Skip to the build
Cost lab

AI chatbot feature (LLM token costs)

StretchBudget $3,400 a month
  1. Step 1 of 3: Fit the load, done
  2. Step 2 of 3: 2Hit the budget
  3. Step 3 of 3: 3Defend your bill

Catalogue

Compute

$589.84 a month

What runs your code.

Kind
Server size
How you pay

Reserved: a 1-year commitment, cheaper, paid all month. Spot: spare capacity, can be taken back with 2 minutes notice.

How many servers you run
Scaling

Fixed: the same number all month. Autoscale: follows the load, billed on the average (plus slack).

Database

$4.69 a month

Where the data lives.

Kind
Capacity mode

On-demand: pay per request. Provisioned: pay for capacity sized to the peak, all month, at a lower price per request.

Logs and monitoring

$13.99 a month

Needs to be kept at least 30 days.

Retention (days)

AI model

$43,500 a month

Billed per million tokens, input and output.

Model for every question
Monthly bill
$44,126
₹41,91,974 at ₹95 per $
Budget
$3,400
₹3,23,000

Monthly bill $44,126 (budget $3,400) · meets peak ✓ · availability ✓

$40,726 over budget, requirements met.

Checks

Handles the peakpassed
At peak 6.85 requests/s reach your servers (15 ms of CPU each, run at 60% load) = 0.17 vCPU. That is 1 x m6i.2xlarge; you allow 2. DynamoDB takes the load either way (on-demand scales itself; provisioned is sized to the peak here).
Availability: 99.5% (a few hours of downtime a month is fine)passed
The brief allows a single zone; the money is better spent elsewhere.
Latency: p99 under 4000 mspassed
No cold starts on the request path, and reads come from memory where it matters.
Logs kept at least 30 dayspassed
Retention is 30 days.
The choices fit the workloadpassed
Nothing in the design contradicts the workload.
Within budget ($3,400 a month)failed
$40,726 over budget.

The bill, line by line

Compute

$589.84
  • m6i.2xlarge servers, on-demand$589.84
    2 x m6i.2xlarge x $0.404/h x 730 h = $589.84

Database

$4.69
  • DynamoDB, on-demand$4.69
    3M reads x $0.0000001425 + 6M writes x $0.00000071 = $4.69
    You pay per request: nothing to size, nothing wasted at 4 am.

Monitoring

$13.99
  • Logs ingested$13.40
    20 GB x $0.67 = $13.40
  • Logs stored (30 days)$0.59
    20 GB x 30 / 30.4 days = 19.7 GB x $0.03 = $0.59

Network

$17.52
  • Load balancer$17.52
    730 h x ($0.0239 + 0.01 LCU x $0.008) = $17.52
    LCU: the larger of new connections (25 a second) or 1 GB an hour.

AI model

$43,500
  • Claude Sonnet 5.5: every question$43,500
    16.5B input x $2/M; 1.1B output x $10/M = $43,500

Mumbai (ap-south-1), 730 hours a month. List prices checked 2026-10-19: the price table below has the sources and what is left out.

Where the prices come from