The decision deadline has passed. On 23 April 2026 Render replaced its workspace plans, gave legacy workspaces until 1 August 2026 to opt in, and said the ones that did not "will automatically move to the corresponding new plan" on that date (Render docs, New Workspace Plans). That date is now behind us. Render published no separate notice that the migration finished, and the docs page above still reads as though the deadline is in the future, but the pricing page sells only the new plans and the bandwidth doc lists only the new plan names, so the changeover is done in every place a customer would look.
Two things happened in that swap. The per-seat fee disappeared, which is genuinely good if you have a team. And the included outbound bandwidth dropped by a factor of twenty, which almost nobody is talking about. Render is a good product with a real trade-off buried in its pricing, and a lot of what is written about it still quotes numbers that stopped being true in April.
What Render actually is
Render is a managed platform-as-a-service. You point it at a Git branch, it builds and runs your code, and it handles TLS, health checks, and rollouts. It was started by Anurag Goel, an early Stripe employee (TechCrunch), and took $100 million at a $1.5 billion valuation on 17 February 2026 for $258 million total raised (Render blog). It is not going anywhere, which matters when you pick a host.
The product surface is deliberately small. As of 6 August 2026 the service types are Web Service, Static Site, Private Service, Background Worker, Cron Job, Workflow (still beta), Render Postgres, and Render Key Value (Render docs). Notice what is missing: no serverless functions, no edge runtime. Render runs long-lived processes. That is the whole personality of the platform, and it decides most of what follows.
The plan change, and the part that got quieter
Here is Render's own legacy-versus-new comparison, condensed. All figures re-checked on 6 August 2026 against Render's New Workspace Plans doc and the pricing page.
| Feature | Legacy Hobby | New Hobby | Legacy Professional | New Pro |
|---|---|---|---|---|
| Plan fee | Free | Free | $19 per member/month | $25/month flat |
| Team members | 1 | 1 | Billed per seat | Unlimited |
| Outbound bandwidth | 100 GB included | 5 GB included, then $0.15/GB | 500 GB included | 25 GB included, then $0.15/GB |
| Custom domains | 2 | 2, then $0.25/mo each | Unlimited | 15, then $0.25/mo each |
| Build pipeline | 500 mins/month | 500 mins/month | 500 mins/member/month | 1,000 mins/month, then $5 per 1,000 |
| Services | Unlimited | Up to 25 (all types, including suspended) | Unlimited | Unlimited |
Legacy Organization at $29 per member per month becomes Scale at $499/month flat, with SSO, SCIM, and org audit logs moving from not-included to included, and the HIPAA premium going from "20% of compute + $250/month minimum" to "20% of compute (no minimum)". Scale's bandwidth stays at 1 TB. Its custom domains go from unlimited to 25 included, and its build pipeline goes from 500 minutes per member to 5,000 per workspace.
Read the bandwidth row twice. Hobby went 100 GB to 5 GB. Professional went 500 GB to 25 GB. Both are 20x cuts, and the overage is $0.15 per GB.
Render's stated reason, from its own FAQ, is that "the plans replace per-seat billing with flat monthly fees, so your plan cost stays predictable as your team scales" and that "bandwidth and custom domains are billed based on usage, so you only pay for the resources you actually use". Render draws no causal line between the two, so I will mark the next sentence as my reading rather than theirs: seat fees were paying for a lot of egress, and when the seat fees went, the included egress went with them. If you are a five-person team, this is a large discount. If you are one person serving files, it is a large increase. The move was also one-way, whether you chose it or the calendar chose it for you. Render's FAQ answer to "Can I move back to a legacy plan?" is still "No."
What it costs
Render's pricing FAQ is blunt about the three-part bill: "Render charges for three things: your workspace plan (flat subscription), metered features like bandwidth (usage-based), and compute for your applications (usage-based)" (Render pricing). Compute is per service and prorated to the second, on top of the workspace fee.
Instance types for web services, private services, and background workers, as of 6 August 2026. Render's FAQ says compute pricing was not part of the plan change, and none of these numbers moved. The Free tier is web services only; private services and background workers start at Starter.
| Instance type | Price | RAM | CPU |
|---|---|---|---|
| Free | $0/month | 512 MB | 0.1 |
| Starter | $7/month | 512 MB | 0.5 |
| Standard | $25/month | 2 GB | 1 |
| Pro | $85/month | 4 GB | 2 |
| Pro Plus | $175/month | 8 GB | 4 |
| Pro Max | $225/month | 16 GB | 4 |
| Pro Ultra | $450/month | 32 GB | 8 |
| Custom | Contact sales | Up to 512 GB | Up to 64 |
Postgres moved to flexible plans, where the instance type buys compute only and storage is billed separately at "$0.30 per GB per month, prorated to the second" (Render docs). A selection from the 23 tiers published on 6 August 2026:
| Tier | Instance type | Price | CPU | RAM |
|---|---|---|---|---|
| Free | Free | $0 (30-day limit) | 0.1 | 256 MB |
| Basic | Basic-256mb | $6/month | 0.1 | 256 MB |
| Basic | Basic-1gb | $19/month | 0.5 | 1 GB |
| Basic | Basic-4gb | $75/month | 2 | 4 GB |
| Pro | Pro-4gb | $55/month | 1 | 4 GB |
| Pro | Pro-8gb | $100/month | 2 | 8 GB |
| Accelerated | Accelerated-16gb | $160/month | 2 | 16 GB |
That Basic-4gb at $75 costs more than Pro-4gb at $55 is not a typo. The tiers are CPU-to-RAM ratios, not a ladder: Pro is 1:4, Accelerated is 1:8. Two storage rules bite later. You can only increase storage "to any multiple of 5 GB", and "you can't reduce storage for a database". Ever.
Two worked bills, computed from those numbers on 6 August 2026. Render publishes no default or minimum storage for a new paid database, so the storage lines are my assumptions, not quotes.
| Example A: side project, Hobby workspace | Cost |
|---|---|
| Hobby workspace | $0.00 |
| Web Service Starter (512 MB, 0.5 CPU) | $7.00 |
| Postgres Basic-256mb | $6.00 |
| Postgres storage, 5 GB assumed, at $0.30/GB | $1.50 |
| Bandwidth, within the 5 GB included | $0.00 |
| Total | $14.50/month |
| Example B: small production app, Pro workspace, 50 GB egress | Cost |
|---|---|
| Pro workspace (flat, unlimited seats) | $25.00 |
| Web Service Standard (2 GB, 1 CPU) | $25.00 |
| Postgres Basic-1gb | $19.00 |
| Postgres storage, 10 GB assumed, at $0.30/GB | $3.00 |
| Bandwidth: 50 GB used less 25 GB included, at $0.15 | $3.75 |
| Total | $75.75/month |
Two things stand out. The workspace fee is a third of Example B before a single request is served. And 50 GB of egress, which is nothing for a site with images, already triggers overage on Pro.
One billing shape here is genuinely good: cron jobs bill per minute, from $0.00016/minute on Starter, and you pay only while the job runs. Pennies for a nightly script beats $7 a month for an always-on box to run it.
The price cliff is not where people say it is
The received wisdom is that Render gets expensive when you leave Starter. Do the arithmetic and that is wrong. Ratios computed from the instance table above, as of 6 August 2026:
| Step | Ratio |
|---|---|
| Starter $7 (512 MB) to Standard $25 (2 GB) | 3.57x price for 4x RAM, proportionate |
| Standard $25 (2 GB, 1 CPU) to Pro $85 (4 GB, 2 CPU) | 3.40x price for 2x RAM. This is the cliff. |
| Pro $85 to Pro Plus $175 (8 GB) | 2.06x for 2x RAM |
| Pro Plus $175 to Pro Max $225 (16 GB) | 1.29x for 2x RAM, the best step on the board |
| Pro Max $225 to Pro Ultra $450 (32 GB) | 2.00x for 2x RAM |
Starter to Standard is fine value. The real problem there is that nothing exists between 512 MB and 2 GB, so an app that needs 700 MB pays for 2 GB. The bad step is Standard to Pro: 3.4x the money for 2x the machine. If your app is about to outgrow 2 GB, that is the number to plan around.
The free tier and its two footguns
Render's free tier is real, and unusually honest about its own limits. It also survived the plan change untouched. The following is verbatim from Render's Deploy for Free doc, read on 6 August 2026.
Spin-down: "Render spins down a Free web service that goes 15 minutes without receiving any inbound traffic." Spin-up: "This process takes about one minute." Note that number. The figure repeated everywhere online is 30 seconds; Render's own docs say about a minute, and the docs win. You also get 750 free instance hours per workspace per calendar month, which do not roll over.
Then the footgun that eats side projects. "Free Render Postgres databases expire 30 days after creation." After that, "you have a grace period of 14 days to upgrade it to a paid instance type. After the grace period, Render deletes the database (along with all of its data)." This is why older articles are wrong: the expiry was cut from 90 days to 30 effective 20 May 2024, with existing databases grandfathered (Render changelog). A free database is also capped at a fixed 1 GB of storage, one per workspace, and it supports no backups at all, so no point-in-time recovery. The free tier is $0 but it self-destructs, and Example A's $14.50 is the real floor for anything you intend to keep.
Free web services also have an ephemeral filesystem, cannot attach a disk, and get no edge caching, no SSH, and no scaling. Free Key Value is in-memory only and loses everything on restart, including when you upgrade it.
For contrast: Vercel's Hobby plan "restricts users to non-commercial, personal use only" (Vercel docs). Render's free docs state no equivalent prohibition, but I did not read its Terms of Use, so read that as an absence of a documented restriction rather than as permission.
The limits that will actually bite you
Disks kill your deploy story. The biggest practitioner gotcha on the platform, and the docs state it plainly: "Adding a disk to a service prevents zero-downtime deploys", because "When you redeploy your service, Render stops the existing instance before bringing up the new instance." And separately: "You can't scale a service to multiple instances if it has a disk attached" (Render docs, Persistent Disks). The moment you need local state, you lose both graceful rollouts and horizontal scaling. Disks are $0.25/GB/month and can grow but never shrink.
Five regions, and you cannot change your mind. Oregon, Ohio, Virginia, Frankfurt, and Singapore. That is the list. "Render doesn't currently support changing the region for an existing service or database" (Render docs, Regions). Pick wrong and you migrate by hand. Services in different regions cannot talk over the private network.
There is no edge compute. Render's edge caching is a CDN response cache, not a runtime, and "Edge caching is not available for free web services" (Render docs). Your code does not run at the edge. If that is what you need, this is the wrong platform, not a cheaper one.
Real previews cost money and need Pro. Hobby gets single-service previews, which spin up one service from a pull request. The full-stack version, which clones your whole environment including databases, is the one gated: "Preview environments require a Pro plan or higher", and "Preview resources are billed just like regular Render services and are prorated by the second" (Render docs). Horizontal autoscaling, HTTP request logs, the performance build pipeline, and the OpenTelemetry metrics stream are Pro and above too. Support is the one place Hobby is not fenced off: as of 6 August 2026 the pricing table shows community, email, and chat support on every plan, the free one included.
Reliability, measured rather than asserted
Rather than repeat folklore about Render's build queues, I pulled the Statuspage API on 6 August 2026 and counted (status.render.com).
| Metric | Value |
|---|---|
| Incidents in feed | 50 |
| Date range | 10 December 2025 to 27 July 2026 |
| Average | ~6.6 per month |
| Impact breakdown | 27 minor, 18 major, 4 critical, 1 none |
| Deploy, build, or queue related | 16 of 50 |
| Datastore related | 6 of 50 |
| Per month, Dec 2025 to Jul 2026 | 1, 5, 5, 6, 8, 6, 10, 9 (December truncated by the cap) |
Two caveats before anyone quotes this. The feed is capped at 50 entries, so 10 December 2025 is an API limit, not the start of Render's history. And incident count is not uptime: I did not pull component-level availability, and many of these were regional or dashboard-only.
What the data does support is the build-queue reputation. Sixteen of fifty incidents are deploy-related: "Stuck queued deploys" (11 June 2026), "Degraded builds/deploys (all regions)" (11 May and again 20 July 2026), "Delayed/missing autodeploys" (9 June 2026). On 11 June alone there were three separate deploy incidents, plus a platform-latency incident the same day. Six touched datastores, including "HA Postgres Unavailability" on 7 May 2026, which is the high-availability feature you pay roughly double for having an outage.
July was the rough month in this window. All four of the feed's critical incidents landed between 15 and 24 July: "Render Dashboard Unavailable", "Service Disruption in Oregon", and two separate GitHub failures that broke git clones and new service creation. Give Render its due on those last two. A GitHub outage is not Render's outage, but it is still your deploy that does not happen, which is the honest way to read a hosted build pipeline. What I cannot give you is a typical build queue time. Nobody publishes one, and incidents are not latency.
A render.yaml that actually parses
Render's Blueprint spec has moved and most samples online have not. Two traps: the field is now autoDeployTrigger with a quoted value, replacing the deprecated autoDeploy, and postgresMajorVersion must be quoted or YAML coerces it to an integer. Both are quoted in Render's own reference example (Blueprint YAML reference).
services:
- type: web
name: my-app
runtime: node
plan: starter
region: oregon
branch: main
buildCommand: npm ci && npm run build
startCommand: npm start
healthCheckPath: /healthz
envVars:
- key: NODE_ENV
value: production
- key: SESSION_SECRET
generateValue: true
- key: STRIPE_API_KEY
sync: false
- key: DATABASE_URL
fromDatabase:
name: my-app-db
property: connectionString
databases:
- name: my-app-db
plan: basic-1gb
region: oregon
postgresMajorVersion: "18"
The env var forms are the useful part. generateValue: true generates a base64-encoded 256-bit value. sync: false prompts you for the secret in the dashboard instead of committing it, which is where every API key belongs. fromDatabase wires the connection string without either of you typing it. That Blueprint costs $7 plus $19 plus storage at $0.30/GB per month, at 6 August 2026 prices.
The PaaS premium, quantified
This is the section that decides it. Render Standard is 2 GB and 1 CPU for $25/month, plus the workspace fee, with no bandwidth of its own. A DigitalOcean Basic Droplet with 2 GiB, 1 vCPU, 50 GiB SSD, and 2,000 GiB of transfer is $12/month (DigitalOcean pricing, read 6 August 2026).
So Render is about twice a comparable VPS on compute alone, and closer to four times once a single-service production setup carries the $25 Pro workspace fee on top. What you buy for that: git-push deploys, zero-downtime rollouts, health checks, managed TLS, managed Postgres with point-in-time recovery, and not being a sysadmin at 2 a.m. Whether that is worth the $13 to $38 a month of difference is the actual decision, and for plenty of people it obviously is.
Egress is where it stops being close. Cost of 1 TB outbound, computed from each vendor's published rate on 6 August 2026:
| Platform | 1 TB outbound |
|---|---|
| Render, over the included pool at $0.15/GB | $150.00 |
| Netlify, 20 credits/GB | up to ~$205 on the smallest top-up pack, far less on higher Pro tiers |
| Railway, $0.05/GB | $50.00 |
| Fly.io, $0.02/GB in North America and Europe | $20.00 |
| DigitalOcean, within the droplet's 2,000 GiB allowance | $0.00 |
| Hetzner Cloud, free to 20,000 GB, then €0.001/GB | $0.00 |
Render is 7.5x Fly.io on egress. If the workload is bandwidth-heavy, Render is the wrong shape at almost any scale, full stop. (The Netlify figure is a derivation, not a published rate: Netlify prices bandwidth at 20 credits per GB and sells add-on credits at $5 per 500 or $10 per 1,500, so the per-GB number depends on your tier. See Netlify pricing. The Hetzner figures come from the traffic data behind its own cost calculator; I dropped the instance-price comparison I ran in July because the SKU I quoted no longer appears in Hetzner's published data.)
One distinction worth naming: Railway meters actual consumption rather than allocated capacity (Railway pricing, read 6 August 2026). That, not the headline rate, is the real Render-versus-Railway split, because a mostly-idle app on Railway costs far less than sticker math suggests while Render bills the full instance price regardless.
About Heroku, accurately
Render is usually pitched as the modern Heroku, and the 2026 chapter of that story is being overclaimed badly. Here is what the primary sources actually say.
Heroku killed its free plans, with free dynos and data services shutting down on 28 November 2022, because, per Bob Wise's "Heroku's Next Chapter" post, "our product, engineering, and security teams are spending an extraordinary amount of effort to manage fraud and abuse of the Heroku free product plans" (Heroku blog). That is the vacuum Render grew into.
Then on 6 February 2026 Heroku announced it is "transitioning to a sustaining engineering model focused on stability, security, reliability, and support", with "an emphasis on maintaining quality and operational excellence rather than introducing new features" (Heroku blog). What stops: "Enterprise Account contracts will no longer be offered to new customers." What does not: pricing, billing, the core platform, production support, and existing Enterprise subscriptions. The post sets no end-of-life date and says outright, "There is no change for customers using Heroku today." Trade press calling this an imminent death is interpretation. Believe the announcement, not the migration vendors quoting it.
Be equally careful with Render's marketing in the other direction. Render's docs say "a web service on Render with 2GB of RAM is $25/mo. A Heroku dyno with 2.5GB of RAM is $250/mo" (Render docs). Both halves are literally true against Heroku's published prices, and the comparison is cherry-picked: Performance-M at $250 is a dedicated-CPU dyno, while Render Standard is a shared 1-CPU instance. The fair analogue is Heroku Standard-2X at $50 for 1 GB. At $50.00/GB against Render's $12.50/GB, Render is about 4x cheaper per GB, and both sides of that were still true on 6 August 2026. Real, but not the 10x the framing implies.
The verdict
Render is the right host when your workload is a long-running process with a database, your team is more than one person, your traffic is requests rather than bytes, and your time is worth more than $20 a month. The seat-fee removal makes it meaningfully better for teams than it was in March, and managed Postgres with automatic point-in-time recovery, per-second cron billing, and the zero-downtime deploy machinery are all things you would otherwise build badly yourself. Point-in-time recovery now reaches every paid database, with a 3-day window on Hobby and 7 days on Pro and above.
Render is the wrong host when you serve a lot of bytes, when you need edge compute, when you need a region it does not have, when you need a persistent disk and horizontal scaling at once, or when your app is fundamentally static.
That last one describes this site, so I should say it plainly. jwatte.com is Eleventy plus Netlify plus a handful of functions. Render has no edge runtime and no serverless product; the closest thing it sells is a $7/month always-on process to do what a static host does for nothing. Its static sites are free, but they draw from the same workspace pool, since "web services and static sites count against your monthly included amounts of outbound bandwidth and pipeline minutes" (Render docs). On Hobby that pool is now 5 GB. My own site is the counter-example to my own recommendation, and I would rather say so than pretend the tool fits every hand.
If you want the whole map of picking infrastructure that does not quietly meter you, that argument is the spine of The $97 Launch.
Related reading
- The Cloudflare Developer Platform, Explained Like You're Going to Ship on It: the edge-compute platform Render has no answer to, and the contrast that decides a lot of these calls.
- Where Should a Small Business Host Its Website and Apps?: the same decision framed for a brochure site, a booking app, and an agency.
- NeonDB: Serverless Postgres for Side Projects and Scaling: the managed-Postgres alternative if you like Render's app layer but not its database bill.
- Netlify Database, in Practice: what a database branch per deploy preview buys you, versus Render's Pro-gated preview environments.
- Static Site Generators Compared in 2026: if your app is actually static, start here instead.
Fact-check notes and sources
This post was first published on 15 July 2026 and updated on 6 August 2026, after the plan migration deadline passed. Every price, quota, and tier limit in it was re-read on 6 August 2026; the dated historical items below, such as the 2022 Heroku shutdown and the February 2026 announcements, are carried over from the original research. These numbers change often, so verify anything load-bearing against the live pages.
- The plan change and the bandwidth cut: Render Docs, New Workspace Plans (legacy vs new tables, the 1 August 2026 auto-migration, the "no" on moving back) and the changelog entry dated 23 April 2026. As of 6 August 2026 that doc still carries its pre-deadline wording and Render has published no changelog entry marking the migration complete, so the evidence that it is done is indirect: the pricing page offers only Hobby, Pro, Scale, and Enterprise, and the bandwidth doc lists included amounts for those four names and no others. Bandwidth is billed per workspace, not per service: Outbound Bandwidth.
- Pricing: render.com/pricing for all instance tables, workspace fees, disk and Postgres storage rates, cron per-minute pricing, and the three-part billing FAQ. Postgres flexible plans, the storage rules, the 23 tiers, and the point-in-time-recovery windows of 3 days on Hobby and 7 days on Pro and above: Flexible Plans for Render Postgres, cross-checked against the plan comparison table on the pricing page. Instance types by service type: Render Instance Types.
- Free tier: Deploy for Free for the 15-minute spin-down, the "about one minute" spin-up, 750 instance hours, the 30-day Postgres expiry plus 14-day grace, the fixed 1 GB free-database storage and one-per-workspace limit, the ephemeral filesystem, and the static-site bandwidth note. The 90-to-30-day change: Render changelog, 20 May 2024.
- Limits: Persistent Disks (disks prevent zero-downtime deploys and block multi-instance scaling), Regions (still the same five regions on 6 August 2026, no changing after creation), Edge Caching (a response cache, not a runtime), Preview Environments (Pro or higher, billed like real services), Platform Features by Plan (what Pro and Scale unlock), Deploys, Service Types.
- Reliability: counted directly from status.render.com/api/v2/incidents.json on 6 August 2026, replacing the 15 July 2026 pull in the original version of this post. Fifty incidents is the feed's cap, not Render's total history, and incident counts are not an uptime figure. Because the cap is a rolling window, older entries have already aged out: the 15 July pull reached back to 2 December 2025, this one only to 10 December.
- Blueprints: Blueprint YAML Reference for
autoDeployTrigger, the quotedpostgresMajorVersion, and the env var forms. Key Value now runs Valkey: Render changelog. - Company: Render, About; Series C extension (17 February 2026, $100M, $1.5B valuation, $258M total, self-reported 4.5M+ developers); Series C; TechCrunch on the seed round. The founding year is genuinely contested between 2018 and 2019 and I have not resolved it.
- Heroku: Heroku's Next Chapter (free-plan removal, 28 November 2022 shutdown, the fraud-and-abuse quote); An Update on Heroku, 6 February 2026 (sustaining engineering model, Enterprise contracts for new customers only, no EOL date); Heroku pricing. Render's own comparison, treated as vendor marketing: Render vs Heroku.
- Competitors: Railway pricing ($0.05/GB egress), Fly.io pricing ($0.02/GB for the North America and Europe region group; Fly's rates rise to $0.04 in Asia Pacific and $0.12 in Africa and India), Vercel Hobby, Netlify pricing (20 credits per GB of bandwidth, add-on credits at $5 per 500 or $10 per 1,500), DigitalOcean droplets, Heroku pricing (Standard-2X $50 for 1 GB, Performance-M $250 for 2.5 GB), and Hetzner's cloud traffic data, which is the file behind its cost calculator and lists Hetzner Cloud outbound traffic as free to 20,000 GB and then €0.001 per GB. The July version of this post also quoted a Hetzner instance price from cloud_data.json; that SKU is no longer in the file, so the price came out rather than being guessed at.
This post is informational, not engineering or purchasing advice. I have no affiliation with Render, Heroku, Railway, Fly.io, Vercel, Netlify, Cloudflare, DigitalOcean, Hetzner, or any other product or organization named here, and mentions are nominative fair use. Prices, quotas, tier limits, and product details are current as of 6 August 2026 and change; verify against the live pages before you build on them.
