GitHub Copilot lost $20 per user per month for two years. Not one rough quarter. Two years, on a product with millions of paying users, before Microsoft restructured it into five pricing tiers.
And Copilot is not the outlier. OpenAI posted 33% gross margins in 2025. Anthropic runs at roughly 40% on $30 billion in annualized revenue. Even Snowflake, at 66.5%, counts as strong for consumption-based cloud and mediocre by SaaS standards.
AI products run at roughly half the gross margin of traditional SaaS. An operating model built for 80% margins does not survive contact with 50%.
Traditional SaaS built an industry on 80-90% gross margins. That buffer funded 40% sales-and-marketing spend, 20% R&D, and the growth-at-all-costs playbook that defined the 2010s. What changed is recent: AI features moved from demos to production traffic, and inference went from rounding error to one of the largest lines on the P&L. The buffer is half gone. Most companies are still running the playbook it paid for.
Ten Points Left Over
SaaS companies historically spend 40% of revenue on sales and marketing. At 80% gross margins, that leaves 40 points of gross profit to cover R&D, G&A, and operating income.
At 50% gross margins, the same spend leaves 10 points. Ten points for engineering, operations, legal, office space, and profit.
That isn't a business. That's a countdown. The choice is binary: generate twice the revenue per sales dollar, or spend half as much on selling. Most AI companies are doing neither and calling the resulting burn rate "investment in growth."
Where the Numbers Land
This is not one bad survey. ICONIQ Growth tracked AI company gross margins improving from 41% in 2024 to 45% in 2025 to 52% in 2026. Bessemer Venture Partners split AI startups into two archetypes: "Supernovas" hitting $40M ARR in year one at roughly 25% gross margins, and "Shooting Stars" growing more like SaaS companies at 60%. Neither archetype touches 80%.
We keep pulling these datasets hoping to find the exception. The range holds every time. The trajectory is improving, yes, but even the best number in the series, 52%, describes a different business than 82%. What moves is the product mix, and the differences are structural rather than a phase companies grow out of.
| Product Type | Typical Gross Margin | Why |
|---|---|---|
| Pure inference (wrappers, chatbots) | 30-50% | Every user action triggers real compute |
| Hybrid (AI features in SaaS shell) | 50-70% | Non-AI features carry traditional margins; AI features drag the average |
| Outcome-priced (per-resolution, per-result) | 60-67% | Higher pricing power offsets higher cost variance |
Datadog runs at 80.8% gross margins selling monitoring software. Anthropic runs at 40% selling the models Datadog's customers use. Same industry, half the margin. In traditional SaaS the marginal cost of the 10,000th user was effectively zero. In an AI product the 10,000th user costs as much to serve as the first, sometimes more if they run complex multi-step workflows.
The 30-Point Gap, Job by Job
The gap lands differently depending on what you own.
If you run sales
Run the comp math on your own plan. A 15% commission on an 80% margin deal consumes 18.75% of that deal's gross profit. The same commission at 50% margin consumes 30%, and past a certain point the commission on an incremental deal exceeds the margin it generates.
AJ Bruno's QuotaPath team studied hundreds of comp plans at AI-native companies and found them modifying compensation at 3.5x the rate of average SaaS companies. Some have dropped deal-size commissions entirely and pay a flat $5,000 bonus per new logo. Multi-year deal bonuses have nearly disappeared, because locking a price for three years is a liability when inference costs shift quarterly.
The plans that stick pay per-logo bonuses where the relationship is the asset, speed bonuses for skipping long proof-of-concept cycles, and lower base commissions paired with expansion incentives.
If you're the founder
Traditional SaaS taught you to love power users: high engagement, strong retention, near-zero marginal cost. AI products invert that. One customer makes 200 queries a month, another makes 20,000, and both pay the same subscription. In old SaaS the second customer was your best account. In an AI product it may be your least profitable.
GitHub Copilot again. Some individual users cost Microsoft $80 per month while paying $10, and the answer was not efficiency work. It was pricing: enterprise tiers at $39 per user, reaching 40% margins on $300 million in enterprise licensing.
Without per-customer cost data, the variance between your most and least profitable accounts is invisible. At 80% margins the buffer absorbed it. At 50% it shows up directly in the operating plan.
If you own the finance model
The traditional fundraising formula assumed 80% gross margins: $1 of ARR valued at $8-12 of enterprise value, with Rule of 40 as the quality benchmark. Kyle Poyar at Growth Unhinged estimates AI companies need 2-3x more revenue to reach the same profitability milestones as traditional SaaS.
Put it in a model. A $10M ARR AI company at 50% margins has $5M of gross profit. A $10M ARR SaaS company at 80% has $8M. That $3M gap can be the difference between a Series B and a bridge round.
Public markets have already repriced, with SaaS names shedding $285-300 billion in market value over 48 hours as investors reassessed how AI affects revenue models. Private markets are slower. A founder modeling an AI company on SaaS benchmarks is presenting numbers the cost structure underneath does not support.
ARR Is Now a Vanity Metric
In traditional SaaS, ARR was a reliable proxy for business health because nearly all revenue converted to gross profit. In AI products every customer carries real compute cost, so two companies can report $10M ARR with wildly different gross profit depending on customer mix, usage patterns, and inference spend. OpenAI projected $25 billion in cash burn for 2026 despite $12.7 billion in annualized revenue. Revenue growth and loss growth, moving in the same direction.
Todd Gagne at Ibbaka frames it precisely: "For the first time in twenty years, software companies have to care about marginal cost again." If cost scales with usage, price has to scale with usage.
Three derived metrics inherit the distortion. CAC payback calculated on revenue overstates the real payback when a meaningful slice of each dollar goes to compute. LTV built on revenue rather than gross profit produces numbers that feel good and mean less. A 10x revenue multiple on an 80% margin business implies a very different valuation than the same multiple at 50%.
None of this makes ARR meaningless. It means ARR alone no longer tells you whether the business works. Gross profit does.
Pricing Becomes a Weekly Decision
Krzysztof Szyszkiewicz at Monetizely describes clients waking up to a monthly bill from Anthropic or OpenAI that is 2x the previous month. When your largest cost input moves monthly, annual pricing reviews are too slow.
Fynn Glover at Schematic puts it more precisely: "A credit cost. A usage limit. A model tier. An overage threshold. These are not annual pricing reviews. They are weekly operational decisions."
DeepSeek dropped API prices 50% overnight when it released V3.2 in September 2025, bringing input tokens to $0.028 per million, roughly one-tenth of GPT-5 at $1.25 per million. Any company building on those models saw its cost-to-serve change in a single day.
A provider drops prices 40%. A cheaper model launches for your use case. Your largest customer doubles inference volume. Each of these is a pricing decision whether or not anyone priced it deliberately, and they arrive quarterly, not annually. A company reviewing pricing once a year is working from stale data for eleven months of it.
Won't Cheaper Models Close the Gap?
The common objection: model costs are dropping fast, so margins improve on their own.
The data so far says no. DeepSeek cut inference costs 50% in 2025 through sparse attention. OpenAI dropped GPT-4o pricing 80% between launch and early 2026. Anthropic's per-token costs fell as Haiku and Sonnet got cheaper. And still, Anthropic's gross margins came in 10 percentage points below their own internal projections, and OpenAI's inference costs reached $8.4 billion in 2025.
The mechanism: cheaper models unlock new use cases, which drive more usage, which fills the cost gap. When inference gets 50% cheaper, customers do not pocket the savings. They run 3x more queries. Jevons paradox played out the same way in cloud computing, bandwidth, and storage.
Waiting on the cost curve is the bet SaaS companies made in 2015 when they assumed cloud costs would make on-premise competitive. The curve helps. It does not change the underlying economics. AI products carry meaningful marginal cost per interaction, and the operating model has to account for it.
Rebuilding the Operating Model
Start with what you report. Lead the board deck with gross profit, segment customers by profitability rather than size, and gate growth spending on unit economics computed from cost. Most decks still lead with ARR; gross profit appears three slides deep, if at all.
None of that works without per-customer, per-feature cost attribution underneath it. At 80% margins you could afford not to know your cost-to-serve by customer, because the buffer absorbed the variance. At 50% the attribution is the input everything else reads from.
Comp we covered above: pay on margin, not revenue.
Pricing cadence has to match cost cadence, weekly or at worst monthly. In practice that means continuous cost monitoring, billing flexible enough to adjust without an engineering sprint, and pricing authority sitting with the product team instead of locked inside annual contracts.
The last piece is expectations. AI margins will likely settle around 55-65% at maturity, not 80%. The 80% era funded "grow now, optimize later." At 50%, every month of mispriced product compounds, so the cost discipline starts on day one rather than at Series B.
Where This Lands
AI-native margins are roughly half of traditional SaaS margins, and truthfully that is not a problem to solve. It is a cost structure to operate within. The companies that restructure around it compound advantages in sales efficiency, pricing accuracy, and capital allocation. The ones that keep the old playbook spend their runway wondering why the growth math never closes.
Either way, the adjustment starts with seeing the margin you are restructuring around. Bear Lumen shows it per customer and per feature, continuously, with no invoice exports or month-end spreadsheet reconciliation to maintain.