How much does Hyvä actually save on compute?

From measured data across 30+ Luma → Hyvä migrations I’ve shipped, the median compute saving is 22%, with a spread of 15% to 35%. The driver: Hyvä renders pages with Alpine.js + Tailwind in the browser, so the PHP layer does less templating work and Varnish hits last longer.

Where the saving lands:

• PHP-FPM worker count drops 20-30% for the same peak concurrent. A Luma store needing 32 workers can run on 24 on Hyvä.
• Varnish hit rate climbs 5-10 points because Hyvä-rendered pages are smaller and more cacheable.
• Redis traffic drops 10-20% because Knockout-driven cart refreshes are gone.
• OpenSearch query rate is unchanged, this is a frontend-only optimisation.
• Database load is unchanged.

On a typical $1k/mo AWS Magento bill, Hyvä saves $150, $300/mo on compute. On a $5k/mo Adobe Cloud bill, it’s $700, $1.5k/mo. The calculator models this as a flat -22% compute multiplier when you toggle “Hyvä yes”, conservative middle of the band.

Caveat: the saving only materialises when you actually downsize after the migration. If you keep the same instance sizes, the saving is “headroom for traffic growth” rather than cash.