Non-GPU Cloud Infrastructure Sees Rising Demand From Production AI Agents
Summary
DigitalOcean and Dynatrace say production AI agents lift CPU, storage, database and log usage: 90% of DOCN AI revenue uses core cloud; DT AI cohorts consume 2x.
DigitalOcean (DOCN) and Dynatrace (DT) both said on their quarterly earnings calls on November 3, 2026 that AI agents are driving non-GPU cloud demand: as customers move agents into production, new usage is landing on CPU compute, storage, databases and log monitoring, products that are usually not counted as AI [1][2].
Why AI agents consume ordinary cloud infrastructure
An AI agent uses a whole stack of ordinary cloud resources that a single model call does not. Until recently, most companies used AI by sending a prompt to a large model and getting an answer back, so the spending went mainly to model tokens and the GPUs behind them. An agent writes its own code, calls outside tools and keeps context, which makes it closer to a continuously running program. DigitalOcean CEO Paddy Srinivasan said that over the last 90 to 100 days customers have started deploying production agent workloads, and that beyond large volumes of tokens these need a secure runtime, often called a sandbox, which isolates the code an agent executes from other systems [1]. DigitalOcean's managed agents offering includes CPU-based secure sandboxes, context and memory services, and high-performance storage and databases [1].
Running agents also generates large volumes of operational records, which lifts usage of monitoring software. Logs, traces and similar runtime data are collectively called telemetry, and companies use observability software to collect and analyze it to check that systems are healthy. Dynatrace CEO Rick McConnell explained that agents and AI observability workloads produce more logs, traces and telemetry, which drives additional usage and draws down customers' prepaid platform commitments faster [2]. As long as agents keep spreading, usage-priced compute, database and monitoring services face the same layering of demand.
Usage is rising with AI customers at both the compute and monitoring layers
Dynatrace's figures show AI customer cohorts consuming its platform at more than twice the rate of other customers. Overall platform consumption is growing in excess of 20%, log consumption is growing well over 100%, and infrastructure monitoring has passed $500 million in annualized consumption, growing about 30% [2]. McConnell added that many large AI-native companies and frontier model developers mainly use infrastructure monitoring [2].
At DigitalOcean, AI revenue is now tied to ordinary cloud products. In the third quarter, 90% of AI customer revenue came from customers that also use its core cloud products [1]. CFO Matt Steinfort said the core cloud is accelerating and basic virtual servers are expanding, and that many new customers deploying agent frameworks such as OpenClaw and Hermes do not identify themselves as AI customers [1]. One company sells compute and the other sells monitoring, and both see AI customers carrying usage onto non-GPU products.
AI budgets shift toward usage-priced infrastructure
As agents spread, part of AI spending moves to products whose price lists carry no AI label. Steinfort said it is getting harder to tell who is an AI customer and who is not [1]. Usage-priced vendors are therefore likely to see the change first in consumption metrics, while contract value would only reflect it when customers renew.
This view has two limits for now. McConnell acknowledged heavy budget scrutiny as customers try to offset large token spend, though he said Dynatrace saw no evidence of deals being delayed because of it [2]. On DigitalOcean's call, an analyst noted that annual recurring revenue (ARR) excluding AI is growing only about 10% to 11% [1], so the claim that agents are lifting the core cloud rests mainly on management's attribution. Points to track: whether DigitalOcean's non-AI ARR growth picks up, and whether Dynatrace's fast-growing log consumption converts into larger contract values at renewal.
Companies exposed to this shift
- Datadog (DDOG): Also a usage-priced observability software company. Its CEO said on the August 2026 earnings call that MCP tool calls (an interface agents use to call outside tools) had grown more than 22x compared with Q4 2025 [3]. It sits in the same telemetry layer as Dynatrace: more agents mean more data to collect and analyze.
- Cloudflare (NET): Sells per-use agent sandboxes and containers on its Workers developer platform, the kind of CPU runtime DigitalOcean says agents need. So far its exposure shows up in developer counts rather than in revenue figures [4].
- AMD (AMD): Supplies server CPUs to cloud providers, and management names agentic workloads as one growth direction for server CPUs [5]. If clouds add CPU capacity for agent sandboxes, demand would pass to this layer, but suppliers have not tied related revenue to specific cloud platforms.
Sources
[1] Drillr · DigitalOcean (DOCN) · 2026-11-03 · Earnings call
In Q3, 90% of AI customer revenue came from customers that also consume our core cloud primitives. They're not looking for just GPU capacity. They're building and deploying production agent applications on the whole platform.
[2] Drillr · Dynatrace (DT) · 2026-11-03 · Earnings call
[3] Drillr · Datadog (DDOG) · 2026-08-06 · Earnings call
[4] Drillr · Cloudflare (NET) · 2026-08-06 · Earnings call
[5] Drillr · AMD (AMD) · 2026-08-04 · Earnings call
This article only helps surface industry changes and companies that may be overlooked. It is not a stock recommendation.