Air Pocket: Heavy-User Tools
Jiayao Yu@jiayao
−12.6%
since Oct 13
The idea
I think the AI trade is priced for billions of heavy users when the real number is closer to an amount, mostly developers running agents that ship code and query databases constantly. So I'm skipping the model makers and buying the usage-priced infrastructure those agents actually burn through: databases, monitoring, security, build pipelines and code hosting, with just a small slice of the big clouds since they're the ones spending the most and would feel a slowdown first.
What it owns
- Every new agent and workload needs protection and delivery, and these platforms charge by the traffic and compute they inspect or serve. · CloudflareIts Workers platform lets developers run code at the edge and pay per request, which suits agents making millions of small calls., CrowdStrikeIt protects computers and cloud workloads, and agents spin up far more of those than human teams do., ZscalerIt sits in the path of company network traffic and inspects it for threats, so more agents talking to more services means more to secure., OktaIt manages logins and permissions, and is developing billing for non-human agent identities the way it does for people.23%
- Agents read and write data far more often than people do, so these databases bill for every query and every gigabyte stored. · SnowflakeIt charges for compute consumed rather than seats, so every query an agent fires off adds to the bill., MongoDBIts cloud database bills by storage and query volume, and agents hit databases far more often than people do., ElasticIts search platform is commonly used by agents to look things up, billed on its cloud by usage.21%
- Agent-written code running around the clock generates constant server, log and trace data, all of which these platforms bill to watch. · DatadogIt bills by the volume of servers, logs and traces it watches, and agent-written code running constantly generates far more of all three., DynatraceIt monitors enterprise software on a pay-as-you-consume plan, so more agent workloads in production means more billable activity.15%
- Agents open far more pull requests and trigger far more build runs than human developers, and these platforms charge for that activity. · GitLabIt hosts code repositories and runs build pipelines that agents trigger every time they ship a change., JFrogIt stores and transfers the packages and build files that every software release depends on, charging partly by usage.14%
- They sell compute and storage by consumption to agent-heavy developers, but they also carry data center build-out risk if demand softens. · MicrosoftIt owns the GitHub platform where much heavy agent coding happens and sells Azure compute by usage., AlphabetIts cloud platform sells compute and AI models by usage to the same developers building agent systems., AmazonIts AWS cloud bills by usage and hosts many developer workloads that agents will run on.10%
- These small usage-billed clouds and APIs charge by the request or message, making them natural homes for agent workloads. · DigitalOceanIt offers usage-billed compute and storage designed for developers building and running small AI systems., TwilioIt charges per message, call and API request, so agents that communicate with users on their own create billable volume.9%
- Cash8%