TCO Calculator

When I need to evaluate infrastructure options, I want to compare API and self-hosted costs with break-even analysis, so I can make data-driven decisions on when to migrate to self-hosted infrastructure.

The Challenge

Finance asks: “Should we keep using Claude API or deploy our own infrastructure?” The question seems straightforward, but the answer requires modeling a dozen interacting variables. API costs scale linearly with usage—$3 per million input tokens, $15 per million output tokens for Claude 3.5 Sonnet. Self-hosted costs are mostly fixed—$31.21 per hour for an A100 instance whether you’re running at 10% or 90% utilization.

The break-even point depends on your request volume, average token count, instance utilization, and hidden costs that rarely make it into back-of-envelope calculations. Network egress, storage, load balancers, monitoring, and ops time add 20-40% to raw GPU costs. A spot instance strategy might save 70% on compute but adds complexity and interruption risk.

Most teams either default to API (simpler, pay-per-use) or commit to self-hosting (feels cheaper at scale) without modeling the actual crossover. The wrong choice at 100K requests/month might cost $5K annually. The wrong choice at 10M requests/month could cost $500K.

How Lattice Helps

The TCO Calculator models the complete cost picture for both API and self-hosted options. Instead of comparing API pricing to GPU hourly rates, you compare fully-loaded monthly costs including infrastructure, operations, and hidden expenses.

The calculator doesn’t just compute totals—it finds break-even points. At what volume does self-hosting become cheaper? How many months until you recoup infrastructure investment? When do reserved instances beat on-demand? These answers require running the numbers across your actual scenario, not generic benchmarks.

Configuring Your Scenario

Usage Volume:

Set your expected request volume using the slider:

• Requests per Month: 1K to 10M requests
• The UI shows tick marks at common thresholds (10K, 100K, 1M)

Token Profile:

Enter your average tokens per request:

• Input Tokens: Average prompt length (typical: 500-2000)
• Output Tokens: Average response length (typical: 200-1000)

Example profiles:

Provider Selection:

Check which options to compare:

• Anthropic: Claude 3.5 Sonnet, Claude 3 Haiku
• OpenAI: GPT-4o, GPT-4o-mini
• Self-Hosted: AWS, GCP, Azure GPU instances

Understanding the Results

Cost Comparison Cards:

Each option shows:

• Monthly Cost: Total cost at your volume
• Annual Cost: Projected 12-month cost
• Cost per Request: Unit economics

Example output for 100K requests/month with 1000 input, 500 output tokens:

The recommended option shows a checkmark. At this volume, APIs win decisively.

Break-Even Analysis:

The calculator shows where self-hosting becomes competitive:

Break-even: 575,000 requests/month
Self-hosting becomes cheaper above this volume

A visualization chart shows cost curves intersecting at the break-even point. Below the break-even, API costs are lower. Above it, self-hosted infrastructure pays off.

Advanced Settings

Expand Advanced Settings for detailed configuration:

Self-Hosted Infrastructure:

Hidden Costs:

These hidden costs add 20-40% to raw GPU pricing. The calculator includes them by default so break-even analysis reflects real costs.

Technical Deep Dive

Pricing Model

API Costs:

monthly_cost = (requests x input_tokens x input_price_per_M / 1M)
             + (requests x output_tokens x output_price_per_M / 1M)

Current pricing (as of December 2025):

Self-Hosted Costs:

compute_monthly = gpu_count x gpu_hourly x hours_per_month x (1 - discount)
network_monthly = egress_gb x price_per_gb
storage_monthly = storage_gb x price_per_gb
ops_monthly = ops_hours x hourly_rate
monitoring_monthly = fixed_monitoring_cost

total_monthly = compute + network + storage + ops + monitoring + load_balancer

Instance pricing (AWS, on-demand):

Spot discounts: 60-70% off on-demand Reserved (1yr): 30-40% off on-demand Reserved (3yr): 50-60% off on-demand

Break-Even Calculation

break_even_requests = self_hosted_monthly / api_cost_per_request

months_to_recoup = initial_investment / (api_monthly - self_hosted_monthly)

The calculator solves for the crossover point where API and self-hosted cost curves intersect.

Real-World Scenarios

A startup at 50K requests/month runs the TCO Calculator and finds API costs $2,250/month vs self-hosted at $26,000/month. Break-even is at 575K requests. They stick with Claude API and revisit when they hit 300K requests—well before break-even, but close enough to start planning infrastructure.

An enterprise at 5M requests/month discovers they’re spending $225K/month on API. Self-hosted with reserved H100 instances would cost $98K/month. They’ve crossed break-even by 4x. The calculator shows 4-month payback on infrastructure investment. They start planning migration.

A platform team evaluating GPU options compares A100 vs H100 at their expected 1M requests/month. A100 80GB costs $41K/month with throughput of 500 req/s. H100 costs $98K/month but achieves 1500 req/s. The TCO Calculator shows H100 has lower cost-per-request despite higher hourly rate—throughput matters more than GPU cost.

What You’ve Accomplished

You now have a systematic approach to infrastructure cost analysis:

• Model complete costs including hidden infrastructure expenses
• Find the exact break-even volume for your workload
• Compare multiple providers and instance types
• Make data-driven migration decisions

What’s Next

The TCO Calculator integrates with other Lattice cost intelligence tools:

• Spot Instance Advisor: Model spot vs on-demand trade-offs for self-hosted
• Memory Calculator: Verify model fits on selected GPU
• Model Registry: Pull current API pricing for selected models
• Stack Configuration: Apply TCO-optimized infrastructure to stacks

TCO Calculator is available in Lattice. Model your infrastructure costs before committing to deployment strategy.