CPU Markup Collectors
Dave Morin@davemorin
−8.6%
since Sep 30
The idea
I keep noticing that plain CPU compute, not GPUs, is what actually drives AI app costs, yet the market still prices everything as a GPU story. This take owns the businesses collecting that CPU spend at a markup: pure CPU clouds like Cloudflare, DigitalOcean and Akamai, plus usage-based resellers like Snowflake, MongoDB and Datadog, while deliberately skipping the hyperscalers and GPU names. It holds about 10% in cash to add on dips, since these mid-cap names can swing hard. I'd rethink it if CPU pricing power fades or bills start tracking GPU costs instead.
What it owns
- These companies sell plain CPU cores and edge compute at a markup, with no meaningful GPU business attached. · DigitalOceanIt sells straightforward CPU cores by the hour to developers, a pure version of compute spend without owning the GPU story., CloudflareIts Workers product sells plain CPU time at the edge, exactly the compute AI apps burn money on, without exposure to GPU pricing., Akamai TechnologiesIts Linode business sells plain CPU compute and edge capacity, collecting the markup on ordinary cores without GPU entanglement., FastlyIts edge platform runs customer code on CPUs close to users, so more AI app traffic means more CPU-hours sold at a markup., GoDaddyIt is the mass-market end of the take, millions of small customers paying marked-up prices for plain hosted CPU capacity., Rackspace TechnologyIt resells managed CPU capacity, and if plain compute demand rises the way the take suggests, its recovery gets a tailwind.47%
- Their bills are metered in compute units, so revenue climbs with the same CPU usage that's driving costs. · SnowflakeIt bills customers in compute credits, which are marked-up CPU-hours by another name, so consumption revenue growth tracks AI app workloads., MongoDBIts Atlas service charges by usage, so the CPU spend AI apps rack up on database queries lands directly in revenue., DatadogIt charges per host and per volume of data, so every extra CPU an AI app spins up becomes another billable unit., ElasticIt resells CPU-heavy search and analytics compute, the exact kind of workload AI apps lean on and drive higher consumption., DynatraceIt bills by host units, so its revenue scales with how many CPUs customers run, the same spend outpacing GPU costs.43%
- Cash10%