Consumption Beats Seats
Clay Norris@claynorris
−8.7%
since Oct 1
The idea
I believe AI has broken the old link between software usage and headcount, and companies that charge by consumption rather than by seat are the ones set up to benefit as agents multiply queries, API calls, and compute hours. This basket owns that metered layer directly: data clouds billed per query, observability billed per gigabyte, API and edge rails, GPU compute billed by the hour, and a smaller slice of companies already priced on outcomes.
What it owns
- GPU and cloud compute sold by the hour scales directly with machine activity, making it the purest consumption-based business model. · CoreWeaveIt sells GPU compute by the hour with no seat-based pricing, so inference and training hours scale directly with agent deployments., OracleIts cloud infrastructure is consumption-billed and ramping on AI workloads, so usage revenue directly tracks agent deployment activity., AmazonAWS pioneered pay-per-use infrastructure and remains the largest metered rail for agent workloads, making it the mainstream way to own this repricing., Nebius GroupIt sells GPU compute by the hour as a challenger to larger cloud operators, providing a second way to own AI compute demand.22%
- Databases and data platforms bill per query and per unit of storage consumed, so agent-driven workloads turn directly into usage-based revenue without requiring more seats. · SnowflakeIt bills per query and per credit consumed, so agent workloads multiply queries against warehoused data without requiring more users., MongoDBIts Atlas database is metered by reads, writes, and vector searches, making it the natural database for AI agents billed purely on usage., ElasticIt bills search and vector retrieval on consumption, which agents perform constantly, so its usage-based revenue grows with agent deployments.21%
- Networks and APIs charge per request or per unit of traffic, so higher machine activity directly increases the bill regardless of headcount. · CloudflareIt provides metered edge infrastructure for inference and API calls, charging by request volume, so agent traffic directly increases the bill., TwilioIt charges per message and per API call rather than per seat, making it natively consumption-based for agent-driven communications volume., DigitalOceanIt bills compute by the hour to small teams and builders shipping agent applications, pricing tied to infrastructure usage rather than user count., FastlyIt provides usage-priced edge delivery and compute, offering an asymmetric way to own edge traffic growth if agent workloads drive higher volumes.20%
- Agent workloads emit far more logs and traces than human activity, and these platforms charge by the gigabyte ingested, so machine activity directly multiplies revenue. · DatadogIt charges by the gigabyte of telemetry ingested, and agents doing work generate far more logs and traces than people do., DynatraceIt moved to consumption pricing for monitored workloads, so rising machine activity converts to revenue independent of headcount.14%
- These companies charge per contact, per sale, or per asset rather than per employee, so revenue decouples from headcount and ties instead to activity volume. · ShopifyIt takes a percentage of merchant sales, so revenue follows transaction volume rather than merchant headcount., KlaviyoIt prices per contact and per message sent rather than per marketer, so revenue scales with campaign activity instead of team size., SamsaraIt charges per connected asset like a truck or machine rather than per employee, pricing tied to the thing doing the work.13%
- Cash10%