Cloud Cost per User: Benchmarks by Company Size

Cloud cost per user = total monthly infrastructure spend divided by monthly active users. It is the single most useful unit economics metric for SaaS companies, and most teams are not tracking it. The formula is:

Total monthly cloud spend means your complete infrastructure bill — compute, storage, database, networking, CDN, monitoring, and any managed services. Do not subtract credits or discounts you won't have forever. Use the number you actually pay.

Monthly active users should be MAU, not registered accounts. Someone who signed up two years ago and never returned does not consume infrastructure proportional to an active user. Use the same MAU definition your product team tracks — consistency matters more than precision here.

A startup spending $1,800/month with 3,000 MAU has a cost per user of $0.60. Healthy for the 1K tier. If the same $1,800 bill comes back next month but MAU grew to 6,000, cost per user dropped to $0.30 — a good sign. If MAU stayed flat and the bill jumped to $2,400, cost per user rose to $0.80 — investigate immediately.

Track this number monthly alongside MAU and total spend. The trend matters as much as the absolute value. A declining number means your architecture is working. A flat or rising number means it is not.

At 1,000 MAU, typical cloud cost per user is $0.50–$2.00. At 100,000 MAU, it drops to $0.05–$0.30 — a 5–10x improvement driven almost entirely by fixed costs spreading across more users (SpendArk benchmark data, 2025–2026). Infrastructure has a large fixed-cost component. A database, a load balancer, and a monitoring stack cost roughly the same whether 100 or 10,000 users are hitting them. Here are realistic benchmarks at each tier with concrete examples of what the underlying bills actually look like.

1,000 MAU: $0.50 – $2.00 per user ($500 – $2,000/mo)

At 1,000 MAU you are paying for a baseline production stack that does not scale below a certain floor. A minimal bill: one t3.small EC2 ($15/mo), RDS db.t3.micro PostgreSQL ($25/mo), S3 + CloudFront ($8/mo), a small Redis cache ($16/mo). Add monitoring, a managed queue, and data transfer and you land around $90/mo — that is $0.09/user. But most teams at 1K MAU run staging environments, have multi-AZ databases, and use Lambda. A realistic full-stack bill is $700–$1,200/mo, putting you at $0.70–$1.20/user.

Products above $2.00/user at 1K MAU are usually running infrastructure designed for a much larger scale: overprovisioned databases, always-on workers, or a Kubernetes cluster that is not justified yet. At this tier, managed services and right-sizing are your highest-impact cost actions.

10,000 MAU: $0.20 – $1.00 per user ($2,000 – $10,000/mo)

By 10K MAU, the same fixed infrastructure that cost $800/mo is shared across ten times the users. A typical bill: t3.medium app server with auto-scaling ($45/mo), RDS db.t3.small with read replica ($95/mo), ElastiCache cache.t3.small ($35/mo), CloudFront + S3 ($22/mo), CloudWatch + logging ($18/mo), load balancer ($18/mo). Total around $233/mo baseline plus traffic-driven costs. A well-run 10K MAU product lands at $2,000–$4,000/mo total, or $0.20–$0.40/user.

Products at $0.70–$1.00/user at 10K MAU are often running infrastructure inherited from a smaller scale: oversized RDS instances, missing caching layers, or data pipelines processing every event synchronously. This is the tier where the ROI on optimization first justifies the engineering time.

100,000 MAU: $0.05 – $0.30 per user ($5,000 – $30,000/mo)

At 100K MAU, economies of scale compound. You are likely on Savings Plans or Reserved Instances (1-year no-upfront saves ~30% on EC2), your CDN absorbs the majority of static traffic, and read replicas are handling bulk query load. A well-optimized 100K MAU SaaS runs $8,000–$15,000/mo: multiple t3.large or c5.xlarge instances ($200–$400/mo), RDS db.r6g.large with replicas ($350–$500/mo), ElastiCache ($80/mo), CloudFront ($100–$200/mo), and monitoring ($150–$300/mo). That works out to $0.08–$0.15/user.

At $0.25–$0.30/user at 100K MAU, you are either running a media-heavy or data-heavy product (see workload types below) or carrying architectural debt. This is the tier where a focused engineering sprint on cost optimization routinely pays back 3–5x in the first year.

1,000,000 MAU: $0.01 – $0.10 per user ($10,000 – $100,000/mo)

At 1M MAU, fixed overhead is fully amortized. Reserved Instance and Committed Use coverage is deep. Your team has run multiple optimization passes. A typical 1M MAU product spends $30,000–$60,000/mo: a cluster of c5.xlarge or m6i.xlarge compute ($3,000–$6,000/mo), RDS Aurora multi-writer ($2,000–$5,000/mo), significant CDN spend ($1,000–$3,000/mo), and observability tooling ($1,500–$3,000/mo). That puts cost per user at $0.03–$0.06 for a lean product.

At $0.08–$0.10/user at 1M MAU, the product is likely media-heavy, processes large datasets per user, or runs ML inference per request. Pure API-driven SaaS products with sound architecture should reach $0.02–$0.04/user at this scale.

Moving from 1K to 10K MAU typically cuts cost per user by 60–70% — driven almost entirely by fixed infrastructure costs spreading across more users, not by optimization work (SpendArk benchmark data, 2025). The decline is predictable. It has three compounding drivers, and understanding them helps you forecast where your unit economics will land.

1. Fixed costs amortize across more users

Every production stack has a fixed cost floor: a database instance, a load balancer, minimum compute, monitoring infrastructure, and at least one staging environment. That floor might be $400–$800/mo regardless of whether you have 500 or 5,000 users. When MAU grows 10x without an architecture change, that fixed cost spreads across 10x the users. The variable cost per user — compute cycles, storage growth, egress — is much smaller than most teams expect. For a typical CRUD SaaS app, it is often $0.02–$0.05 per user per month.

2. Commitment pricing unlocks at meaningful spend

AWS, Azure, and GCP all offer significant discounts for committed usage — but those discounts only make financial sense once you have a stable baseline to commit against. AWS 1-year Savings Plans save 30–40% on EC2 and Lambda. 3-year Reserved Instances save up to 72%. At $200/mo in compute, committing is not worth the flexibility tradeoff. At $2,000/mo, it saves $600–$800/mo. At $10,000/mo, commitment pricing alone cuts your bill by $3,000–$4,000/mo — roughly $0.03–$0.04/user at 100K MAU.

3. Optimization ROI justifies engineering time

At 1K MAU, two engineering days on infrastructure optimization might save $50/mo. At 100K MAU, the same two days might save $2,000/mo. The work is proportionally the same — rightsizing instances, adding a caching layer, tuning database queries. The dollar return scales with your spend. Well-run companies naturally invest more in optimization as they grow, compounding the efficiency gains. For industry benchmark comparisons, see the cloud cost benchmark 2026 report.

At scale, you can justify architectural patterns not practical earlier: CDN caching strategies that eliminate origin requests entirely, database connection pooling that reduces RDS compute requirements, async processing that smooths traffic spikes and reduces peak instance sizes. These patterns each shave $0.01–$0.03/user at the 100K–1M tier.

At 100K MAU, a well-optimized API product spends $0.06–$0.10/user. A media-heavy product at the same scale spends $0.25–$0.40/user — a 3–5x difference driven entirely by workload type, not by inefficiency (SpendArk benchmark data, 2025). The benchmarks above apply to a typical API-driven SaaS product. Here are the three main workload categories and how they affect cost per user.

API-heavy: $0.05 – $0.20/user

Products that primarily serve JSON via REST or GraphQL — project management tools, CRMs, productivity apps, developer tools — have the most favorable cost-per-user economics. Main costs are compute and database. Egress is low. Storage grows slowly. CDN caching is highly effective for static assets. At 100K MAU, a well-built API product lands at $0.06–$0.10/user. A project management SaaS at 100K MAU on AWS typically spends $10,000–$15,000/mo ($0.10–$0.15/user). For broader context on lifecycle economics, see cloud cost management for SaaS startups.

Media-heavy: $0.15 – $0.50/user

Products that store, process, or serve user-generated images, video, or large files pay a steep premium. S3 storage costs ($0.023/GB/mo) compound as user base grows. CloudFront egress ($0.085/GB) adds up fast when users download videos or large files. Transcoding and image processing add compute costs per upload. At 100K MAU, a photo-sharing or video platform commonly spends $25,000–$40,000/mo ($0.25–$0.40/user). Key levers: aggressive CDN caching, S3 Intelligent-Tiering, and processing media asynchronously at off-peak hours.

Data-heavy: $0.10 – $0.35/user

Analytics platforms, ML-powered products, and applications that run complex queries per user pay more because computation scales with activity depth, not just user count. A user running a complex report triggers far more database CPU than a user reading a dashboard. At 100K MAU, a data analytics SaaS might spend $15,000–$30,000/mo ($0.15–$0.30/user) — biggest cost drivers being database compute (RDS or Aurora CPU) and data transfer between services. Primary lever: query optimization and result caching to eliminate redundant computation.

At 10K MAU spending $0.90/user on an API product, you are paying 2–3x what a lean competitor at the same scale spends — and that gap compounds. At 100K MAU it becomes an extra $75,000/mo in infrastructure costs your competition does not carry. Benchmarks are a starting point. The more important question is: what does a high cost per user actually signal?

You're above the range for your tier and workload type

If your cost per user is consistently above the high end of the range for your MAU tier and workload category, something structural needs attention. Not a tweak — a fix. The most common structural causes are an oversized database, a missing caching layer, or compute that does not scale down after traffic peaks.

Cost per user is flat or increasing as you grow

In a healthy product, cost per user should fall as MAU grows. Flat or increasing means variable costs are growing faster than your user base. Common causes: each new user triggers disproportionate compute or storage (missing caching, inefficient queries), infrastructure scales up but not down (auto-scaling without a floor does this), or architectural complexity is being added faster than users are being added.

Signals to investigate

Most teams that start tracking cost per user discover they have been above benchmark for months without knowing it. The metric is only useful if tracked consistently. Here is a simple three-step system that works at any company size.

Step 1: Get your total monthly cloud spend

Pull your total bill from your cloud provider's billing console at the same time each month — e.g., the 1st of every month for the prior month. Include all services: compute, storage, database, networking, third-party managed services billed through the provider. If you use multiple cloud providers, sum them. Do not subtract credits that will not always be there.

Step 2: Get your MAU for the same period

Use the MAU your product team tracks. Be consistent: the same definition every month. If your product does not yet track MAU, use weekly active users × 4 as an approximation. Do not use registered accounts — inactive accounts do not consume infrastructure in proportion to active users.

Step 3: Divide and log the result

Cost per user = (Total spend) ÷ (MAU). Log it in a spreadsheet or internal metrics dashboard alongside total spend and MAU. The trend line matters more than any single month. A gradual decline is the target.

Estimating cost per user with SpendArk

Manual billing exports work fine at small scale. When you're planning ahead or comparing providers, though, it helps to model the numbers first. SpendArk's free cloud cost calculator lets you estimate total monthly spend for an AWS, Azure, or GCP setup and compare providers side by side — divide the result by your monthly active users and you have a projected cost per user before you deploy. It's free and needs no account.