokens.

tokens(call) = ⌈ AWS_cost(call) / $0.00001 ⌉

The ceil() is intentional — we round up so the platform never under-charges on fractional tokens. On large jobs the rounding effect is negligible; on micro-calls (like a single inference cell) it's the only enforcement of a per-call floor.

2. Where the dollars come from

# Authoritative constants. eu-west-3, ARM64 Graviton2, May 2026.
GB_SEC     = 0.0000166667     # Lambda 10GB-class compute, $/GB·s
INVOKE     = 0.0000002        # Lambda request, $/invoke
S3_GBMO    = 0.023             # S3 standard, $/GB·month
S3_PUT     = 0.000005         # S3 PUT, $/op
S3_GET     = 0.0000004        # S3 GET, $/op
DDB_RW     = 0.00000125       # DynamoDB on-demand write, $/op

def aws_cost(call):
    return (
        call.lambda_gb_seconds * GB_SEC
      + call.lambda_invokes    * INVOKE
      + call.s3_mb_month * 1e-3 * S3_GBMO
      + call.s3_puts           * S3_PUT
      + call.s3_gets           * S3_GET
      + call.ddb_writes        * DDB_RW
    )

3. Measured costs (May 2026)

Wall-clock and AWS cost from =POTENTIAL on real data, hitting the production endpoint. The dispatcher routes locally below 500 rows (zero Lambda fee) and to snakebatch's v6 workers above. Cloud columns price the Lambda invocations at AWS list rate; local rows are bundled into the EC2 fixed cost (already in §7).

4. The 150,000-token starter pack

Every new sign-up gets 150,000 tokens on first verification. That's $1.50 of AWS cost, sells for $7.50 face value. Concretely it covers:

Generous enough for a serious evaluation, finite enough that we know who's serious. Starter pack is one-time; refills are paid (or credit-granted for design partners).

5. Storage is permanent — and trivially priced

Every trained model is retained in S3 indefinitely. We do not garbage-collect. The user can list, rehydrate, or explicitly delete any model. See architecture §7 for the mechanism; the cost shape is:

storage_tokens_per_day = ⌈ total_MB × (0.023 / 30) / 0.00001 ⌉

Storage is so cheap relative to compute that it's effectively free for any realistic deployment. The reason it's metered at all is to keep the books honest — not to discourage retention.

6. Three billing tiers

7. Fixed costs are tiny

8. A worked example: design partner with one workbook

cdana@monce.ai signs up. Receives 150K starter tokens. Over month 1:

• Sweeps =POTENTIAL.SCAN across a 50K-row factory dataset — ~10K tokens.
• Trains a Délai classifier on 15K rows, 15 layers — ~5K tokens.
• Predicts Délai on 800 upcoming orders — ~1K tokens.
• Audits 30 specific predictions for QA — ~30 tokens (embedded model = $0 inference, but each =AUDIT hits cloud once for the lookalike fetch).
• Stores 4 trained models all month — ~4 tokens.
• Total: ~16,000 of 150,000 starter tokens. 90% of the pack still available.

The design-partner experience converts to standard tier when the pack runs out — or when the partner asks to.

9. The line

"Five times AWS cost — rounded up — on every cell that fires.
Predictable, auditable, and stamped on every workbook."