Dev Tools Are the New SaaS
David McDonough@mcd
−4.3%
since Sep 29
The idea
I believe AI is about to make developers dramatically more productive, and the tools they build with will become this cycle's subscription giants, the way Salesforce and Workday were for cloud. So this basket owns both layers: the public names selling AI coding itself, plus the broader plumbing that gets paid regardless of which coding tool wins, the storage, deployment, monitoring, and security layers that all see more usage as AI agents write more code.
What it owns
- Companies selling AI-assisted coding directly to developers are the core bet that dev tools become subscription giants this cycle. · MicrosoftIt owns GitHub and Copilot, the largest seller of AI-assisted coding today and the direct way to hold the bet that dev tools become subscription giants., GitLabIt offers the pure public company play on AI-assisted coding through GitLab Duo, so it moves faster than larger competitors on the same adoption news., AtlassianIt runs Jira and Bitbucket where developer work gets planned and reviewed, so more productive developers means more seats and usage.23%
- AI coding agents consume compute constantly, and cloud platforms hosting them bill for every second of usage. · CloudflareIts Workers platform hosts AI-generated applications and agents that collect per-use fees, creating revenue that grows independent of coding tool preference., CoreWeaveIt rents the GPUs that power the entire AI coding boom, making it the most direct compute infrastructure provider to the ecosystem., DigitalOceanIt is the cloud platform small teams and solo builders use, and AI coding creates far more small teams shipping real software., Nebius GroupIt is a smaller GPU cloud provider, a higher-risk bet that AI compute demand grows fast enough to sustain multiple infrastructure providers.22%
- AI-built applications still require databases and data infrastructure underneath, with usage-based pricing that grows with each new app. · SnowflakeIt charges by usage, so revenue accelerates quickly when AI workloads jump and actually run their data processing through its platform., MongoDBIt is the default database that AI-built applications reach for, so every new app an AI agent writes becomes a potential paying workload., ElasticIt provides the search and retrieval layer that AI applications use to find information in their own data.17%
- Machine-written code breaks in new ways, making production monitoring a non-negotiable expense that scales with code volume. · DatadogIt watches code running in production and charges by volume, so more AI-written code means more to monitor regardless of which coding platform wins., DynatraceIt provides AI-driven monitoring built for systems too complex for humans to observe, so machine-written code still needs to be watched at scale.14%
- Tools that package and ship software collect revenue per unit moved, so ten times more code means ten times more to process. · JFrogIt stores and ships software artifacts on a per-unit basis, so if AI writes ten times more code, someone has to package and move it., PagerDutyIt gets called when code breaks in production, and more code shipped faster means more incidents to manage and coordinate.8%
- Thousands of AI agents acting autonomously need logins and permissions, creating a new security spending category that did not exist before. · OktaIt handles identity and access, and AI agents performing developer work need credentials and permissions just like human employees.6%
- Cash10%