We liked Next.js, and we built a lot of successful software on it. But over the last year, keeping up with the App Router and debugging opaque cache directives started to feel like a second job. We recently migrated our core SaaS architecture to TanStack Start, dropping our local dev startup time from 12 seconds down to just 2.
Here is exactly why we made the switch, and what the numbers look like on the other side.
The Core Signal:
Waiting on Turbopack: Our initial local route loads took 10 to 12 seconds under the full weight of the App Router and RSCs. Switching to a Vite-based runner dropped that to 2 seconds, completely changing the momentum and mood in our Slack channels.
Production memory leaks: We watched Next.js processes climb to 9GB locally and repeatedly OOMKilled our Kubernetes pods in production. Moving to TanStack’s explicit Node architecture eliminated these opaque router cache memory leaks entirely.
The RSC security tax: The recent React2Shell (CVSS 10.0) vulnerability proved that implicit Server Components turn your UI protocol into a security-critical transport. We wanted explicit JSON boundaries back, not a massive new attack surface to patch.
The multi-cloud ceiling: Next.js heavily incentivizes Vercel's edge network, which scales unpredictably for our workloads. Compiling to a standard Node build unlocks standard container hosting, saving us thousands in infrastructure costs.
The Contrarian Take:
The industry insists that compiler magic and implicit server boundaries are the only way forward for React. We disagree.
For highly interactive SaaS applications, predictability beats abstraction every time. Explicit route loaders and strict TypeScript boundaries prevent the exact classes of runtime bugs that implicit frameworks tend to push into production. We traded the magic for explicitness, and the code is simply easier to reason about.
The Migration Playbook:
We didn't pause our product roadmap for a massive, six-month rewrite. We used a phased, strangler-fig approach—putting an Nginx proxy in front and moving our highest-friction dashboards first. It wasn't perfectly smooth, and Vite SSR definitely has its quirks.
Read the exact migration playbook, our hard metrics, and the trade-offs we hit →
isanudelhi•1h ago
Here is exactly why we made the switch, and what the numbers look like on the other side.
The Core Signal:
Waiting on Turbopack: Our initial local route loads took 10 to 12 seconds under the full weight of the App Router and RSCs. Switching to a Vite-based runner dropped that to 2 seconds, completely changing the momentum and mood in our Slack channels.
Production memory leaks: We watched Next.js processes climb to 9GB locally and repeatedly OOMKilled our Kubernetes pods in production. Moving to TanStack’s explicit Node architecture eliminated these opaque router cache memory leaks entirely.
The RSC security tax: The recent React2Shell (CVSS 10.0) vulnerability proved that implicit Server Components turn your UI protocol into a security-critical transport. We wanted explicit JSON boundaries back, not a massive new attack surface to patch.
The multi-cloud ceiling: Next.js heavily incentivizes Vercel's edge network, which scales unpredictably for our workloads. Compiling to a standard Node build unlocks standard container hosting, saving us thousands in infrastructure costs.
The Contrarian Take: The industry insists that compiler magic and implicit server boundaries are the only way forward for React. We disagree.
For highly interactive SaaS applications, predictability beats abstraction every time. Explicit route loaders and strict TypeScript boundaries prevent the exact classes of runtime bugs that implicit frameworks tend to push into production. We traded the magic for explicitness, and the code is simply easier to reason about.
The Migration Playbook: We didn't pause our product roadmap for a massive, six-month rewrite. We used a phased, strangler-fig approach—putting an Nginx proxy in front and moving our highest-friction dashboards first. It wasn't perfectly smooth, and Vite SSR definitely has its quirks.
Read the exact migration playbook, our hard metrics, and the trade-offs we hit →