Software Economics: Low Distribution Cost, High Fixed Commitments

Why software can be cheap to reproduce while remaining expensive to build, host, support, and keep useful.

“Zero marginal cost” describes one boundary

Once a software product exists, supplying another download or permitting another user to access the same code can require little additional copying cost. That is the useful intuition behind near-zero marginal cost. It is not a statement that the next customer is free. Cloud compute, storage, bandwidth, payment fees, customer support, security, implementation, and sales all add costs, and some products require substantial human work per account.

The fixed costs are also unusual. Engineering, product design, testing, documentation, and compliance are incurred before many of the future users arrive. A company can spread those costs across more customers, but it must first finance them and keep the product compatible and secure. Low distribution cost creates an opportunity for operating leverage; it does not guarantee that the company will capture the value.

Research and development is the main fixed commitment

Software companies spend on engineers, product managers, testing, infrastructure, and security before revenue appears. Under U.S. GAAP, many research and development costs are expensed as incurred, while some software and internal-use costs are capitalized after specified technical stages. The accounting boundary therefore differs from the economic history of the product.

Microsoft's 2024 Form 10-K reports $29.5 billion of research and development expense and explains that it includes product-development payroll, third-party development, and purchased software amortization. The filing is evidence of reported spending, not proof that every dollar creates a durable asset or that a high R&D ratio is inherently good. Microsoft's 2024 Form 10-K shows how a large software and cloud business presents this cost boundary.

Scale can lower delivery cost without creating a moat

When a platform serves more customers with roughly the same core code, average development cost per customer can fall. That can support lower prices, higher margins, or more investment. Competitors can still copy features, customers can switch, and a new architecture can reset the basis of competition. The relevant question is whether the company has an advantage in product quality, distribution, data, integrations, brand, or switching costs that survives price competition.

Network and data effects can reinforce scale when additional users improve matching, fraud detection, content, or liquidity. They can also be weak when users multi-home, data is portable, or the value comes from a standard available to every provider. “Winner-take-most” is a possible outcome in a particular category, not a law of software.

Cloud delivery restores a variable cost

On-premises software pushed servers, storage, and much of the operating work to the customer. A cloud provider or SaaS vendor takes more of that work back: compute, storage, networking, uptime, security, and support become part of the service. The incremental cost remains low relative to many physical goods, but it is not zero and can rise sharply with data-intensive workloads.

Microsoft's filings show this boundary in practice: in later reporting, cloud growth increased cost of revenue as infrastructure scaled, while engineering investment continued. A cloud company must therefore track gross margin, usage intensity, service levels, capacity commitments, and the cost of serving each workload rather than extrapolate an on-premises margin to a hosted service.

Subscription changes cash and revenue timing

Perpetual licenses can produce a large transaction at delivery. A subscription spreads the promised service over a term, often with cash collected upfront and revenue recognized over time. During a transition, reported revenue can grow more slowly than bookings or cash, while support and engineering costs continue. The accounting change is real, but it does not by itself show whether the customer relationship became more durable.

Measure annual recurring revenue, renewal, expansion, churn, deferred revenue, billings, cash collection, and support cost together. A subscription that requires expensive implementation and loses customers at renewal has different economics from one that is embedded in a workflow and renews with little incremental work.

Professional services reveal how much human work remains

Implementation, customization, training, and managed services can be essential to the customer receiving value. A high services share may mean the product is complex, the customer needs help changing its process, or the vendor is effectively doing the work that the software was supposed to automate. A low services share may reflect a self-serve product, mature integrations, or work performed by partners outside the reported company.

How to test software leverage

Separate code-copying cost from cloud and support cost. Compare customer growth with engineering, sales, support, and infrastructure expense. Follow cohorts through renewal and expansion. Check whether price increases hold volume, whether users can export data, and whether competitors can offer compatible alternatives. When comparing companies, adjust for capitalization of development, stock compensation, acquisition accounting, and outsourced labor.

The conclusion should remain conditional. Low incremental distribution cost is a powerful operating property. It becomes durable economic advantage only when the product remains useful, the delivery system scales, customers continue to pay, and the company can fund the fixed work that keeps the software alive.

Inside CompanyGraph

CompanyGraph tracks the cash-conversion print live: companies whose operating cash flow margin, free-cash-flow share of operating cash flow, and cash flow relative to sales all sit in elevated ranges.