CluenexHow does traditional per-seat SaaS make money?
Traditional enterprise software sells access per user per month. A customer signs an annual contract granting 500 employees access to an application for a fixed rate, such as $40 per user each month, generating $240,000 in Annual Recurring Revenue (ARR).
This pricing model created the modern software boom because of its unit economics. Once software code is written and hosted on standard cloud infrastructure, adding another user costs fractions of a cent in server bandwidth. Because incremental delivery costs are negligible, mature per-seat software companies regularly produce gross margins between 75% and 85%.
Per-seat licensing also offers unmatched revenue visibility for investors. Customers pay upfront or commit to multi-year contracts, which appear on the balance sheet as deferred revenue and Remaining Performance Obligations (RPO). Unless the customer lays off staff or cancels the contract at renewal, revenue arrives predictably every single quarter.
How do enterprise AI platforms price their products?
Enterprise AI platforms monetize using three distinct structures: raw consumption, hybrid seat add-ons, and agentic outcome billing.
1. Token and compute consumption: Customers pay strictly for what their systems process. Pricing is quoted per 1,000 or 1,000,000 tokens (word fragments), per gigabyte processed, or per GPU second of compute. For example, a customer using an enterprise retrieval model might pay $2.50 per million input tokens and $10 per million output tokens. If the customer runs complex data queries one month and idles the next, the bill fluctuates accordingly.
2. Hybrid per-seat add-ons: Vendors layer a fixed fee on top of an existing per-seat license, but pair it with strict fair-use consumption boundaries. Microsoft adopted this model for Microsoft 365 Copilot, charging a flat list price of $30 per user per month on top of existing business licenses. The vendor absorbs the computational costs within regular usage, but reserves the right to throttle or charge overages if query volumes spike.
3. Outcome and agent-based billing: Instead of charging for human seats or raw tokens, providers bill for completed business actions. When Salesforce introduced its Agentforce platform in late 2024, it set a benchmark pricing standard of $2 per conversation or resolved customer service task. This model decouples software revenue from human employee headcounts and aligns pricing directly with business labor replacement.
Why does AI software have lower gross margins than traditional SaaS?
AI queries require graphic processing units (GPUs) to run continuous mathematical matrix calculations for every single prompt. This ongoing computational load means every interaction incurs hardware depreciation, cloud compute fees, and electrical power costs.
In traditional SaaS, running code accounts for a tiny share of sales, leaving the rest of the revenue to cover research and marketing. In enterprise AI, inference costs are billed directly to Cost of Goods Sold (COGS). When a user submits hundreds of complex document analysis prompts per day, the software vendor must pay cloud providers for GPU time to generate those answers.
Investors can study this dynamic in detail through our guide on gross, operating, and net margins. Across the software sector, pure AI workloads often deliver gross margins between 50% and 65%, trailing the broader benchmark where the median 61.5% stands at 61.5%. Unless a company owns its own energy infrastructure or deploys smaller, optimized open-weight models, AI features inevitably drag down corporate gross margins.
How does token billing change revenue visibility for investors?
Token-based billing shifts software financial analysis from contract accounting to consumption tracking. Under pure consumption, a signed $1 million enterprise deal is merely an allowance, not recognized revenue.
In standard SaaS, investors evaluate ARR by multiplying current contracted subscriptions. In usage-based AI models, ARR is an annualized snapshot of the prior month or quarter. If an enterprise customer finishes a major data migration project or pauses an internal AI pilot, revenue drops immediately without any formal contract cancellation.
This shift also exposes software platforms to prompt optimization headwinds. Enterprise engineering teams continually refine their applications to reduce token waste, cache common responses, and swap large foundation models for smaller task-specific models. When a customer optimizes its code and cuts prompt tokens by 40%, the vendor's revenue from that client falls by 40% overnight, even though software usage remained identical.
Investors evaluating these business models can review our breakdown on reading 10-K filings to see how consumption billing appears in customer concentration notes and deferred revenue schedules.
Per-seat SaaS vs AI pricing models compared
The table below highlights the core differences between traditional seat software and consumption or outcome-driven AI platforms.
| Metric | Traditional Seat SaaS | Token / Compute Billing | Outcome / Agentic Billing |
|---|---|---|---|
| Core Pricing Unit | Named user or seat | Input/output tokens or GPU seconds | Resolved task, ticket, or workflow |
| Gross Margin Range | 75% to 85% | 50% to 65% | 60% to 75% |
| Revenue Predictability | High (contracted upfront ARR) | Low (variable month to month) | Medium (tied to business activity) |
| COGS Sensitivity | Very low (minimal server cost) | High (direct cloud GPU expense) | Moderate (depends on task efficiency) |
| Growth Ceiling | Capped by corporate headcount | Tied to volume of processed data | Tied to total business task volume |
| Balance Sheet Signal | High deferred revenue balance | Low deferred revenue, billed in arrears | Variable credits or prepay balances |
Where does this comparison break down?
The distinction between seat SaaS and consumption software is not absolute, and several practical market dynamics can blur these lines.
First, many modern enterprise vendors run hybrid agreements. A corporation might commit to an annual minimum spend of $500,000 in platform credits to secure volume discounts. The vendor collects predictable cash upfront, insulating its quarterly reports from month-to-month usage dips until the customer exhausts or expires the credit pool.
Second, inference costs decline over time. Advances in semiconductor design and algorithmic model quantization consistently reduce the cost per million tokens. An AI task that costs $0.05 to process today may cost $0.005 two years from now. If an AI vendor can maintain stable pricing while inference costs collapse, its gross margins will expand back toward traditional SaaS levels.
Finally, platform stickiness can offset margin pressure. As explored in our analysis of software switching costs as a moat, software that embeds deeply into enterprise workflows retains customers regardless of billing style. High retention gives vendors the pricing power needed to protect their operating profits.
What this means for you
If you invest in individual software stocks through a standard brokerage account, you cannot evaluate an enterprise AI company using old SaaS rules alone. Here are practical checks to run before buying:
- Track gross margin direction over four quarters: Check whether gross margin is expanding or contracting in the company's 10-Q filing. If revenue is growing 30% but gross margin drops from 74% to 62%, the company is subsidizing heavy GPU inference costs to buy customer adoption.
- Inspect Remaining Performance Obligations (RPO): Look at the gap between current revenue and total RPO. If RPO growth slows while management cites strong token usage, customers are using the platform on short-term commitments rather than locking into multi-year contracts.
- Evaluate Net Revenue Retention (NRR): In our guide on customer churn and retention, we examine how existing accounts expand. A healthy consumption platform should demonstrate an NRR above 120%, proving that existing clients naturally process more tokens over time.
- Verify corporate health scores: You can search any covered software stock on Cluenex at cluenex.com/stocks/ to see its business quality scored from 0 to 10 against its industry peers, highlighting profitability and balance sheet stability directly from SEC filings.