Payroll Turns Into Compute
The idea
I believe startups now hit real revenue with far fewer people, skipping junior hires and routing that money into cloud and compute instead. So I own both sides of that shift in equal halves: the hyperscalers and chipmakers that power the compute, and usage-billed software that earns more as small teams run more of it, while I avoid anything priced per seat. It is spread across twenty names, mixing steady cloud owners with smaller growth bets, with a modest cash reserve since this plays out over years. If teams start growing again or billing shifts back to seats, the thesis breaks.
What it owns
- Monitoring, network and messaging tools bill by traffic and volume, so their revenue grows with compute instead of headcount. · Cloudflare, Inc.It carries web and AI traffic and its developer tools charge by usage, so more automated traffic from small teams means more revenue., Datadog, Inc.It bills by the servers and data it monitors, not by how many people use it, so a small team running lots of compute pays more., Twilio Inc.It charges per message and per call, so an AI agent texting customers for a three-person startup pays it the same as a big call center would., Akamai Technologies, Inc.It earns from moving web traffic and is building out a cloud for running AI close to users, both billed by usage rather than seats., Dynatrace, Inc.It prices monitoring by consumption, so it gets paid for how much software runs, not how many engineers a customer employs.27%
- The biggest clouds collect most of the compute bills startups pay instead of junior salaries. · Amazon.com, Inc.When a five-person startup skips junior engineers and leans on AI tools, a big share of that saved salary ends up as an AWS bill., Alphabet Inc.It rents out both regular servers and its own TPU chips, so it collects on the compute bills that small teams run up instead of payroll., Microsoft CorporationAzure is where small teams rent compute that replaces people they didn't hire, and it also sells the AI coding tools that make skipping junior hires possible., Oracle CorporationIt has turned itself into a landlord for AI compute, with a big backlog of contracts to rent out servers and power.21%
- Databases and data platforms charge by how much data is stored and processed, not by how many people log in. · MongoDB, Inc.It charges developers by database usage rather than per seat, so it earns more as small teams ship software with fewer engineers., Snowflake Inc.It charges for the computing customers use, not per employee, so a tiny team crunching lots of data pays more., Elastic N.V.It sells search and retrieval that AI apps use to find data, and its cloud product charges by what customers use.18%
- Every extra hour of compute a small team buys runs on chips from these companies. · NVIDIA CorporationAlmost every dollar a small team spends on AI compute runs through its chips somewhere down the line, so bigger compute bills mean more chips bought., Broadcom Inc.It builds the custom AI chips and networking gear that big clouds and AI labs use to run the models small teams pay for., Advanced Micro Devices, Inc.It is the main alternative to Nvidia for AI chips, so it gains as compute spending grows and buyers want a second supplier.15%
- Smaller clouds sell compute straight to lean teams and AI companies, so they grow faster when payroll shifts to servers. · DigitalOcean Holdings, Inc.Its whole business is selling simple cloud servers to small teams and solo developers, exactly the companies running with way fewer people., CoreWeave, Inc.It rents out GPU servers by the hour, which is a direct play on rising compute bills from lean teams shifting payroll to infrastructure., Nebius Group N.V.It is a GPU cloud built for AI teams, the kind of lean startups that spend on compute instead of headcount.9%
- Cash10%