AI Startup Gross Margins: What Investors and Buyers Want to See

Why AI Gross Margins Deserve Closer Attention

AI applications incur inference costs when customers use the product. More activity can also require additional hosting, third-party data, or human support. If revenue stays fixed while those costs increase, greater engagement can reduce the gross profit earned from a customer.

Gross margin measures the percentage of revenue remaining after cost of revenue:

Gross Margin (%) = (Revenue − Cost of Revenue) ÷ Revenue × 100

At a 60% gross margin, each dollar of recognized revenue leaves 60 cents of gross profit to help cover operating expenses.

Lightspeed illustrates this relationship with four hypothetical businesses that have identical revenue growth and operating spending. Over five years, its 20% gross margin business burns a cumulative $160 million, while its 80% margin business generates $27 million in cumulative cash flow. The example shows how different delivery costs can change the capital required to support the same growth.

What Is a Good Gross Margin for an AI Startup?

There’s no single target that fits every AI application. Product complexity, pricing, customer mix, and the amount of human involvement all affect delivery economics.

• ICONIQ’s July 2026 State of AI report draws on a survey of roughly 300 software-company executives building AI products. It reports average AI product gross margins of 45% in 2025 and projections of 53% in 2026 and 59% in 2027. Those forward-looking figures reflect respondent expectations, not guaranteed results or targets every startup should adopt.
• Bessemer’s 2025 research highlights a different distinction. Among 20 selected high-growth AI startups, its ten fastest-scaling “Supernovas” averaged roughly 25% gross margins. Its “Shooting Stars” averaged approximately 60%. This is a study of selected standout businesses, not a representative industry sample.

Use those findings to frame questions about your business. How much of your margin reflects temporary launch costs? Which customer segments are economically attractive? Does greater adoption improve gross profit, or expose a pricing problem?

Which AI Costs Belong in Ongoing Cost of Revenue?

Start by identifying what it takes to deliver the products and services customers purchase. Looking only at a model provider’s invoice can leave out meaningful delivery costs.

The table below provides a starting point for review with your finance team. Final treatment depends on the activity, applicable accounting guidance, and your company’s policies. Some development costs may require capitalization rather than immediate expense recognition.

A shared cloud bill or an engineer’s job title doesn’t settle classification. One engineer may spend time both improving the product (Opex) and operating customer workloads (Cost of Revenue). Your accounting policy needs to address how the Engineers expenses are consistently attributed to the appropriate expense bucket.

An AI inference provider’s invoice may cover customer workloads, product development, and employees using AI for internal finance, HR, or sales workflows. Classify spending by the activity it supports, rather than treating the entire bill as cost of revenue. Where your tools allow it, use separate projects, API keys, or billing tags to distinguish these uses before spending becomes difficult to reconstruct.

Moving an expense between cost of revenue and operating expenses changes reported gross margin. It doesn’t reduce the underlying cost or improve total operating profit. Operational improvement requires a change in pricing, delivery efficiency, or the mix of business you serve.

Why Blended Gross Margin Can Hide a Problem

Consider an AI application with two customers who each pay a $1,000 monthly subscription. Customer 1 uses the product moderately, while Customer 2 uses it heavily. Both generate $1,000 in monthly recognized revenue, but their inference costs differ. Assume the company spends $3,000 to acquire each customer. All figures are hypothetical. To isolate the effect of inference usage, the example holds all other costs of revenue at a combined $100 per customer per month. In practice, these costs may also rise with usage.

Together, the two customers generate $2,000 in monthly revenue and $1,000 in gross profit, producing a blended gross margin of 50%. That average hides the difference between Customer 1’s 75% margin and Customer 2’s 25% margin. If more customers move toward heavy usage while subscription prices stay fixed, the company’s blended margin falls.

Customer acquisition cost (CAC) payback measures how long monthly gross profit takes to recover acquisition spending. Customer 1 repays the $3,000 acquisition cost in four months ($3,000 ÷ $750), while Customer 2 takes twelve months ($3,000 ÷ $250). The combined payback is six months ($6,000 ÷ $1,000). These calculations assume stable monthly economics and no churn during payback, and exclude cash collection timing differences.

Segment reporting makes the issue visible. Review gross margins by product, pricing tier, customer cohort, and usage level where the data supports it. For some businesses, implementation-heavy customers or workflows requiring frequent human intervention deserve separate analysis.

If you earn both software subscription revenue and implementation or professional services revenue, report revenue and gross margin separately for each, and reconcile them to the company total. This helps explain whether a lower blended margin reflects software delivery costs or a larger services component. It also lets you assess whether implementation work (which is often lower margin) supports adoption and retention at an acceptable cost.

Use this analysis to inform pricing and product decisions. If you also track contribution margin or a “fully loaded” management metric, define the included costs and reconcile it to your financial statements. Don’t use the same label for calculations with different cost bases.

What Makes a Margin Improvement Plan Credible

Declining inference prices can help, but they don’t guarantee better margins. Customers may use more tokens, move to more expensive capabilities, or expect lower prices.

Andreessen Horowitz makes a useful counterpoint to judging AI businesses solely on current margins. Its analysis emphasizes model routing, pricing and packaging, conversion to paid and enterprise plans, and customer retention. Present margins need to be interpreted alongside product value and the economics of acquiring and retaining customers.

Translate those possibilities into a forecast your team can test. If you expect model routing to lower costs, identify the workloads eligible for cheaper models and verify that quality remains acceptable. If you plan to reduce human review, measure the intervention rate and the work required to change it.

Pricing changes need similar discipline. Usage allowances, overages, or higher-value tiers may improve economics, but your forecast should account for rollout timing and potential effects on conversion and retention. Don’t assume every existing customer immediately moves to a new plan.

Track the difference between expected and realized improvement each month. An improvement plan is more useful when engineering, finance, and commercial leaders agree on the assumptions and own the actions behind them.

How Gross Margin Questions Can Affect M&A Diligence

In a recent client sale process supported by Burkland, a review of historical gross margins became a major focus of diligence and led to a valuation adjustment. The buyer argued that the company was incorrectly attributing some expenses to operating expenses that should have been classified as part of cost of revenue. As a result, the buyer argued that the gross margin that was originally presented to them was inflated, and they negotiated for a lower purchase price vs. what was offered in the initial term sheet.

The practical lesson is to provide a transparent overview of expense classification methodologies before a transaction begins. A buyer may want to understand whether historical margins are repeatable, whether delivery depends on work excluded from the calculation, and whether projected improvements require additional investment.

What to Prepare Before Investor or Buyer Diligence

Build a concise margin package that connects reported results to the supporting records:

• A written cost classification policy explaining included expenses and shared-cost allocations.
• Monthly gross margin history reconciled to the financial statements, with explanations for material changes.
• Customer or segment analysis showing where delivery economics differ and how allocations were made.
• A forecast connecting margin improvement to specific actions, owners, timing, and downside assumptions.

Be transparent about cloud credits, temporary discounts, and pricing concessions that affect historical economics. If you present a view excluding a temporary benefit, show the adjustment separately and reconcile it to reported results.

Use recognized revenue for financial-statement gross margin calculations. Keep ARR and usage run-rate analysis clearly labeled. Burkland’s AI ARR You Can Defend explains how to substantiate the revenue side of diligence, while Financial KPIs for AI Startups provides a broader framework for monitoring performance.

The goal is to make your economics understandable and your improvement plan testable. Burkland’s Accounting and Fractional CFO teams can help you establish consistent reporting, examine delivery costs, and prepare financial information for fundraising or M&A diligence.