Quick Answer
Cloud computing cost is what a provider charges for on-demand compute, storage, network, and AI capacity, calculated as services x unit price x usage volume. There is no standard rate. Bills run from a few thousand dollars a month at seed-stage startups to more than $110 million a year at scale, and worldwide public cloud spending passes $870 billion in 2026.
When Figma filed to go public in July 2025, one line in its S-1 got more attention than the revenue chart. The company was spending roughly $300,000 every day on AWS, and it had just renewed a five-year commitment to spend at least $545 million more.
That works out to about 12% of Figma’s $821 million in revenue, supporting 13 million monthly active users at a 91% gross margin. The internet gasped. Figma’s CFO, presumably, did not, because the company signed that commitment on purpose and its margins can carry it. The part worth gasping at is different: it took an SEC filing for anyone outside the company to see the number.
Cloud spend is now a board-level line item growing 21% a year, and AI is rewriting what’s inside it. Here’s what cloud computing actually costs in 2026, what drives the number, and how to calculate and control yours.
How much does cloud computing cost in 2026?
There’s no single price for cloud computing, but there is a single formula. Your total cost equals the services you select, times the unit price you pay, times the volume you use. Everything on your bill, from a $5 storage bucket to Figma’s $100 million+ a year, reduces to those three levers.
Written out, the cloud cost formula looks like this:
Total Cloud Cost (TC) = Services (S) x Unit Price (P) x Volume (V)
The service is which provider and which products you use. The unit price is the rate you pay, negotiated or on demand. The volume is how much you consume, and how that compares to what you planned. Two companies in the same industry can run the same workload at wildly different costs because they pull these levers differently.
Some real-world anchors help calibrate. A seed-stage startup burning through provider credits might spend a few thousand dollars a month. Scaling SaaS companies routinely run five to six figures monthly. Figma’s disclosed spend works out to roughly $100 million a year, and nobody at Figma is panicking.
What you should spend depends on your revenue model. The smartest teams track cloud spend as a percentage of revenue rather than staring at the raw total until it stares back.
Report
Finance needs to prove AI’s return: CloudZero report
260 senior finance leaders (more than half CFOs) told us why the speed of seeing AI spend, not the size of it, separates who pulls ahead on AI from who gets burned.
What changed the cost of cloud computing?
AI changed it, structurally and permanently. For nearly two decades, a cloud bill was compute, storage, and network. In 2026 a fourth pillar arrived: AI workloads, spanning GPU instances, managed AI services, and per-token model APIs. It’s the fastest-growing spend category in the history of cloud computing.
The numbers make the case. Gartner projects worldwide AI-optimized infrastructure spending to grow 96% in 2026, reaching $42 billion, then $66 billion in 2027. The overall public cloud market grows 21.3% this year. AI infrastructure is growing more than four times faster than the overall public cloud market.
In 2026, the AI bill also became permanent. Gartner’s data shows global spending on inference ($23.3 billion) surpassing spending on training ($19 billion) for the first time. Training is a project cost that ends. Inference is a usage cost that scales with every customer interaction, “creating sustained demand for AI-optimized infrastructure,” as Gartner analyst Hardeep Singh puts it.
The consumption model changed too. Model APIs price by the token, so costs now move with prompt length and output size, not instance hours. Pricing pages update constantly across ChatGPT, Claude, and every other major model, and a single H100 GPU instance can cost more per hour than an entire fleet of general-purpose servers.
Finance teams are feeling it. CloudZero’s State of AI Costs research projected average monthly AI spend rising 36% year over year to $85,521 in 2025, yet only 51% of organizations can confidently evaluate the return on that spend. “Enterprise AI budgets are coming under greater scrutiny,” notes Gartner analyst Arunasree Cheparthi. Scrutiny without visibility is just anxiety, and that’s the gap most companies are living in.
How much do companies spend on cloud computing today?
Worldwide, companies will spend about $850 billion on public cloud services in 2026, up 21.3% from 2025, according to Gartner’s latest forecast. Gartner attributes the acceleration directly to AI integration, and projects the market will reach $1.48 trillion by 2029.
The 2026 spend picture, sourced and dated:
| Benchmark | 2026 figure | Source |
|---|---|---|
| Worldwide public cloud end-user spending | ~$850B, up 21.3% | Gartner, 3Q25 forecast update |
| AI-optimized infrastructure spending | $42B, up 96% | Gartner, August 2026 |
| Inference vs. training spend | $23.3B vs. $19B | Gartner, August 2026 |
| AI platforms and models spending | $64B, up 63.4% | Gartner, July 2026 |
| Data center systems spending | $788B+ | Gartner, April 2026 |
| Cloud waste rate | 29% of cloud spend | Flexera 2026 State of the Cloud |
| Average monthly AI spend per org | $85,521 projected, up 36% | CloudZero, 2025 State of AI Costs |
The visibility line deserves a hard look. A decade of tooling has taken the easy savings: practitioners in the FinOps Foundation’s State of FinOps 2026 report diminishing returns, with one describing having “hit the ‘big rocks’ of waste” and now facing “a high volume of smaller opportunities that require more effort to capture.” Then AI arrived on top of that. In the same survey, 98% of respondents now manage AI spend, up from 63% a year earlier and 31% the year before that. The hard part is no longer finding the waste. It is seeing the new spend at all.
CloudZero’s own research explains why: visibility. In our 2026 survey of 260 finance executives, 135 of them CFOs, only 22% could tie AI spend to business outcomes, while 87% said they needed to within the year. Sixty percent admitted they are already spending more on AI than they can justify. You cannot govern spend you cannot see, and most companies still cannot see it.
What drives cloud computing costs?
Four things drive your bill in 2026: compute, storage, network, and AI workloads. Understanding how each is priced is the difference between forecasting your cloud computing cost and getting surprised by it.
- Compute. Processing power, memory, and the instances that deliver them. Providers offer general-purpose, compute-optimized, and memory-optimized instance types, billed by the hour or second. Compute is usually the largest line on the bill, and the one with the most pricing levers attached.
- Storage. Object, block, and file storage, billed per GB per month, with rates that vary by access tier and region. Storage looks cheap per unit and gets expensive at scale, which is why S3 pricing alone sustains an entire discipline of lifecycle management.
- Network. Data transfer out (egress) is the classic budget ambush: your data moves in free and pays rent on the way out. One notable shift: the big three stopped charging data transfer fees for customers migrating off their platforms, with Google moving first in January 2024 and AWS and Microsoft following in March, though Microsoft’s waiver still requires closing the account and claiming credits. Meanwhile IPv4 address fees became a structural cost across all major providers, at roughly $0.005 per hour per address, about $3.65 a month, on every public IP you hold. Small line, every resource, forever.
- AI workloads. The new fourth pillar. GPU and accelerator instances carry premium hourly rates, managed AI platforms bill for training and hosting, and model APIs bill per token. AI spend behaves differently from the other three: it’s spikier, it’s harder to attribute to products and customers, and idle GPU time burns money faster than any idle resource before it.
What are the hidden costs of cloud computing?
The costs that catch teams off guard usually aren’t on the pricing page. Watch for these:
- Data transfer between regions and zones. Cross-region replication and cross-zone chatter accumulate quietly, and rates differ by region pair. Architecture choices you made for resilience show up as line items.
- Data retrieval fees. Archival storage tiers charge to get your data back. Teams that tier aggressively without modeling retrieval patterns end up paying the discount back with interest.
- Support plans and third-party tooling. Provider support tiers are a percentage of your bill, and the surrounding toolchain, from monitoring to security, scales with your footprint.
- Commitment penalties and unused reservations. Committing to capacity you don’t use is prepaid waste. It’s one of the most common failure modes in cloud budgeting, and it’s about to repeat itself at larger scale with GPU reservations.
- Idle AI infrastructure. The newest hidden cost and the most expensive per hour. A GPU cluster idling between training jobs is the priciest way ever invented to do nothing, and it rarely shows up as “waste” in a standard billing view. Our guide to cloud waste covers how to surface it.
How does cloud pricing work across AWS, Azure, and GCP?
All three major providers structure cloud computing pricing the same way: a flexible on-demand rate, discounts for commitment, and deep discounts for interruptible capacity. The names differ, the mechanics don’t, and the price you actually pay depends far more on how you blend the three than on any rate card.
| Pricing model | How it works | Typical discount | AWS / Azure / GCP names |
|---|---|---|---|
| On-demand | Pay per hour or second, no commitment | None (baseline) | On-Demand / Pay-as-you-go / On-demand |
| Committed use | 1 to 3 year usage commitment | Up to 72% | Savings Plans, Reserved Instances / Reservations, Savings Plans / Committed Use Discounts |
| Spot / preemptible | Interruptible capacity the provider can reclaim | Up to 90% | Spot Instances / Spot VMs / Spot VMs |
On-demand is the most flexible and the most expensive per unit; it’s the right default for unpredictable workloads. AWS Savings Plans and their Azure and GCP equivalents trade flexibility for savings on steady workloads. Spot capacity suits fault-tolerant batch jobs, where a 90% discount outweighs the interruption risk.
GCP adds one twist worth knowing: Sustained Use Discounts apply automatically once an instance runs more than 25% of the month, no commitment required. Between that and its early move to scrap exit fees, Google has been the aggressor on pricing structure. For a full breakdown of how the big three stack up, see our AWS vs. Azure vs. GCP comparison.
Mature cloud operations blend all three models: committed coverage for the baseline, on-demand for bursts, spot for batch. Where you set those ratios is one of the most consequential decisions in cloud economics, and it’s a finance decision as much as an engineering one.
The provider landscape itself is also wider than it was. GPU-focused clouds like CoreWeave now compete with the hyperscalers for AI workloads, sometimes at materially different rates. Our guide to cloud service providers maps the full field.
Is cloud computing cost-effective?
Usually, yes, but earned rather than automatic. The cloud converts capital expense into operating expense and trades ownership for elasticity. Whether that trade pays off depends entirely on how well you govern consumption after you sign.
The counterexamples are instructive precisely because they’re rare. 37signals famously walked away from the cloud for owned hardware and later reported saving $2 million in 2024, with another estimated $1.3 million a year from exiting S3. That math works when workloads are stable, predictable, and self-managed by a strong infrastructure team. Most companies fail at least one of those three tests.
Figma made the opposite bet in the same market and its margins didn’t blink: 91% gross margin while spending 12% of revenue on AWS. The difference between the two isn’t that one is right. It’s that both actually know their numbers, which puts them ahead of most of the market.
Cost effectiveness in 2026 comes down to unit economics. Total spend rising isn’t a problem if revenue is rising faster; total spend falling isn’t a win if you cut capacity your customers needed. The question that matters is cost per customer, per feature, per transaction, and increasingly per AI interaction. When you can see those numbers, the cloud’s flexibility is worth its premium. When you can’t, some slice of that 29% waste rate has your name on it.
How do you calculate your cloud computing costs?
Start from workloads, not from price lists. Inventory what you run today, including the machines, databases, and storage behind each application, plus indirect costs like administration. If you’re migrating, that on-premises baseline is your Total Cost of Ownership starting point. If you’re cloud-native, your current bill plays the same role.
Then price the target state. Every provider offers a calculator where you enter instance types, storage volumes, and transfer estimates. They’re accurate on unit prices and optimistic on volume, the way gym memberships are optimistic about January. Whatever usage number you enter, stress-test it against growth.
For instance-level decisions, CloudZero Advisor is a free tool that recommends instance types and sizes by workload profile and budget, with comparisons across EC2, RDS, ElastiCache, and more. It also estimates infrastructure costs from CloudFormation templates before you deploy, which turns cost from a postmortem into a design input.
Two additions to the classic method for 2026. First, price your AI usage separately: token costs scale with adoption, not with infrastructure, so model them against customer growth. Our guide to managing AI spend covers how. Second, budget migration and dual-running costs explicitly; the months where old and new environments overlap are where forecasts usually break.
How do you keep cloud and AI spend under control?
Five levers control cloud and AI spend: rightsize continuously, push committed coverage above 70% of baseline, eliminate idle resources including GPUs, architect for cost, and track cost per unit rather than totals. All five cut waste or improve unit economics. None of them cut the capability your customers pay for.
- Rightsize continuously. Most environments are overprovisioned. Analyze real utilization over a rolling window and downsize instances consistently running below 40%. In our experience this alone trims compute spend by 20% to 30%. Committed coverage is the share of your steady-state usage covered by a one- to three-year commitment rather than paid at on-demand rates. In our experience, above 70% of baseline is the right target for most steady workloads.
- Push committed coverage above 70% of baseline. Cover stable workloads with commitments, let on-demand absorb the bursts. Every point of coverage below that threshold is money left at on-demand rates.
- Eliminate idle resources, especially GPUs. Unattached volumes, orphaned snapshots, and idle load balancers are the classics. Idle GPU time is the new heavyweight. Automate detection for both.
- Architect for cost. Cloud architecture decisions are pricing decisions: storage tiering, serverless for event-driven work, spot for batch, and scaling strategy all set your cost floor before anyone opens a billing console.
- Track cost per unit, not just totals. Spend growth is healthy when the business grows faster. The only way to know is measuring cost per customer, per feature, and per AI interaction.
Each of those levers is a discipline of its own, and our guide to cloud cost management tools compares the platforms that operationalize them.
CloudZero approaches this from the visibility side. The platform ingests spend from AWS, Azure, GCP, Kubernetes, Snowflake, Datadog, and AI platforms including OpenAI and Anthropic, then maps every dollar, tagged or not, to the people, products, and customers driving it.
Finance gets cost per customer and COGS they can defend; engineering gets cost signals inside their workflow. When AI spend jumps by a third in a year, that visibility is the difference between explaining the number and apologizing for it.
For most software companies, cloud is now one of the largest operating costs on the books. It’s growing 21% a year, and AI is pushing from below. The companies that win aren’t the ones that spend least. They’re the ones that can see what every dollar returns.
Schedule a demo to see your cloud and AI spend mapped to the products and customers driving it, take the self-guided tour, or start with a free cloud cost assessment.
Cloud computing cost FAQs
