Before CloudZero, there was a budget: $3,000. Erik Peterson was building a cloud-based scanning system at an application security company. His team had to make the architecture fit the money. A few weeks later, the project was finished. The bill was $2,972. Twenty-eight dollars remained, along with a question that would become considerably more valuable: what happens when engineers know the price before they choose the design?
- CloudZero connects infrastructure bills to products, features, customers, and teams.
- Its bet: engineers make better cost decisions when they can see their consequences.
- AI gives that old problem a new meter, counting tokens alongside cloud resources.
In Peterson’s account to AWS, cost became a requirement of the project, alongside its technical demands. That sounds modest. Yet cloud computing had made buying infrastructure almost indistinguishable from writing software. A developer could increase a company’s spending without encountering anything resembling a purchasing department. Finance would discover the decision later, in an invoice.
01 / The receipt is missing the plot
Peterson started CloudZero in 2016, with Matt Manger identified as co-founder in its early materials. The company initially described its work as serverless reliability management. Today, it sells the missing financial connection. Its software collects costs from services including AWS, Azure, Google Cloud, Snowflake, and AI providers, then organizes them around the business. A cloud account is useful to an infrastructure team. Cost per customer is useful to someone deciding whether a subscription price makes sense. Those are different views of the same expenditure.
The underlying expertise is allocation: matching financial records with technical usage. AnyCost brings disparate charges into a common model. CostFormation and Dimensions supply rules for grouping costs by product, team, feature, or customer. Shared databases and Kubernetes workloads complicate the exercise because several customers can consume the same infrastructure. Resource tags help, but a label on a server cannot explain every business relationship.
02 / The expensive thing everybody liked
Consider Drift, the conversational marketing company. Its engineers had formed a five-person task force called COGS Hunters, led by chief architect Freedom Dumlao. In CloudZero’s customer account, the group spent roughly 80% of its time reducing costs. Manual AWS tags and retrospective billing investigations still left the team struggling to connect spending to products.
CloudZero helped expose a troublesome feature: customized messages for website visitors. Popularity made it more expensive to run, and Drift offered it in the free tier. Charging customers was one possible response. The team kept the feature free and found ways to make it cheaper to operate. The case study reports an 80% reduction in that feature’s cost and up to $2.4 million in annual AWS savings overall.
The change in perspective matters. The original problem looked like an oversized AWS bill. The useful problem was an expensive piece of the product. Once costs had owners and context, engineering could choose a specific intervention. The platform located the question; people still had to answer it.
“The most valuable tool in my day-to-day work.”Freedom Dumlao / Drift chief architect
03 / A larger bill can hide a better business
This places CloudZero in FinOps, the practice of managing technology spending across engineering and finance. Customers include Coinbase, Nubank, Klaviyo, and Skyscanner. The audience is a software business with enough infrastructure complexity that the monthly total has stopped explaining much. Product managers need margins; finance needs forecasts; engineers need to know where to investigate.
Its alternatives include Cloudability, CloudHealth, Finout, Vantage, and cloud providers’ own cost tools. CloudZero’s emphasis is engineering ownership and business-level unit costs, with allocation that can work beyond tags. Buyers should compare those capabilities on their own data. A dashboard demonstrating a neat sample workload proves less than one explaining an awkward shared database.
Same bill. Different economics.
on a fixed $100,000 cloud bill
Slide customer volume to see why total spend alone cannot measure efficiency. This example excludes other costs and revenue.
04 / Now the meter counts tokens
AI extends the attribution problem. An invoice can identify an API key while saying little about which customer benefited or whether a feature earned its keep. In May 2026, CloudZero introduced Real-Time AI Spend, capturing usage through an agent on an engineer’s machine or a connection to an LLM gateway. AI Hub adds Claude-powered analysis and access through an MCP server and packaged coding-agent tools.
The promise is a shorter path from a cost question to the evidence behind it. It depends on what gets connected and how costs are defined. A token counter cannot supply missing customer context, and a revenue comparison cannot establish product quality. Teams need meaningful usage data and business measures before a precise-looking number deserves their confidence.
05 / Selling the shared set of numbers
CloudZero’s commercial model is a subscription, quoted according to the scale and complexity of a customer’s environment. Its current pricing page includes unlimited users, sources, dimensions, and dashboards. There is no public dollar tariff. That makes the practical buying question whether better allocation and engineering action justify the license and the work of connecting data.
Investors have funded the experiment: a $32 million Series B in June 2023, followed by a $56 million Series C in May 2025, led by BlueCrest Capital Management and Innovius Capital. MongoDB made a strategic investment. An AWS collaboration announced the following month described plans for AI-assisted analysis and optimization. The company now reports more than $14 billion in cloud and AI spending under management.
06 / Copy the question, then assign an owner
The lesson readers can borrow starts before any purchase. Pick a unit that matters - a customer served, a transaction completed, a feature used. Connect its cost to the team that can change it. Compare the number over time. Keep customer value in the conversation. Drift’s experience suggests that cutting the bill need not mean removing what customers enjoy.
A simple workload may need only native billing tools. A complicated one needs careful allocation and someone willing to act. CloudZero’s published values include “Nothing Is Sacred,” an apt instruction for a business asking engineers to reconsider yesterday’s design. Peterson’s original project left $28 unspent. The enduring idea is to make the budget visible while there is still time to do something interesting with it.