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.
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.
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.
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 typically decreases as production increases, up to a point: