AI Products Trail Software on Gross Margins: ICONIQ
08 Aug 2026
AI companies are running on structurally thinner margins than the SaaS businesses that preceded them, according to survey data from ICONIQ, and the gap traces back to a cost that traditional software never had to reckon with at the same scale: inference.
The numbers
ICONIQ surveyed roughly 300 executives at software companies building AI products and found:
- 41%: average AI product gross margin in 2024
- 52%: projected average AI product gross margin in 2026
- 75-85%: the gross margin range typical of traditional software
Even with the projected improvement, AI products are expected to remain well below legacy software's margin profile through at least 2026.
Why the gap exists
The report points to a structural difference in cost dynamics. Traditional software has near-zero marginal cost per user once built. AI products don't work that way: every inference call carries a compute cost that scales directly with usage. That creates a direct conflict between margin and product quality on a per-unit basis — better model performance often means more compute, which means higher cost per interaction.
At scale, the report notes, inference tends to become the dominant cost in AI companies, even displacing talent as the primary expense line. That's a notable shift for an industry where engineering and research headcount has historically been the biggest cost driver.
In response, AI-native companies are increasingly moving toward usage-based pricing models — an attempt to align revenue with the variable costs inference imposes, rather than relying on the flatter subscription models that worked for traditional SaaS.
What's not yet clear
The report leaves several open questions. ICONIQ's methodology and its precise definition of "AI product gross margin" aren't specified, and it's unclear which industries or company sizes are represented in the survey sample. There's also no breakdown of how much of the margin gap is attributable to inference costs specifically versus other factors, and no data on whether usage-based pricing adoption is actually correlated with the margin improvements projected for 2026. Beyond that year, there's no visibility into how margins might trend.
Why founders should care
For early-stage founders building AI products, this data suggests a few things are likely true, even if not certain:
- Gross margins in AI products will probably remain below traditional software benchmarks for the foreseeable future, which may affect how investors evaluate AI startups against SaaS comparables.
- The 41%-to-52% projected improvement suggests cost efficiency or pricing strategy could gradually improve, but founders shouldn't assume margins will converge with traditional software anytime soon.
- Because compute cost scales with usage, product quality decisions likely carry direct margin consequences in a way that wasn't true for most software businesses — meaning tradeoffs between model performance and unit economics may need to be made explicitly, not assumed away.
- The move toward usage-based pricing among AI-native companies is likely a rational response to this cost structure, and founders modeling pricing strategy may want to weigh it as an option even if their default instinct is a flat subscription.
The report's core opportunity signal: founders who think carefully about unit economics from day one are better positioned for long-term profitability. In a margin environment where inference costs can quietly become the largest line item, that discipline may matter more for AI startups than it did for the software generation before them.