Marginal cost is what it costs to produce one more unit. Not the average cost across all units. Not the total cost of your operation. Just the incremental cost of going from 999 to 1,000 units, or from 10,000 customers to 10,001.
This matters enormously for pricing, and it's where the economics of digital products diverge completely from physical products. A car manufacturer's marginal cost is $20,000-$30,000 per vehicle (materials, labor, assembly). A SaaS company's traditional marginal cost per additional user was close to zero. Then AI came along and changed the math.
The Formula
Marginal Cost = Change in Total Cost / Change in Quantity
MC = ΔTC / ΔQ
If producing 100 units costs $10,000 and producing 101 units costs $10,080, the marginal cost of the 101st unit is $80.
Why Marginal Cost Sets the Pricing Floor
The most basic pricing rule in economics: never price below marginal cost in the long run. If it costs you $80 to produce one more unit and you sell it for $60, you lose $20 on every incremental sale. Volume doesn't fix this. More sales means more losses.
The pricing floor is marginal cost. Everything above that is contribution margin that goes toward covering fixed costs and generating profit.
In practice, the pricing target is usually 2-5x marginal cost, depending on competitive dynamics and the value delivered to customers.
The Digital Product Revolution (and Its AI Disruption)
For decades, the defining feature of software economics was near-zero marginal cost. Build the software once (high fixed cost), then distribute it to millions of users at essentially no incremental cost. This is why SaaS companies achieve 80% gross margins and why investors valued them at premium multiples.
Product Type | Marginal Cost per Unit | Gross Margin |
Traditional software | ~$0 | 80-90% |
Cloud storage | $0.02-0.10/GB | 60-70% |
Streaming video | $0.10-1.00/hour | 20-40% |
AI inference (2024-2025) | $0.50-5.00/interaction | 40-60% |
Physical manufacturing | $10-10,000+/unit | 15-45% |
AI disrupted the zero-marginal-cost model. Every AI interaction requires compute resources: GPU time, memory, electricity. ChatGPT-style applications have meaningful per-query costs. This is why AI-powered SaaS products often have gross margins 15-20 points lower than traditional SaaS. The "build once, sell infinitely" model doesn't apply when every customer interaction costs real money.
Marginal Cost and Economies of Scale
Marginal cost typically decreases as production increases, up to a point:
- Bulk purchasing discounts reduce material costs per unit
- Production learning curves improve efficiency
- Fixed cost spread means overhead per unit declines
- Supply chain optimization at scale
But at very high volumes, marginal cost can increase (diseconomies of scale): production bottlenecks, overtime labor, supply chain strain, quality control complexity.
The sweet spot is where marginal cost is minimized. Economists call this the optimal production level, and it's where the business is most efficient.
What's Changed Recently
AI inference costs are the biggest shift. Companies building AI-powered features face a new cost structure where every user action has a meaningful marginal cost. This is driving a fundamental rethink of pricing models: usage-based pricing, tiered plans with interaction limits, and hybrid models that combine subscription fees with consumption charges.
Dynamic pricing tools leverage marginal cost data to set prices in real time. Airlines, hotels, ride-sharing, and increasingly e-commerce adjust prices based on current marginal cost, demand, and competitive pricing.
Near-zero marginal cost businesses (information products, digital media, software) continue to scale dramatically, but the category is narrowing as AI features add real variable costs.
Frequently Asked Questions
How is marginal cost different from average cost?
Average cost = total cost / total quantity. Marginal cost = cost of one additional unit. When marginal cost is below average cost, average cost is declining (you're gaining efficiency). When marginal cost exceeds average cost, average cost is rising (you're hitting constraints).
Can marginal cost be zero?
Effectively yes, for pure digital products. Sending one more email, serving one more webpage, or letting one more user download an app has near-zero marginal cost. But "near-zero" isn't actually zero: there's bandwidth, storage, and processing overhead, just at fractions of a cent.
How should I use marginal cost in pricing decisions?
Marginal cost is your pricing floor, not your target. Price based on customer value, not cost. But never price below marginal cost long-term. For premium products, price at 5-10x marginal cost. For commodity products, price at 1.5-2x marginal cost.
Why are AI companies struggling with marginal cost?
Because AI model inference (generating responses, processing data) requires GPU compute that costs real money per interaction. Unlike traditional software where serving another user costs almost nothing, serving another AI query costs $0.01-$5.00 depending on model size and complexity. This fundamentally changes unit economics.
Sources & References
- Investopedia. "Marginal Cost." investopedia.com
- Shopify. "What Is Marginal Cost?" shopify.com
- Simon-Kucher. "Marginal Cost Pricing." simon-kucher.com
- Corporate Finance Institute. "Marginal Cost." corporatefinanceinstitute.com
Written by Conan Pesci | Last Updated: April 2026