Key details

  1. Render announced the expanded compute catalog August 26–27, 2026.
  2. More than 15 new memory-optimized plan options were added.
  3. At 2+ CPU tiers, users can choose multiple RAM quantities independently of CPU count.
  4. Four-CPU web services can now choose 8GB, 16GB or 32GB RAM.
  5. New 12-CPU tiers for web services, private services and background workers offer up to 96GB RAM.
  6. New plan IDs encode specifications, such as `4c-32g` and `12c-96g`.
  7. Legacy plan IDs remain supported across Render tooling.
  8. Render says existing plans receive no pricing changes.

What builders should take away

  1. Profile actual peak and sustained memory before moving to a higher-RAM plan; choose the smallest CPU tier that satisfies real compute demand rather than treating RAM and CPU as one scaling unit.
  2. For agent and retrieval workloads, measure memory growth over long sessions and concurrent jobs, not only a short synthetic request.
  3. Compare a single large-memory instance with multiple smaller instances on both cost and failure-domain risk.
  4. Update Render CLI, Terraform provider and other integrations before using new spec-based plan IDs, even though legacy identifiers remain backward compatible.
  5. Record the plan ID and specs explicitly in infrastructure configuration so future plan-catalog expansion does not make an old descriptive name ambiguous.

What changed

On August 26–27, 2026, Render expanded its service compute catalog with more than 15 new CPU/RAM combinations. At every tier with two or more CPUs, customers can choose higher-memory options independently of CPU count; for example, four CPUs can now be paired with 8GB, 16GB or 32GB of RAM. Render also added 12-CPU web-service, private-service and background-worker plans with 24GB, 48GB or 96GB of RAM. At the same time it replaced descriptive plan names with spec-based IDs such as `4c-32g`, while keeping legacy IDs backward compatible across the API, CLI, SDK, Blueprints and Terraform integrations.

Why it matters

Memory-heavy application economics do not always scale with CPU. Agent harnesses can hold long-lived session state, retrieval systems keep indexes or embeddings resident, and databases or workers may need large heaps without saturating processors. Previously, getting more memory could force teams into a higher CPU tier they did not need. Render’s new catalog makes RAM a more independent cost/sizing decision and raises the ceiling for stateful application workloads without requiring a move to another hosting model.

RAM can now scale independently at the same CPU tier

Render’s expanded catalog adds multiple memory ratios at 2, 4 and 8 CPUs, including 2c-8g, 2c-16g, 4c-32g, 8c-16g and 8c-64g options for supported service types. That lets teams tune for memory pressure without automatically stepping up processor count.

The top long-running service tier reaches 12 CPUs and 96GB

Web services, private services and background workers can use new 12-CPU plans with 24GB, 48GB or 96GB RAM. Render explicitly cites agent platforms and other memory-intensive workloads as a motivation. Cron jobs also receive additional memory-heavy combinations, although their plan matrix differs from continuously running services.

Existing deployments do not need a naming migration

Render renamed 'instance types' to 'compute plans' and introduced spec-based IDs. Existing legacy identifiers such as `standard`, `pro` and `pro_ultra` remain valid in tooling and map to identical existing specifications and prices. Teams can adopt the new IDs gradually; the entirely new memory configurations have only the new spec-based identifiers.

The economic benefit depends on the new-plan prices

Render says existing plans keep their prices unchanged. The significance of the new choices is that a workload can select a different resource ratio rather than overprovisioning CPU solely for RAM. Builders should compare the new plan’s actual monthly rate with their current tier and with horizontal scaling; more memory on one instance may reduce CPU waste but can also increase failure concentration.

What to watch next

  • Pricing and adoption of the new high-memory plans relative to competing hosting providers.
  • Whether Render extends independent CPU/RAM sizing to smaller sub-2-CPU tiers or offers even larger memory ratios.
  • Whether autoscaling gains memory-aware policies that make the expanded plan matrix easier to use dynamically.
  • Operational evidence from long-running agent, vector-search and database workloads on the 64GB and 96GB tiers.

Still unclear

  • Render’s claim that the new ratios reduce overprovisioning is workload-dependent; CPU, memory bandwidth, I/O and horizontal scaling can still dominate performance.
  • Existing plan prices are unchanged, but the new plan options have their own rates and should be compared against alternatives rather than assumed cheaper.
  • A larger single instance can reduce resource waste while increasing the impact of an instance-level failure.

Sources

Direct reading behind this dossier.

4 sources
Updates to Render Compute Plans
Render Docs primary_documentation

Migration details and confirmation that existing plan pricing is unchanged.

Discussion

Discussion is reader-contributed. Comments are not part of the BTN dossier or its editorial evidence.

0 visible comments

Join the discussion

Keep comments useful and relevant. Reader contributions may be moderated and are not BTN editorial evidence.

Sign in to comment