Backstage vs Port vs Cortex 2026: Which IDP to Actually Buy

The Internal Developer Portal pitch sounds the same in every vendor deck: developers waste hours hunting for service ownership, deployment runbooks, and “who do I ping if checkout breaks at 3am.” A portal centralizes all of that. Click button, ship code. Easy.

The reality, after watching three teams I know stand these things up, is messier. One team spent six months building a Backstage catalog only to have half the engineers ignore it. Another bought Port and had a working catalog with golden paths in three weeks. The third bought Cortex specifically because their CTO wanted scorecards on every service and didn’t care about self-service deployments.

So which one should you actually buy in 2026? It depends on factors most vendor comparison pages skip — and Port’s recent $100M Series C at an $800M valuation makes the question more interesting than it was a year ago, because the commercial alternative to Backstage now has runway to actually compete.

Why everyone is suddenly buying an IDP

Platform engineering went from a Gartner buzzword to a budgeted line item somewhere around 2024. By 2026, if your engineering org has more than 50 people and you don’t have a platform team, you’re an outlier. The CNCF released a literal documentary about Backstage in March 2026 — that’s the level of mindshare we’re talking about.

The forcing functions are pretty mundane. Microservices counts crept up. Service ownership got fuzzy. Onboarding a new engineer started taking three weeks of “ask Slack” instead of two days of “follow the docs.” Someone in finance noticed that engineers were spending real percentages of their time on coordination rather than shipping. Cue the portal.

I’ll be honest, the productivity numbers vendors quote are mostly nonsense. McKinsey-flavored claims of “30% developer productivity gains” don’t survive a real measurement attempt. What does survive: faster onboarding (weeks to days), fewer “who owns this?” pings, and incident response that doesn’t depend on tribal knowledge. Those are real wins. They’re also harder to put on a slide.

Backstage: free as in puppy

Backstage is Spotify’s open-source developer portal, donated to the CNCF in September 2020 and graduated to Incubating status in March 2022. It’s now undergoing the security audit needed to become a Graduated project. Adoption is enormous — over 3,400 known adopters and 96% internal adoption inside Spotify itself.

The pitch is appealing: free, extensible, owns the standard, plugin ecosystem covers everything from PagerDuty to Argo CD. If you’ve used GitHub or BT’s internal portal in the last few years, you’ve used Backstage.

Here’s the part the vendor blogs are right about: self-hosting Backstage is a real engineering investment. The numbers I’ve seen quoted by Roadie and the Platform Engineering community Slack are consistent — 2 to 4 dedicated engineers, 12 to 18 months to a stable production rollout, three-year TCO often north of $2M when you load in salary, infra, and opportunity cost. Senior platform engineers in the US load to roughly $250K each. The math gets ugly fast.

A few specific traps I’ve watched teams fall into:

• The TypeScript tax. Backstage is a React/TypeScript app. Most platform teams are Go or Python shops who suddenly need to maintain a frontend. Plugin upgrades break things. Material UI migrations eat sprints.
• Catalog drift. catalog-info.yaml files start clean, then rot as services get renamed, archived, or forgotten. Without enforcement (a custom reconciler, GitHub Actions linting, scorecards), the catalog becomes lies-as-a-service within 18 months.
• Software templates that nobody uses. The “scaffolder” feature is genuinely good but requires curation. Templates that aren’t maintained get stale, developers stop trusting them, you’re back to copy-pasting from service-template-old.

None of this means Backstage is bad. Spotify, GitHub, American Airlines, and a long list of large orgs run it successfully. But “free” is doing a lot of work in those vendor decks. If you don’t have a 5-person platform team or a strong appetite for owning a moderately complex app, Backstage will drain budget through engineering hours instead of invoices.

Port: the commercial bet that might actually work

Port raised $100M Series C in December 2025 at an $800M valuation, led by General Atlantic with Accel and Bessemer participating. Total funding sits around $158M. Customers include GitHub, BT, Visa, Sonar, StubHub, and Nando’s. Their public claim is 300% revenue growth year-over-year, which I won’t independently verify but matches the hiring patterns I see on LinkedIn.

What Port actually does well is the data model. Where Backstage gives you opinionated entities (Component, API, Resource, System, Domain) and a YAML catalog, Port gives you “Blueprints” — arbitrary entity types you define with whatever properties matter to you. Want to model GPU clusters as first-class entities with utilization, cost-per-hour, and assigned ML team? Define a blueprint, ingest from your billing API and Prometheus, done.

This sounds abstract until you’ve tried to bend Backstage’s catalog to model something it wasn’t designed for. I once spent a Saturday trying to represent a Snowflake warehouse as a Backstage entity in a way that wasn’t horrifying. Port’s flexibility wins in any org with non-standard infrastructure.

Pricing-wise, Port has a free tier (15 seats, 10K entities, 500 automation runs) and paid plans starting around $19/user/month, but real enterprise pricing is custom and roughly tracks “platform fee + seats + entity count.” Vendor pricing pages omit the meaningful numbers, as always.

The Port pitch in 2026 is heavily AI-flavored — they’re rebranding around “Agentic Engineering Platform” with their funding announcement. Whether that’s substantive or marketing depends on how much you trust agent demos. The underlying portal works regardless. I’d evaluate Port on the catalog and self-service actions, not on the AI agent demo, and let the agent stuff prove itself over the next 12 months.

Where Port is weaker: it’s still a younger product than Backstage. Some integrations require their MCP server setup or webhooks rather than a dedicated plugin. The community is much smaller than Backstage’s, so when you hit an edge case, you’re filing a support ticket instead of finding a Stack Overflow answer.

Cortex: scorecards-first, deployments-second

Cortex takes a different angle. Where Backstage and Port emphasize self-service (“click button, get environment”), Cortex emphasizes engineering quality measurement. Service catalogs, yes — but the headline feature is scorecards: did this service set up Datadog alerts, did the on-call rotation get configured, does it have a runbook, does it pass the production-readiness checklist.

Cortex is YC-backed, has been around since 2019, and lists a customer list heavy on financial services and large enterprise. Pricing lands around $65/user/month based on what’s been publicly reported, with custom enterprise terms and the usual gotchas — auto-renew with 5-10% annual escalation, professional services fees that are negotiable if you push.

The model is more rigid than Port’s. Entities are largely services, with some flexibility for resources and domains, but you’re not modeling arbitrary things the way Port lets you. That’s a feature if you want guardrails — Cortex has opinions about what good engineering hygiene looks like. It’s a constraint if your environment doesn’t fit those opinions.

Where Cortex shines: an engineering leader who wants to measure team health and production readiness across hundreds of services. Scorecards plus the integration set (PagerDuty, Datadog, Sentry, GitHub) give you a coherent view of “is anyone actually maintaining this service?” without writing your own data pipeline.

Where it doesn’t: if your team mostly wants self-service environments and infrastructure provisioning, Cortex is more about visibility than action. You’ll end up bolting on Terraform Cloud or backstage-style scaffolders to fill that gap.

Feature comparison, with the caveats vendor pages skip

The “setup time” row is where vendors get cute. Backstage demos look fast because the demo is a clean install. Production-grade Backstage with auth integration, populated catalog, working CI plugin, and at least one functioning template is a multi-quarter project. Port and Cortex demos are more honest because the SaaS abstracts the setup, though “import from GitHub and manually clean up the catalog” is still a real first month.

Pricing math that actually means something

Skip the vendor’s TEI report. Run the back-of-envelope yourself.

For a hypothetical 200-engineer org with 300 services:

Backstage (self-hosted): 3 platform engineers fully loaded at $250K = $750K/year in salary, plus $20-30K/year in hosting, plus the opportunity cost of those engineers not building product features. Year one is realistically $800K-$1M when you include onboarding and trial-and-error. Year two and beyond drops to roughly $750K-$800K steady state. If those engineers would’ve existed anyway because you have a platform team, the marginal cost is closer to “what else they could’ve built.”

Port: Custom pricing, but the public ranges I’ve seen for orgs at this size are $50K-$150K/year all-in, with implementation typically a few weeks of one engineer’s time. Total year one: $80K-$180K.

Cortex: $65/user × 200 users × 12 months = $156K/year base, plus implementation and integration costs. Total year one: $180K-$220K.

The “Backstage is free” math doesn’t survive contact with actual engineering salaries. The honest framing is: Backstage is cheaper if your platform team would exist anyway, more expensive if you’d otherwise not need to hire one. Port and Cortex shift cost from headcount to invoices, which finance prefers because it’s predictable and doesn’t show up in your engineering ratio.

This is also where I disagree with most vendor blogs that pitch “Backstage TCO is $2M, we’re $200K, easy choice.” That math ignores that a team with strong platform engineers gets second-order benefits beyond the portal — better infra, better deployment automation, better incident response. Backstage isn’t really a portal, it’s a justification for a platform team. Port and Cortex are portals.

When you actually shouldn’t buy any of them

Worth saying out loud: most companies under 50 engineers don’t need an IDP. A README with the on-call rotation, a Notion page with the runbook links, and terraform apply from a shared repo will get you embarrassingly far. I’ve seen 30-engineer startups buy Cortex because the CTO came from a 5,000-engineer org and assumed they needed it. Six months later, the catalog was empty and the contract was up for renewal.

The threshold I use as a rough rule: if you have more than 50 engineers, more than 100 services or microservices, and at least one full-time platform engineer, an IDP starts to pay for itself. Below that, Slack, GitHub, and discipline are cheaper than any tool.

The other “don’t buy yet” signal: if your service catalog has no source of truth today, an IDP won’t fix that. The portal becomes garbage-in-garbage-out. Spend a quarter writing a script that reconciles your AWS resources, GitHub repos, and PagerDuty rotations into a normalized list before you spend money on a portal that depends on that list existing.

How I’d actually decide

If I were picking today, my decision tree looks like this:

• Platform team of 3+ engineers, willing to own a TypeScript app, want full control: Backstage. Use Roadie or Spotify Portal if you want managed Backstage without the FTE cost.
• Smaller platform team, want self-service environments and golden paths in weeks not quarters, comfortable with custom data models: Port. The blueprint flexibility is the differentiator and the SaaS removes the maintenance tax.
• Engineering leader who cares more about measuring quality than provisioning infra, services are the primary entity, want strong scorecards out of the box: Cortex.
• Under 50 engineers: save the money. Write a README. Revisit in 18 months.

The honest answer is that all three of these are good products in 2026. The bad outcome isn’t picking the wrong one — it’s picking any of them without a clear owner, then watching the catalog rot.

If you do nothing else after reading this, ask whoever’s championing the IDP purchase: “who is going to own this in 18 months, and what does success look like?” If they can’t answer, neither Backstage nor Port nor Cortex is going to save you.

Sources used while writing this post: Port’s Series C announcement, TechCrunch on Port’s $800M valuation, CNCF’s Backstage documentary announcement, and Roadie’s analysis of Backstage TCO. Pricing details current as of April 2026 — verify against vendor pages before signing anything.