Why One Startup Abandoned AWS Lambda Web Functions for Custom Compute

When AWS introduced Lambda Web Functions, infrastructure teams hoped it would streamline running web applications without heavy custom setups. However, after experimenting with the feature, startup Easel found that execution environments continue billing for memory across five to seven minutes of idle time. While CPU is billed strictly during active use, memory charges accrue for the entire life of the environment, a factor that becomes costly for platforms managing thousands of intermittently used applications.

Easel attempted to mitigate this by programmatically terminating environments after ten seconds of inactivity, but AWS logged those manual shutdowns as runtime failures. Unwilling to rely on that workaround in production, the team decided to run their own compute hosts for application execution while retaining traditional Lambda for overflow traffic.

When benchmarked against the same Next.js application, the company's self-operated infrastructure answered warm requests up to three times faster and achieved cold starts roughly 3.5 times faster. Although the tests represent early findings from a single region, the custom setup provided better control over resource scheduling and eliminated unexpected idle memory expenses.

What to watch next: Monitor whether AWS introduces configurable idle timeouts for Lambda Web Functions to address developer concerns over idle memory billing.