top of page
Rechercher

SaaS is cracking. Figma fell first. Who's next?

  • Photo du rédacteur: benjamin. brl
    benjamin. brl
  • 28 avr.
  • 5 min de lecture
Pic from Carnets d'entrepreneurs podcast, check it out
Pic from Carnets d'entrepreneurs podcast, check it out

Two data points.

Figma's stock is down roughly 85% from its August 2025 peak.

Not because they missed a quarter. They closed 2025 with $1.06 billion in revenue, up 41% year over year, and a net dollar retention of 136%, the best numbers they had posted in two and a half years. In a different decade, those figures underwrite a $30 billion market cap.

The stock fell anyway.

Last month I met Ivan Rimbaud. He runs a pest control business in the 18th arrondissement of Paris, called Halte Nuisibles. Rats, cockroaches, pigeons, a fleet of technicians, the whole thing. He told me he had rebuilt his entire SaaS stack internally. No more subscriptions. Just infrastructure to maintain.


These two facts belong in the same story.


The 2010s answer

SaaS was a rational response to a real constraint. In 2012, if you wanted a tool that actually fit your company, like a workflow dashboard, an internal CRM, or a custom scheduler, you needed engineers, months, and infrastructure. Most companies did not have all three. So you rented someone else's generic version instead of building your own specific one.

That was the deal. Thousands of SaaS companies were built on the assumption that the deal would hold indefinitely.

It stopped holding.

The part most people miss: it is not because engineers got faster. It is because the ability to build tools escaped engineering entirely. Ivan Rimbaud did not hire developers. He used AI to assemble the tools his business actually needed, and now pays for the servers they run on instead of a dozen monthly invoices.

The bottleneck was never really "can this be built." It was "can someone inside your company build it." That line just moved. Not by a little.


What the market actually saw in Figma

Figma is not standing still either. They launched their own AI product, Figma Make, and by BTIG's account, it is working: weekly active users up roughly 70% quarter over quarter since the broader March rollout. They wired Figma into AI coding agents via an MCP server, letting Claude Code, Cursor, and Copilot read Figma designs directly, and now write them back. Revenue grew. Retention grew.

The market did not care.

The market is not betting against Figma's revenue. It is betting against the default that makes Figma's revenue possible. "We need a design tool, let's subscribe to Figma" was a one-step decision for a decade. For a rising share of use cases it is now two steps: "Do we need a design tool, or do we need the output, and can we generate it directly?"

On April 17, 2026, Anthropic launched Claude Design. Three days earlier, Anthropic's Chief Product Officer, Mike Krieger, resigned from Figma's board. Figma fell roughly 7% in a single session.

Not because Claude Design is ready to replace Figma at scale. It is not, yet. Because the market no longer believes the category is safe.

The signal is broader than one Anthropic launch. Net-new AI-native design tools are being built from scratch. Pencil.dev launched in January 2026 and crossed 100,000 users within weeks. Founded by Tom Krcha, an ex-Adobe and Miro product leader, the tool stores design files in your Git repository, versions them like code, and gives AI agents full read-write access to the canvas. Figma's own MCP server gives agents read-only access. Open-source alternatives ship alongside it. The category is being re-architected from first principles by independent teams with no SaaS revenue to defend.

This is what repricing looks like. Not a crash. A slow correction of an assumption that held for fifteen years.


Your stack was assembled under the old default

Every SaaS line on your P&L was bought under one rule: rent first, build only if critical. That rule was correct when you bought it. It is not automatically correct today.

The question is not "build everything." For each line on your stack, the question is whether the old logic still holds.

Some categories still make sense to rent. Cloud infrastructure. Compliance and security tooling. Anything with real network effects. Anything sitting on multi-tenant data you cannot reproduce. These are not cracking.

The categories that are cracking sit further down the stack. Internal workflows. Dashboards. Glue tools between systems. Niche CRMs. Lightweight analytics. Anything you bought because "building it in-house would cost more than the subscription." That calculation has changed. Most companies have not re-run it.

A CEO does not need to cancel everything. A CEO needs to know which assumption under each line is still true.


This is not an AI migration

Replacing SaaS with AI-assisted builds is not the same as replacing SaaS with AI dependency.

The tools Ivan Rimbaud uses every day at Halte Nuisibles run on his own servers. They do not call Claude or OpenAI at runtime. Claude Code helped him build them. It is not called to operate them. The build is assisted. The deployment is independent. That distinction matters when you are projecting your ten-year cost base.

But the build itself is not free, and the build-time price is not frozen. Claude Code, Cursor, and Codex are still in early pricing mode. Today's price is a go-to-market price, not a mature one. A CEO who replaces ten SaaS subscriptions with heavy reliance on one build tool has traded ten small risks for one bigger one. If that tool doubles or triples its price, the math changes. Not fatally, but not negligibly.

The hedge is obvious. Build with whichever tool is cheapest this quarter. Keep the option to swap. MCP and open-source alternatives are what make that option real.


The question

What is the first subscription you would cancel if you had to rebuild it in 48 hours? And why have you not?

Ivan Rimbaud already answered that question for his entire stack. Figma's board is answering it under market pressure, in public. Your turn is coming whether you run it deliberately or let the market run it for you.


Sources



Commentaires

bottom of page