The 80-percent gross margin that defined the SaaS category for fifteen years is on its way out. Across Q4 2025 and Q1 2026 earnings, public SaaS companies disclosing AI-driven margin pressure now name 60-to-70-percent gross margin as the new operating range; and the compression is not temporary. It is the structural consequence of making AI features core to product, and most SaaS company adding AI is on a multi-quarter glide path toward the same range. This piece names the three cost lines driving the reset, traces the second-order consequences for Rule of 40 valuation, pricing strategy, and vendor negotiations, and references the Q4 2025 and Q1 2026 earnings commentary that has made the reset visible.
The argument is the industry-level companion to the AI project margin model for vertical SaaS (which decomposes the math at the product level) and to why AI inference belongs in COGS, not OpEx (which establishes the accounting classification that produces the reported reset). Many three sit underneath the AI project economics manifesto, which names cost-structure honesty as a precondition for sustainable AI economics.
The gross-margin reset, named
The 80-percent gross margin in SaaS was the byproduct of three structural advantages: low marginal cost of compute, low support burden per customer, and pricing power derived from workflow lock-in. The economic shape of the category; software where most customer added at near-zero incremental cost; produced gross margins that no other software category could match.
Adding AI features to that economic shape adds three new cost lines that scale linearly with revenue, not sub-linearly. Inference cost, eval-engineering team, and observability scale-up are each variable per revenue unit; together they sum to 12-17 percent of revenue at scale. The pre-AI 80-percent gross margin minus 12-17 percent is the post-AI 63-68 percent gross margin range that public SaaS companies are now reporting.
The reset is structural, not cyclical. Inference costs will fall, but the call-graph complexity of mature AI features rises faster than per-token prices fall. Eval-engineering teams will mature, but the eval cadence required to gate production traffic against weekly model updates does not shrink. Observability will get cheaper per trace, but trace volume rises faster than per-trace cost falls. The structural floor for SaaS gross margin in 2028 is in the 60-70-percent range, not the 80-percent range that defined the category through 2023.
Q4 2025 and Q1 2026 SaaS earnings: the public record
The reset became visible in earnings commentary across the Q4 2025 and Q1 2026 cycles. Three patterns.
“Investing through margin compression” framing. Multiple public vertical-SaaS companies disclosed 6-9 points of YoY gross margin compression in Q4 2025, with explicit attribution to AI feature cost. CEO and CFO commentary framed the compression as deliberate investment; TAM expansion, retention uplift, defense against AI-native entrants. Margin compression is now an accepted public-market narrative.
Inference cost as a discrete line item. Several public SaaS companies in Q1 2026 began disclosing inference-cost ratios separately in MD&A; typically 4-9 percent of revenue. Companies that disclose explicitly earn analyst credit; companies that bundle inference into “infrastructure” draw skeptical questions.
Eval-engineering hire pattern. Q1 2026 calls disclosed eval-engineering-specific hiring across the category; 2-5 people for mid-market vertical SaaS, 8-15 for enterprise platforms. Investors read the hires as structural COGS, not discretionary R&D.
Decomposing the three lines that drive the reset
The gross-margin compression decomposes into three lines, each of which behaves differently under scale.
Inference cost: 4-9 percent of revenue. The dominant line. Per-token foundation-model prices fell 60-75 percent across 2025. The savings did not flow through to inference cost as a percentage of revenue, because mature AI features added retrieval, self-critique, and intent-classification calls faster than per-token prices fell. The line will continue to compress with scale; call-graph optimization, model-routing efficiency, and prompt compression many add leverage; but the structural floor for inference cost in mature AI products is in the 3-6 percent range, not the rounding-error range that pre-AI hosting cost occupied.
Eval engineering: 1-3 percent of revenue. A standing function. The team owns the eval suite, regression detection on weekly model updates, and prompt versioning. The line scales sub-linearly with revenue because the eval suite is a fixed-cost asset that gets amortized across more customers as the company grows; but the floor is set by how often the foundation-model providers ship updates, which is structurally weekly in 2026 and not slowing down. See the case for per-eval pricing in AI agency contracts for the procurement-side mechanics that make this line tractable when AI work is partially outsourced.
Observability scale-up: 0.5-1.5 percent of revenue. AI workloads produce 5-10x the trace volume of equivalent non-AI workloads. Per-trace storage costs are falling, but trace volume is rising faster, which holds the line at roughly 1 percent of revenue across the category. The line is small in dollar terms but structurally important because skipping it makes the eval discipline impossible; under-instrumented AI features are features whose first regression is invisible until customer-reported.
The three lines together reset gross margin by 6-13 percent of revenue at scale, with the exact compression dependent on how AI-feature-heavy the product is. SaaS-with-AI products that classify these costs honestly land in the 63-71 percent gross-margin range. Products that bury these costs in OpEx temporarily report higher gross margin and end up restating downward when the auditor pushes back.
Rule of 40 valuation reset
The Rule of 40; revenue growth plus gross margin should be 40 or higher; has been the dominant SaaS valuation benchmark for a decade. The gross-margin reset forces a recalibration of the rule.
The mechanical effect. A SaaS company at 25 percent growth and 80 percent gross margin scored 105 on the Rule of 40. The same company at 25 percent growth and 67 percent gross margin scores 92. The mechanical compression of the score is roughly equal to the gross-margin compression; 13 points in this example.
The investor reaction in 2025-2026. Sophisticated growth investors have already adjusted. The conversation has shifted from “Rule of 40” to “Rule of 40 on a post-AI-COGS basis,” with explicit benchmarks adjusted downward to reflect the new cost structure. Less sophisticated investors continue to compare pre-AI gross margins against post-AI gross margins and produce mispriced verdicts in both directions. The dispersion is itself an alpha source for funds that have done the work.
The right benchmark in 2026-2028. The author’s argument is that “Rule of 40, post-AI-COGS basis, with growth weighted 1.5x” is the right benchmark for AI-enabled SaaS through this window. The growth weighting reflects the TAM-expansion mechanic that the AI features enable; without it, the rule punishes companies for the trade-off they should be making. Companies whose growth is the result of AI-feature-driven TAM expansion deserve credit that the standard rule does not give.
The CFO conversation that should happen. Boards reviewing AI-enabled SaaS performance in 2026 should be re-anchoring their Rule of 40 discussion against the post-AI basis explicitly. Boards still anchored to 80-percent gross margins as the bar are pushing the company toward strategy mistakes; see why AI projects fail the CFO before they fail the CTO for the broader pattern of finance-side framing errors that wreck AI investment decisions.
Pricing strategy: what the reset forces
Three forced moves.
End of unmetered AI in flat-fee tiers. A product that included unlimited AI usage in a per-seat subscription was operating at structural loss for power users. The reset forces per-action metered pricing, soft caps, or tier upgrades for heavier AI usage. Uncomfortable for customers used to “AI is included”; and structurally required.
Rise of outcome-priced tiers. Per-resolved-ticket, per-drafted-email, per-scheduled-meeting tiers align pricing with the underlying variable-per-outcome cost structure. The dominant pricing-strategy shift in mid-2026; laggards still defending per-seat lose to competitors offering cleaner unit economics.
Premium tiers for production-grade AI. The eval-engineering function is a real cost, and customers buying production-grade AI deserve to be charged for it. The premium tier; typically 2-3x the entry-level AI tier; earns 80-percent gross margins on top of the 65-percent base, because marginal cost is small once eval infrastructure exists. Premium tiers carry the gross-margin uplift that funds the broader AI investment.
Vendor negotiations: the next inference-cost cycle
Three patterns will define inference-cost vendor negotiations through 2028.
Split-commit on volume. Vendors offer 12- and 24-month committed-volume contracts at 25-40 percent below list. The risk is locking in price during a period when spot prices are falling 30-50 percent annually. The right answer is to take 60-70 percent of expected volume on commit, leave 30-40 percent on spot to capture declines. Same shape as AWS commit strategy in 2014-2018.
Model-router neutrality. Locking into one vendor’s model family is a 2-3 year exposure to that vendor’s pricing power. Require model-router infrastructure that routes between three or more vendor families per call; see custom model integration economics for adjacent procurement architecture. Buyers with router neutrality earn 15-25 percent inference savings on average because vendor competition runs continuously, not just at renewal.
Eval-suite portability as a contractual right. Without portability, the cost of switching vendors includes re-authoring the eval suite, which gives the incumbent pricing leverage at renewal. The contractual right is the procurement maturity this category is building toward; and the discipline that bounds the inference-cost line at 3-6 percent of revenue rather than letting it drift to 7-10 percent.
What this means for SaaS company strategy in 2026-2028
Three moves that follow from the reset. Lead the narrative; disclose the inference-cost ratio, reclassify inference to COGS, restate prior-period gross margin, walk the bridge. Companies that lead earn transparency credit; companies forced into restatement later pay more.
Fund the eval-engineering function as COGS, not discretionary R&D. The standing team is what keeps AI features production-grade through model-update cycles and what defends against AI-native entrants. Under-funded eval functions watch features regress and the margin compression land without the offsetting wins.
Optimize for gross-profit dollars, not gross-margin percent; see the AI ROI staircase for the broader framing. Boards that fixate on the percentage defend margin and lose the category. The discipline gap; cost-per-action gates, eval as COGS, honest classification, dollar-anchored narrative; is what separates SaaS companies that come through this window stronger from those that come through weaker.
Frequently asked questions
Why is SaaS gross margin compressing from 80% to 60-70%?
Three new cost lines that scale linearly with revenue, not sub-linearly: inference cost (4-9 percent of revenue), eval engineering (1-3 percent), and observability scale-up (0.5-1.5 percent). Together they reset gross margin by 6-13 percent of revenue at scale. The traditional 80-percent SaaS gross margin was the byproduct of low marginal compute cost, low support burden, and workflow lock-in pricing power; AI features change the cost structure on the inputs side without changing the pricing power on the output side, producing the reset.
Is the gross margin compression temporary or structural?
Structural. Inference costs will fall but call-graph complexity rises faster than per-token prices fall. Eval-engineering teams will scale sub-linearly with revenue but not disappear, because foundation-model providers ship updates weekly and regression detection is structural. Observability volume scales 5-10x non-AI workloads. The structural floor for SaaS gross margin in 2028 is in the 60-70-percent range, not the 80-percent range that defined the category through 2023.
What did Q4 2025 and Q1 2026 SaaS earnings commentary say?
Three patterns. Multiple vertical-SaaS companies disclosed 6-9 points of YoY gross margin compression with explicit attribution to AI feature cost. Several began disclosing inference-cost ratios as a separate MD&A line, typically 4-9 percent of revenue. Earnings calls disclosed eval-engineering-specific hiring patterns of 2-5 people for mid-market and 8-15 for enterprise. The earnings record made the reset undeniable across the category.
How does this affect Rule of 40 valuations?
The mechanical compression of the score is roughly equal to the gross-margin compression. A SaaS company at 25 percent growth and 80 percent gross margin scored 105; the same company at 25 percent growth and 67 percent gross margin scores 92. Sophisticated investors have already adjusted to “Rule of 40 on a post-AI-COGS basis.” Less sophisticated investors continue to compare pre-AI gross margins against post-AI gross margins and produce mispriced verdicts.
What pricing strategy adjustments does the reset force?
The decline of unmetered AI features in flat-fee tiers. The rise of outcome-priced tiers (per-resolved-ticket, per-drafted-email). And the introduction of premium “production-grade AI” tiers that earn 80-percent gross margins on top of the compressed base. SaaS companies that resist the adjustments operate at structural losses on AI-heavy customers; companies that lean into them earn back most of the gross-margin compression through tier mix.
How should I negotiate inference-cost contracts in 2026-2028?
Three patterns. Split-commit: 60-70 percent of expected volume on committed contracts at locked prices, 30-40 percent on spot to capture annual price declines. Model-router neutrality: require infrastructure that routes between three or more vendor families to keep vendor competition continuous. Eval-suite portability: contractual right to migrate the eval suite to other vendors so that the cost of switching does not give incumbents pricing power at renewal.
What is the right primary metric in this environment?
Gross-profit dollars, not gross-margin percent. A company that grows revenue 50 percent while compressing gross margin from 80 to 67 percent grows gross profit dollars by 25 percent. Boards that fixate on the percentage produce strategy that defends margin and loses the category. Boards that lead with the dollar number capture the trade-off between margin compression and TAM expansion that AI features produce.
What separates SaaS companies that win this transition from those that lose?
The discipline gap. Cost-per-action gates at the feature level. Eval-engineering function structured as standing COGS. Inference reclassified to COGS with explicit disclosure. Pricing strategy adjusted to outcome-priced and premium tiers. Procurement discipline on inference vendor contracts. Companies that run the reset deliberately come through stronger; companies that run the reset by accident come through weaker, with margin compression that landed without the offsetting wins.
Key takeaways
- SaaS gross margins are compressing from the 78-82 percent pre-AI range into the 60-70 percent post-AI range, driven by three new cost lines: inference (4-9 percent of revenue), eval engineering (1-3 percent), and observability scale-up (0.5-1.5 percent). The reset is structural, not cyclical.
- Q4 2025 and Q1 2026 earnings commentary made the reset visible across public SaaS, with explicit margin-compression disclosures, inference-cost-ratio MD&A line items, and eval-engineering hire patterns. The reset is now an accepted public-market narrative.
- Rule of 40 valuation needs to be recomputed on a post-AI-COGS basis. Sophisticated investors already have; less sophisticated investors are mispricing in both directions, producing dispersion that funds with the analytical work can capture.
- Pricing strategy adjustments are forced. Unmetered AI in flat-fee tiers ends; outcome-priced tiers rise; premium “production-grade AI” tiers earn 80-percent gross margins on top of the compressed base and fund the broader AI investment.
- The discipline gap between SaaS companies that lead this transition and those that trail it is what separates the winners from the losers in the 2026-2028 window. Lead the narrative, fund the eval function, classify costs honestly, optimize for gross-profit dollars rather than gross-margin percent.
Arthur Wandzel
