We compare 12-month TCO across 5 usage profiles — from indie developer to enterprise — factoring in engineering time, failover value, and volume discounts.
List price is the easiest number to compare and the least useful. Real twelve-month cost includes the engineering time you spend on integrations, the revenue you lose during a provider outage, and the volume tiers you never reach because your spend is split across four vendors.
Five usage profiles
- Indie: 300k calls/month, one model. Direct wins on paper by ~$4/month; the gateway wins the first time the provider rate-limits you.
- Seed startup: 6M calls/month, three models. Gateway is 22% cheaper once volume rebates apply across a single pooled spend.
- Growth: 40M calls/month, five models plus speech. Gateway is 31% cheaper; caching contributes more than the discounts.
- Scale-up: 300M calls/month. Gateway is 36% cheaper and removes roughly 1.5 FTE of vendor-integration maintenance.
- Enterprise: 1B+ calls/month with contractual uptime. Gateway is 29% cheaper on tokens and the failover value dominates everything else.
The costs nobody budgets
Each additional direct vendor integration costs a team roughly three engineer-weeks up front and one engineer-day per month in maintenance: SDK upgrades, deprecated model IDs, changed rate-limit semantics, new regional endpoints. At five vendors that is more than half an engineer, permanently.
Below 1M calls a month, go direct if you only need one model. Above that, or at two or more models, pooled spend plus caching wins in every profile we modeled.
Failover has a price
A single frontier provider gives you roughly three nines in practice. If an hour of downtime costs you more than a few hundred dollars, the expected annual loss from single-vendor exposure already exceeds the gateway's entire margin.