AIOps Pricing Models Explained

Per-host, per-user, per-event, per-GB, and the "contact us" wall, what each model rewards, what it punishes, and a 3-year TCO comparison at three deal sizes.

Why pricing model matters more than sticker price

The sticker price is what you sign for in year one. The pricing model is what you'll be fighting about at renewal. A platform that's $40 per host in year one becomes $70 per host in year three because growth was 35% and the volume tier reset. The model, not the price, is what compounds.

The four production AIOps pricing models in 2026 are per-host, per-user, per-event, and per-GB. Most vendors blend two of them. The fifth model is "contact us," which isn't a model, it's a negotiation tactic. Each rewards a different growth pattern. Pick the model that aligns with how your environment will grow, and you'll be the rare buyer who has a smaller bill at renewal.

Per-host (Datadog, New Relic)

Per-host pricing charges by the count of monitored servers, containers, or pods. Typical 2026 list price: $30-$45 per host per month for infrastructure monitoring, $50-$80 per host for full APM. Volume discounts kick in around 200 hosts.

What it rewards. Stable host counts. If your infrastructure is steady-state and you scale on workload (not on host count), per-host pricing is predictable.

What it punishes. Containerised, ephemeral environments. A single Kubernetes node can run 30 pods, each counted as a host depending on the vendor's definition. Auto-scaling groups that briefly spawn 200 hosts during a traffic peak get billed for the peak. Datadog's container fee, $1 per container above the included quota, is where most surprise renewals come from. A 400-host customer at $40/host/month plus 12,000 containers at $1 each pays $192k/year in monitoring before adding logs or APM.

The right buyer. Companies with 100-500 stable hosts, low growth, predictable workloads. The wrong buyer is anyone running serverless, lambda, or aggressive Kubernetes auto-scaling, the host count fluctuates wildly and the bill follows.

Per-user (PagerDuty, Incident.io)

Per-user pricing charges per engineer who can receive pages or run incidents. Typical 2026 list price: $20-$50 per user per month at the operational tier, $60-$120 at the enterprise tier with full features.

What it rewards. Small, expert teams. A 12-person SRE team at PagerDuty's $41/user/month enterprise plan is roughly $6k/year, predictable and cheap.

What it punishes. Wide on-call coverage. The moment you give every engineer a pager, modern incident management practice, the bill scales with headcount. A 200-person engineering org at $41/user/month is $98k/year. A 600-person org is $295k. PagerDuty's pricing is the reason Incident.io and Rootly built per-incident or hybrid models.

The right buyer. Teams of 10-50 engineers with a centralised on-call rotation. Wrong buyer: companies practicing "you build it, you run it" with hundreds of engineers in the on-call population.

Per-event (Splunk, Sumo)

Per-event pricing charges per log line, alert, or metric data point. Typical 2026 unit price: $0.0005 to $0.002 per event after volume tiers.

What it rewards. Quiet, high-quality environments. If your alert volume is 50k/month, per-event pricing is cheap.

What it punishes. Noisy systems and chatty applications. A single misbehaving microservice that emits 20M debug log lines a day during an incident can add $400-$1,200 to a single day's bill. The compounding problem is that growth in alert volume usually correlates with system complexity, not with revenue, so the bill grows faster than the business.

The right buyer. Quiet, mature environments with disciplined alert hygiene. Wrong buyer: teams in active scale-up mode where every new service adds noise.

Per-GB ingest (Splunk, ELK)

Per-GB charges by the volume of telemetry ingested. Typical 2026 price: $0.40 to $2.50 per GB depending on retention and compression. Splunk Enterprise sits at the high end ($1.60-$2.50/GB at most contract sizes); cheaper ingest tiers exist for log archive but the analytics fees stack on top.

What it rewards. Aggressive log filtering. Teams that drop debug-level logs at the agent and only ship structured production traffic can get extraordinary value from per-GB pricing.

What it punishes. Verbose logging. A single Java service emitting stack traces on every error can add 200GB/day during an incident. At $1.50/GB, that's $300/day per noisy service. The 3-year TCO surprise on per-GB platforms is consistently 1.8x to 2.4x the year-one estimate.

The right buyer. Teams with mature log discipline and willingness to filter aggressively. Wrong buyer: anyone planning to "log everything and figure it out later."

The "contact us" trap

Most enterprise AIOps vendors hide pricing behind a "contact us" page. The justification is "every customer is different." The actual reason is price discrimination, they want to know your budget before they quote.

The defence is simple. Always ask for the pricing model in writing on the first call, before any discovery. "What is your unit price for the smallest tier you sell? What's the unit price at our scale? What overages am I exposed to?" If a vendor can't answer these in 15 minutes, they're optimising for negotiation leverage, not customer success.

The second defence is the 90-day quote validity. Whatever number lands in writing must hold through the procurement cycle. Vendors who renegotiate after 30 days are training you to never trust their numbers.

Three-year TCO worked examples

Small (50 engineers, 80 hosts, 600GB/day logs). Per-host platform at $45/host = $43k/year, growing 25% YoY = $43k + $54k + $67k = $164k three-year. Per-user platform at $35/user = $21k/year flat, $63k three-year. Per-GB at $1.50/GB = $329k/year, $1.07M three-year. Mid-market companies get crushed by per-GB unless they aggressively filter.

Medium (200 engineers, 400 hosts, 3TB/day logs). Per-host at $40/host = $192k/year, $635k three-year with 25% growth. Per-user at $40/user = $96k/year, $317k three-year. Per-GB at $1.20/GB = $1.31M/year, $4.34M three-year. The per-GB model is where a CFO's eyes get wide.

Large (800 engineers, 1,500 hosts, 12TB/day logs). Per-host at $35/host = $630k/year, $2.1M three-year. Per-user at $45/user = $432k/year, $1.43M three-year. Per-GB at $0.95/GB (negotiated) = $4.16M/year, $13.7M three-year. At enterprise scale, the pricing model swings the contract by a factor of ten.

The lesson: pricing model selection is the single biggest commercial decision in an AIOps purchase. Even small teams should run the three-year TCO before they sign, the numbers are usually shocking enough to change the shortlist.