On October 30, Vercel announced that Data Cache, the cache behind fetch and unstable_cache in Next.js, is moving to Runtime Cache. Your code stays the same, but limits and billing change, and existing free Hobby teams switch to the new system on November 29.
Data Cache moves to Runtime Cache: what exactly changes
A quick reminder: a cache keeps a ready result, so the next request does not have to ask the database or an external API again.
Until now, Vercel had two separate caches for data. Data Cache handled fetch caching and the unstable_cache function in Next.js. Runtime Cache handled the newer options: the 'use cache: remote' directive in Next.js 16 and the getCache function from the @vercel/functions package. According to Vercel's blog post, after the migration all four APIs run on one service. SvelteKit and Nuxt can use Runtime Cache too.
The key details from Vercel's changelog:
- Data Cache is leaving beta. The switch covers every Next.js version and needs no code changes.
- Teams created on or after October 30 use Runtime Cache from the start.
- Existing Hobby teams migrate automatically on November 29, 2026. Pro teams will get a notice with their migration date, and Enterprise customers will hear from their account team.
- Hobby and Pro teams can opt in early from the team's CDN settings, but the new pricing applies as soon as they do.
What it costs
Data Cache reads and writes were not billed separately until now. After the migration they follow Runtime Cache rates: from $0.40 per million read units and from $4 per million write units, depending on the region. One unit is 8 KB of data, according to the Runtime Cache documentation. The Hobby plan includes 1 million read units and 200,000 write units free each month, shared by all projects on the team. The Pro plan draws on its $20 monthly infrastructure credit, and usage above it is billed at regional rates.
The documentation also explains what happens when a Hobby team goes over the limit: you pay nothing extra, but Runtime Cache is paused for the rest of the billing period, typically up to 30 days. The app keeps working, it just hits your database or external API more often.
In return, there is more room. On Pro, each project and each environment gets its own allocation, 8 times the old team-wide Data Cache pool, and Enterprise storage grows 10 times. On Hobby, all projects on the team still share a single cache.
How it works and what to do now
Runtime Cache is regional: each region keeps its own copy of the data, and a tag invalidation typically reaches the other regions within a few hundred milliseconds. In Next.js, invalidating a tag means, for example, calling revalidateTag('blog') after you publish a new blog post. Vercel says the service handles more than 440 million reads and over 18 TB of data per hour.
Before migrating, Vercel suggests checking your Data Cache usage in the Observability tab and cutting unnecessary reads with its optimization guide. You can turn off caching for fetch and unstable_cache in CDN settings, but then more requests go to your data source, and the migration happens anyway.
What this means if you are learning to code
If you deploy your learning projects on Vercel's free plan, you do not need to change any code. Still, open Observability before November 29 to see how many reads your apps make. With little traffic, a million reads a month is usually plenty, but one project with an unlucky cache setup can use up the limit for all of them, because the Hobby pool is shared.
It is a good moment to learn the cache layers in Next.js. The CDN cache stores whole HTTP responses, while Runtime Cache stores the data your function fetches from a database or an API. A simple exercise to start with: in one server component, call fetch with cache: 'force-cache' and next: { revalidate: 3600 }, then check in Observability how many reads and writes it produces.
You can learn the basics of rendering and caching in the App Router in our Next.js course. For wider context, read our articles on Next.js and Vercel, and if fetch and promises are new to you, start with the JavaScript documentation. You do not need to know the prices to the cent yet. Just remember that every read from a cloud cache has a price and a limit.
Also today
- Secrets leak at Northflank. The app deployment platform reported an incident on October 30 and told customers to assume all environment variables had been decrypted and to rotate their keys, starting with AI model API keys. The same day, Wiz counted nine AI developer platforms that disclosed incidents in October, with attackers hunting mainly for keys stored in environment variables. Source, Wiz analysis
- New findings on the fake Claude Code extension in Cursor. Phorion explained that the malicious code in versions 2.1.312 and 2.1.313 offered on Cursor's marketplace runs as soon as the extension loads, adds an SSH key labelled
svc-telemetry, and that uninstalling the extension alone does not remove the backdoor. Cursor updated its notice with new indicators of compromise. Source, Cursor's notice - A shared robots.txt signal for AI training. A new IETF draft defines the
__usage_ai-trainingtoken: aUser-Agent: __usage_ai-traininggroup withDisallow: /means the site objects to its content being used to train models. Google plans to support it in 2027, and OpenAI said it will honor the signal. Source, Google's blog - Claude Code 2.1.314. A new
permissions.blockWritesOutsideWorkingDirectoriessetting blocks file writes outside the working directories in every permission mode, andclaude -pnow exits with error code 1 instead of 0 when startup fails. Source - pnpm 12.17.0. It adds
pnpm peers check --globaland the:prod,:optionaland:peerselectors inoverrides, and files that pnpm creates get normal permissions again, which fixes Docker images run by a different user. Source - PostgreSQL 19 is later than usual. Robert Haas, a project committer, expects the release in November or later. In his view, AI tools are finding many more bugs, and the number of fixes back-patched to the older branch grew by more than 74% this year. Source
Frequently asked questions
Do I need to change my Next.js code after the move to Runtime Cache?
No. Vercel says fetch caching and unstable_cache switch over on every Next.js version with no code changes. What changes is billing and limits, not the API.
When does a free Hobby team move to Runtime Cache?
Existing Hobby teams move on November 29, 2026. Teams created from October 30 use Runtime Cache right away, and everyone else can opt in earlier in the CDN settings.
What happens if I go over the free Hobby limit?
According to the documentation, Vercel will not charge you, but it pauses Runtime Cache for the rest of the billing period, typically up to 30 days. Your site keeps working, but requests then go straight to the data source.
Sources
- Data Cache is moving to Runtime Cache, Vercel, 2026-10-30
- Simplifying runtime caching on Vercel, Vercel, 2026-10-30
- Runtime Cache, Vercel Docs, 2026-08-28
- Optimize Runtime Cache on Vercel, Vercel, 2026-10-30
- Security incident, Northflank, 2026-10-30
- Black October for AI Developer Platforms, Wiz, 2026-10-30
- Official Claude Code Extension Compromise: Security Advisory, Phorion, 2026-10-30
- Security notice: malicious version of Anthropic.claude-code (2.1.312 and 2.1.313), Cursor, 2026-10-29 (updated 2026-10-30)
- Machine-Readable Training Preference Signal, IETF Datatracker, 2026-10-30
- Supporting a shared AI training control for websites, Google Search Central, 2026-10-30
- Claude Code v2.1.314, Anthropic (GitHub), 2026-10-30
- pnpm 12.17.0, pnpm (GitHub), 2026-10-30
- Postgres 19 is late, and it's AI's fault, Robert Haas, 2026-10-30
This article was prepared with the help of AI from primary sources and reviewed by the CodeWorlds editorial team.