The True Cost of Observability: Why Per-GB Pricing Is Broken

Your observability bill just hit five figures. Again. You're not ingesting significantly more data than last month, yet the invoice keeps climbing. You pull up your platform's usage dashboard and find yourself in a familiar frustration: you can't tell whether you're paying for value or excess.

This is the reality for thousands of DevOps teams and SREs using traditional observability platforms. The per-gigabyte pricing model—once considered the gold standard for "pay for what you use"—has become a financial liability that obscures true costs, encourages data hoarding by vendors, and creates perverse incentives that work against your operational needs.

The problem isn't observability itself. Visibility into your systems is essential. The problem is the economic model that's been normalized across the industry. Let's examine why observability cost per GB is fundamentally broken and what modern alternatives exist.

The Hidden Tax of Per-GB Observability Pricing

Per-gigabyte pricing seems straightforward at first: ingest a gigabyte, pay a fixed rate. Simple math, right? In practice, it's far more complicated—and far more expensive—than it appears.

The Cardinality Trap

One of the most insidious problems with observability cost per GB models is how they interact with high-cardinality data. Every unique combination of tags, labels, or fields creates a distinct time series. A single service with user IDs in its metrics can generate millions of cardinality explosions.

Traditional observability platforms don't charge you fairly for this problem. They charge you based on ingest volume, but the real computational cost—and the actual limitation you hit—is cardinality. A customer ingesting 10 GB of low-cardinality metrics data might cost them $500/month, while another customer ingesting 10 GB of high-cardinality data hits a hard wall at $50,000/month because the platform's index can't handle it.

This means teams start making architectural decisions not based on what's best for their systems, but on what won't bankrupt them. You stop logging user IDs. You strip out request parameters. You sample more aggressively. Each of these decisions degrades your observability.

Sampling Penalties and Data Blindness

To control per-gigabyte costs, many teams implement aggressive sampling strategies. Ingest only 1 in 100 log lines. Keep metrics at 30-second intervals instead of 10. Sample traces at 10%.

The math seems to work: if you sample 99% of your logs, you'll only ingest 1% of the volume, right? Wrong. You've also eliminated your ability to see rare but critical issues. That one error condition affecting 0.1% of requests? You'll never see it with 99% sampling. The one slow query that only happens on Tuesdays? Gone.

According to a 2023 survey by the Cloud Native Computing Foundation, 67% of organizations using sampled observability data reported missing critical incidents in the past year. The cost savings were real, but they came at the expense of actual observability.

The Retention Ratchet

Traditional per-gigabyte platforms make their margin not just on what you ingest, but on how long you keep it. Want 30-day retention instead of 7 days? That's roughly 4x the storage cost. Want 90-day retention for compliance? You're looking at 13x the base cost.

This creates a hidden penalty for teams that need longer retention windows. Financial services companies, healthcare organizations, and anyone dealing with regulatory requirements suddenly face exponential cost multipliers that have nothing to do with their actual data volume or operational needs.

How Industry Incumbents Lock You In

The per-gigabyte model isn't accidental. It's a feature, not a bug—for the vendors. Here's how the economics work:

• Revenue visibility: A $10/GB pricing model with no caps means usage uncertainty translates directly to revenue uncertainty. Venture-backed companies need predictable growth, so they build in guardrails and soft limits that push customers toward higher consumption.
• Vendor lock-in: Once you've ingested terabytes of data into a proprietary system, the cost of leaving—both in terms of data portability and analyst ramp-up—keeps you locked in. You can't easily migrate to a cheaper alternative without losing historical data access.
• Margin expansion: As your business grows and you ingest more data, the per-gigabyte model automatically increases your bill faster than your business grows. A startup doubling in size doubles its observability bill. A mature company with 10x the traffic pays 10x for observability, regardless of whether their operational complexity actually increased proportionally.

The result is an industry where average observability spend is among the fastest-growing cloud costs—often 30-50% year-over-year for growing companies, according to multiple surveys from Gartner and Forrester.

Why Your Current Bill Is Probably Higher Than It Should Be

Let's ground this in concrete numbers. A mid-market SaaS company with 20 microservices, moderate log volume, and standard metric retention might reasonably expect to pay $3,000-5,000/month for observability with a per-gigabyte model.

But in practice, they're likely paying $8,000-15,000 because of:

• Cardinality expansion: Metrics with service names, instance IDs, container IDs, and request types create explosive cardinality. What looks like "manageable" metric volume explodes in cost once cardinality pricing kicks in.
• Infrastructure redundancy: They're running their own log aggregation layer because the observability platform's querying is too slow, adding Elasticsearch costs on top of their observability bill.
• Multiple tools: They're using one platform for logs, another for metrics, a third for traces, because no single platform's pricing model makes sense for all three data types. Each adds its own per-gigabyte multiplier.
• Historical data taxation: They're paying inflated rates on 60-90 day old data because they need it for compliance, even though it's never queried.

The observability cost per GB model creates a situation where you're not just paying for the value you get—you're paying penalties for architectural choices, data diversity, and operational needs that have nothing to do with actual platform utilization.

The BYOO Alternative: Flat Subscription, Owned Infrastructure

An alternative model is gaining traction among forward-thinking observability platforms: BYOO—Bring Your Own Observability. Instead of paying per gigabyte to a vendor, you maintain your own data infrastructure and pay a flat subscription for the intelligence layer on top of it.

LeashStack pioneered this approach. You keep your logs in your own S3 buckets. Your metrics stay in your own time-series database. Your AI-powered querying, alerting, and analysis run on your own AWS Bedrock instance. You pay one predictable monthly subscription—no per-gigabyte surprises, no cardinality penalties, no retention multipliers.

How This Changes the Economics

Under BYOO architecture, the cost structure fundamentally shifts:

• Storage costs are your domain. You negotiate S3 rates directly with AWS or use reserved capacity. You're not paying a vendor markup on storage. A terabyte of S3 storage costs roughly $23/month; a terabyte ingested into a traditional platform costs $300-500/month or more.
• Compute scales with your needs. Whether you're running simple metric checks or complex AI analysis, you control the compute resources. You pay AWS for what you use, not a platform vendor's margin on top of their infrastructure.
• Retention becomes free. Historical data in S3 doesn't cost more than recent data. You can keep years of logs for compliance without penalty. Querying old data might cost slightly more in compute, but you're not paying exponentially higher subscription rates.
• No cardinality hostage situation. High-cardinality metrics don't create hard limits or surprise billing. Your infrastructure scales with your data patterns, not against them.

The Intelligence Layer: Where BYOO Platforms Compete

If storage and compute are commoditized, what justifies the platform cost? Pure intelligence: how effectively the platform helps you understand and act on your data.

This is where LeashStack's feature set differentiates. With data remaining in your infrastructure, the platform focuses entirely on making that data actionable:

Natural Language Log Querying: Instead of learning yet another query syntax, you can ask questions in plain English. "Show me all payment errors in the last hour" gets translated by the AI into a structured query against your logs. No DSL to learn, no query documentation to reference—you describe what you need, and the system figures out how to find it.

AI-Powered Log Summarization: When an incident occurs, you don't manually dig through thousands of error log lines. One click generates a comprehensive incident report, extracting error patterns, affected services, and timeline of events. What used to take 30 minutes now takes 30 seconds.

Semantic Log Search with Embeddings: Find issues by meaning, not just keyword matching. Search for "authentication timeouts" and surface logs about connection resets, timeout exceptions, and auth service errors—all the semantically related problems—without writing OR conditions for every variation.

Predictive Alerting and Statistical Anomaly Detection: Rather than waiting for thresholds to breach, LeashStack uses Z-score, MAD (Median Absolute Deviation), and rate-of-change analysis to detect anomalies before they become incidents. More advanced: predictive alerting forecasts when metrics will breach thresholds, giving you hours of warning instead of minutes of reaction time.

Alert Correlation and AI Incident Analysis: When 50 alerts fire simultaneously, traditional systems leave you to manually correlate them. LeashStack groups related alerts and analyzes their root cause, explaining why they fired together and what the actual problem is.

Configuration as Code and Observability Auto-Discovery

Setup is where many platforms lose teams. LeashStack includes infrastructure auto-discovery that scans your AWS account and generates monitoring configurations automatically. It produces Terraform, CloudFormation, or AWS CLI commands for monitoring gaps you didn't know you had.

Dashboard generation from natural language means describing what you want to monitor—"show me API response times and error rates for each service"—and the AI builds the dashboard. No clicking through UI menus, no manual metric selection. The platform generates Grafana-compatible dashboards, meaning you're not locked into LeashStack's visualization layer.

Concrete Example: Real Cost Comparison

Let's compare two companies on different observability models.

Company A: Traditional Per-GB Platform

• Ingests 500 GB/month of logs, metrics, and traces
• Average per-GB cost: $15
• Monthly data cost: $7,500
• Cardinality overages (metrics exceed platform limits): +$2,000
• 90-day retention multiplier (3x base rate): +$4,500
• Additional Elasticsearch cluster (platform queries too slow): +$1,200/month
• Total: $15,200/month

Company B: BYOO (LeashStack)

• Same 500 GB/month, now stored in S3: ~$12/month
• AWS Bedrock API calls for AI analysis: ~$800/month
• Compute for anomaly detection and search: ~$400/month
• LeashStack flat subscription (teams): $2,000/month
• Total: $3,212/month

Company B pays 79% less while maintaining superior observability—no cardinality limits, no sampling penalties, full historical retention, and more advanced AI capabilities.

At annual scale, that's $143,000 in savings. For a team of 20 engineers, that's roughly $7,000 in observability cost per engineer per year, versus $36,000 on the traditional model.

What to Watch For When Evaluating Alternatives

If you're considering moving away from per-gigabyte pricing, here's what to evaluate:

Data Portability and Lock-In

Ensure your data remains accessible in standard formats. Can you export all logs to standard JSON? Can you retrieve metrics in Prometheus format? A BYOO platform should make this trivial because your data is already in your infrastructure—there's no vendor lock-in by design.

Query Flexibility

Does the platform force you into its query language or UI, or can you use your existing tools? LeashStack accepts Prometheus remote_write, imports Grafana dashboards, and works with Loki. Your existing investments in monitoring should still be useful.

True Flat Pricing with Transparency

Watch for hidden escalation clauses. "Flat pricing" that doubles when you hit certain API call thresholds, or that has separate seats/user pricing, is not actually flat. Legitimate flat pricing should be clear: you pay one number, and it doesn't change based on usage.

Actual Intelligence Features

Cheap storage doesn't help if you can't gain insights from it. Evaluate whether the platform actually helps you find issues faster or if it's just a cheaper dumb storage layer. Natural language querying, anomaly detection, and incident correlation should be core features, not add-ons.

The Path Forward: Reclaiming Observability Economics

The per-gigabyte observability model solved a real problem when storage was expensive and unpredictable. In 2024, with cloud storage commoditized at pennies per gigabyte, the model has become a tax on growth rather than a fair pricing mechanism.

Companies that break free from this model gain two advantages: lower costs and better observability. The cost savings come from eliminating vendor markup on storage and eliminating cardinality penalties. The observability improvements come from focusing on intelligent analysis rather than data volume limitations.

BYOO architectures like LeashStack represent a shift in how observability vendors compete: not on who can ingest the most data cheaply, but on who can extract the most insight from data you control. It's a healthier competitive dynamic for customers.

If your observability bill is growing faster than your business, if you're making architectural compromises to control costs, or if you're running multiple tools because no single platform's pricing model makes sense for all your data—you're experiencing the failure mode of per-gigabyte pricing.

There's a better way. Evaluate alternatives that put you in control of your infrastructure and your costs.