Market data showing how AI companies are changing pricing to capture value from tokens, API calls, compute, and other usage signals
AI pricing is moving toward models that can reflect variable consumption and rapidly changing unit economics. Gartner's September 2025 forecast put worldwide AI spending at nearly $1.5 trillion for 2025 and above $2 trillion for 2026. At the same time, AI products can incur costs that scale with tokens, API calls, compute, model choice, agent behavior, and other consumption variables. That makes pricing architecture a core part of AI unit economics rather than a packaging decision made once at launch.
For companies building consumption-aware AI monetization, billing infrastructure has to meter usage accurately, support hybrid commercial structures, and let pricing evolve as costs and customer behavior change. The 42 statistics below show how quickly that shift is occurring and why revenue teams increasingly need a flexible usage data foundation.
Key takeaways
- Hybrid pricing is dominant among AI agents: Orb's 2026 study of 80 AI agent companies found that 95% use hybrid pricing.
- Usage-based pricing is nearly universal in that sample: 91.3% use usage pricing, up from 83.3% in Orb's 2025 study.
- Broader SaaS data shows a mixed market: High Alpha reports that companies monetizing AI are split across subscription, hybrid, usage, and outcome models rather than converging on one structure.
- Price-to-cost alignment matters: FTI Consulting found that nearly 89% of B2B SaaS providers in its research believe consumption models best align price with AI-specific costs.
- AI budgets are difficult to govern: McKinsey found that 93% exceeded AI budgets in its May 2026 Enterprise AI FinOps survey.
- AI unit economics move quickly: Stanford reports that GPT-3.5-equivalent inference cost fell more than 280-fold between November 2022 and October 2024.
- AI usage can scale dramatically: OpenAI reports average API reasoning-token consumption per organization increased about 320x over 12 months in its enterprise data.
Hybrid models dominate Orb’s AI agent sample
AI agents are a useful window into modern pricing because their cost structure is inherently variable. Orb's 2026 State of AI Agent Pricing study analyzed 80 AI agent companies at a 90% confidence level. The findings show that most companies in the sample combine multiple pricing mechanisms, including subscriptions and usage-linked charges, rather than relying on a single pricing mechanism.
1. 95% of AI agent companies use hybrid pricing
Orb's 2026 research found that 95% use hybrid pricing. In this sample, combining multiple pricing mechanisms is the norm rather than an edge case. For billing teams, that means subscriptions, usage charges, credits, commitments, and other components often need to operate together inside one commercial model.
2. 91.3% of AI agent companies use usage-based pricing
Usage-based pricing appears in 91.3% of companies in Orb's 2026 sample. Tokens, API calls, actions, compute, and AI-generated outputs are all examples of measurable consumption that can become billable units when they track product value or cost.
3. 71.3% of AI agent companies retain a subscription component
Subscriptions remain part of the model for 71.3% of companies in the study. This reinforces why the practical shift is toward hybrid monetization rather than a simple replacement of recurring subscriptions with pure pay-as-you-go pricing.
4. 94.7% of companies using subscriptions also use usage-based pricing
Among the AI agent companies that charge a subscription, 94.7% add usage pricing. A recurring fee can anchor revenue and customer commitments while usage charges capture differences in consumption above or alongside that base.
5. Only 3.8% of AI agent companies use outcome-based pricing
Orb found that 3.8% use outcome pricing, down slightly from 4.5% in 2025. Outcome pricing can align charges closely with delivered results, but it also requires reliable outcome definitions, attribution, and billing-grade measurement.
6. Hybrid pricing rose from 92.4% to 95% year over year
Orb's study shows hybrid adoption moving from 92.4% to 95% between its 2025 and 2026 samples. The small increase is notable because hybrid pricing was already close to saturation in the AI agent category.
7. Usage-based pricing rose from 83.3% to 91.3% year over year
Usage-based pricing increased from 83.3% to 91.3% in Orb's annual AI agent research. That eight percentage-point increase indicates that usage has become an increasingly common component of AI agent monetization.
Broader AI monetization remains mixed
Orb's AI agent sample is especially usage-heavy. High Alpha's 2025 SaaS Benchmarks provide a broader view of companies monetizing AI and show that subscriptions remain common while hybrid and variable models continue to expand.
8. 53% of companies monetizing AI use subscription pricing
High Alpha reports that 53% use subscription pricing among companies monetizing AI in its benchmark. Subscription remains the largest single category in that dataset, even as other approaches gain share.
9. 31% of companies monetizing AI use hybrid pricing
High Alpha found that 31% use hybrid pricing. This structure can pair a predictable base with variable usage or outcome charges, which is useful when product value and delivery cost do not scale cleanly with seats.
10. 11% of companies monetizing AI use pure usage-based pricing
Pure usage-based pricing accounts for 11% of AI monetizers in High Alpha's benchmark. The result is a useful distinction: usage-linked monetization is expanding, but pure usage pricing is still a minority model in the broader SaaS sample.
11. 5% of companies monetizing AI use pure outcome-based pricing
High Alpha reports that 5% use outcome pricing as their primary model. That low share is consistent with the operational difficulty of measuring and attributing outcomes well enough to invoice against them.
12. Companies using hybrid pricing had 105% median NRR in High Alpha's benchmark
Companies using hybrid pricing recorded 105% median NRR, the highest median net revenue retention among the pricing approaches High Alpha discusses. This does not establish causation, but it does show why teams are paying close attention to models that combine recurring and variable revenue.
13. Consumption pricing adoption showed 43% median year-over-year growth
High Alpha reports 43% median growth in adoption of consumption-based pricing. The data points to continued experimentation with models that tie charges to measured product use.
14. Outcome pricing adoption showed 65% median year-over-year growth
Outcome-based pricing adoption showed 65% median growth in High Alpha's benchmark. The absolute share remains small, but the growth rate reflects active experimentation with pricing that is tied more directly to delivered value.
AI pricing is shifting toward cost and value alignment
FTI Consulting's 2025 B2B Pricing Survey shows why consumption models are attracting attention even when many providers still use subscriptions. The underlying issue is alignment: AI products can create variable delivery costs and variable customer value, so a fixed recurring fee may not capture either one on its own.
15. Nearly 89% believe consumption pricing best aligns price to cost
FTI reports that nearly 89% of B2B SaaS providers in its research believe consumption models provide the best alignment between price and AI-specific costs, including compute, data processing, and model maintenance.
16. 83% of surveyed SaaS providers bundle AI with existing products
FTI found that 83% of providers bundle AI with existing products. That helps explain why subscription pricing remains common even when the economics of the AI component are usage-sensitive.
17. 66% of SaaS providers see incremental willingness to pay for AI
FTI reports that AI creates incremental willingness to pay for 66% of SaaS providers in its research, whether AI is sold standalone, as an add-on, or bundled with a price increase.
18. Only 17% of providers sell AI standalone or as an add-on
Only 17% sell AI separately, according to FTI. Most providers are embedding AI into broader offerings, which can make the pricing unit and cost allocation less obvious than in a standalone AI service.
19. About 51% to 58% of subscription and consumption providers bundle AI with an increased price
For subscription and consumption models, FTI says approximately 51% to 58% bundle AI with the current product at an increased price. This is one way providers can monetize AI while preserving familiar packaging.
20. 63% of outcome-priced providers do not raise bundle prices
FTI reports that 63% do not increase the bundle price under outcome-based models. Because outcome pricing already charges for success, the AI value can be reflected in the outcome price itself rather than through a separate bundle uplift.
AI budget growth and adoption are increasing pricing pressure
AI pricing decisions are being made against rapidly expanding budgets. As AI moves from pilots into production workflows, organizations have to govern a larger and more variable cost base.
21. Survey respondents projected 36% growth in average monthly AI spend for 2025
CloudZero's March 2025 survey of 500 U.S. software engineers at manager level and above found that average monthly AI spend was projected to rise 36%, from $62,964 in 2024 to $85,521 in 2025. This was a projection rather than a measured year-end result.
22. 60% of surveyed organizations use generative AI tools
CloudZero found that 60% use generative AI, making generative AI the most widely used AI-tool category in its 2025 research. CloudZero also notes that these tools can introduce substantial costs through compute-heavy inference, tokenized API pricing, and retraining overhead, reinforcing the need to track consumption as adoption scales.
23. 93% of surveyed organizations exceeded their AI budgets
McKinsey's May 2026 Enterprise AI FinOps survey found that 93% exceeded AI budgets. The survey included 120 enterprise participants and 75 qualified respondents across five major industries.
24. Enterprise-wide AI adoption is associated with nearly fourfold AI spend
McKinsey reports that AI spend is nearly four times higher as organizations move from isolated use cases to enterprise-wide adoption. A pricing model that works at pilot scale therefore needs to remain economically sound as customer usage expands.
25. Enterprise leaders expected about 75% LLM budget growth
Andreessen Horowitz's 2025 survey of 100 CIOs across 15 industries found that enterprise leaders expected about 75% budget growth over the following year. The finding reflects how quickly model and application usage can move from experimentation into recurring spend.
26. Innovation budgets fell from 25% to 7% of LLM spend
In the same a16z research, innovation budgets represented just 7% of LLM spend, down from about one quarter a year earlier. The shift toward central IT and business-unit budgets indicates that AI is becoming a recurring operating category rather than a temporary experiment.
AI cost management still has major visibility gaps
Variable consumption is difficult to price well when organizations cannot attribute the underlying cost or forecast demand. The next set of statistics shows why metering, usage visibility, and forecasting have become central to AI FinOps and monetization.
27. Only 51% strongly agree they can track AI ROI effectively
CloudZero found that only 51% strongly agree they can track AI ROI effectively. Reliable cost attribution matters because finance teams need to connect AI consumption to revenue, productivity, or other business outcomes.
28. 57% still use spreadsheets for AI cost management
CloudZero reports that 57% use spreadsheets as a core AI cost-management method. Manual tracking becomes harder to scale as usage sources, pricing tiers, models, and business units multiply.
29. 15% have no formal AI cost-tracking system
CloudZero found that 15% lack formal tracking. Without a structured cost view, it becomes difficult to forecast AI spend or connect product consumption to unit economics.
30. 78% of surveyed IT leaders reported unexpected charges tied to consumption-based or AI pricing
Zylo's 2026 SaaS Management Index, based on a survey of 218 IT leaders, found that 78% experienced unexpected charges tied to consumption-based or AI pricing in the prior year. The finding highlights the importance of spend visibility and customer-facing controls around variable billing.
31. 20% to 30% of AI spend is often unaccounted for
McKinsey says its experience shows that 20% to 30% of AI spend is often unaccounted for because investments are fragmented across vendors, tools, and commercial models.
32. Token usage can vary by as much as 30x for the same task
McKinsey reports that token usage can vary by up to 30x for the same task. That variability makes cost forecasting harder because the unit of user activity does not always map predictably to the unit of model consumption.
33. Only 20% to 25% of companies have mature AI FinOps practices
McKinsey finds that only 20% to 25% of organizations have mature AI FinOps practices. The gap between deployment and financial governance creates a need for better attribution, forecasting, policy, and cost controls.
34. Organizations with high forecasting maturity save 10% more on AI spend
Organizations with strong forecasting maturity save 10% more on AI on average than peers, according to McKinsey. Granular usage data helps support scenario planning before costs appear on an invoice.
35. About one-third of surveyed organizations achieved 20% to 30% savings through optimization
McKinsey reports that about one-third achieved savings of 20% to 30% through active optimization during the prior three months. Better consumption data can help identify the usage patterns where those optimization opportunities exist.
36. Prompt caching can reduce repeated input-token costs by up to about 90%
McKinsey notes that prompt caching can reduce repeated input-token costs by up to 90% in suitable workloads such as retrieval-augmented generation and agents with large, stable prompt prefixes. Cost structures can therefore change materially as engineering teams optimize how models are used.
AI unit economics and usage volumes are moving fast
AI pricing has to operate against two forces at once: the cost per unit can fall quickly, while total usage can rise rapidly. That combination makes static pricing especially difficult to maintain over long periods.
37. GPT-3.5-equivalent inference cost fell more than 280-fold
Stanford's 2025 AI Index reports that the cost to query a model at GPT-3.5-equivalent MMLU performance fell more than 280-fold, from $20 per million tokens in November 2022 to $0.07 by October 2024. The finding shows how quickly model economics can change.
38. AI hardware price-performance costs declined about 30% annually
Stanford also reports that hardware price-performance costs declined about 30% per year. Falling infrastructure cost can create room for margin expansion, lower customer prices, or new packaging, depending on how a provider chooses to compete.
39. 37% of surveyed CIOs report using five or more AI models
Among the 100 CIOs surveyed by Andreessen Horowitz, 37% use five-plus models, up from 29% in the previous year's survey. Multi-model environments make cost allocation and pricing more complex because equivalent product actions can incur different costs depending on routing and model choice.
40. Weekly ChatGPT Enterprise messages grew about 8x
OpenAI reports that weekly ChatGPT Enterprise message volume grew about 8x in aggregate since November 2024. Growth at that rate demonstrates how rapidly AI consumption can expand once tools become embedded in repeatable workflows.
41. API reasoning-token consumption per organization increased about 320x
OpenAI reports that average API reasoning-token consumption per organization increased about 320x over 12 months. For companies that monetize or absorb model usage, consumption growth at this scale makes metering and cost-aware pricing increasingly important.
42. More than 9,000 OpenAI API organizations have processed over 10 billion tokens
OpenAI says more than 9,000 organizations using its API have processed more than 10 billion tokens, while nearly 200 have exceeded one trillion tokens. These volumes illustrate why AI billing systems need to treat usage data as production-grade financial infrastructure.
How usage-based billing addresses AI pricing challenges
The statistics point to a consistent operating requirement: AI companies need billing infrastructure that can meter high-volume consumption, model complex pricing, support pricing iteration, and give customers visibility into variable spend. Orb brings those functions together on a raw usage events foundation designed for usage-based and hybrid revenue models.
Real-time usage metering
AI applications can produce large streams of billable activity across tokens, API calls, compute time, model requests, and product-specific actions. Orb's metering infrastructure stores granular raw usage events, supports high-volume ingestion, and makes usage available for billing, financial analysis, and customer-facing experiences. Retaining raw usage events gives teams a detailed source of truth for billable metrics, backfills, historical analysis, and pricing changes.
Flexible pricing model support
The market data shows that modern AI monetization rarely fits one model. Orb's price modeling supports seat-based, hybrid, prepaid credits, contract commitments, tiered, bulk, pay-as-you-go, and dimensional structures. Orb's dimensional price groups support pricing across multiple usage dimensions, such as region, instance type, and environment, using a single pricing configuration for dimension combinations.
That flexibility is paired with price evolution, which lets teams simulate proposed pricing on real data, build customer-specific rollout plans, and execute changes through an auditable workflow. For AI companies whose unit economics move quickly, pricing can evolve without rebuilding the billing data foundation for every iteration.
Granular visibility and spend controls
Variable pricing works best when customers can see and manage their consumption. Orb's Experience Kit powers advanced dashboards and pricing experiences that help customers plan and understand usage across dimensions such as teams, locations, and metrics. Orb's spend controls add spend thresholds, alerts, workflow triggers, and automated actions so rising usage can be surfaced before it becomes a surprise.
These capabilities are particularly relevant when 78% of IT leaders report unexpected charges tied to consumption-based or AI pricing and 93% of respondents in McKinsey's survey report exceeding AI budgets. Usage transparency is therefore part of the product experience, not only a finance workflow.
Pricing simulation on real usage data
AI pricing teams often have to change rates, packages, or value metrics as real consumption becomes visible. Orb Simulations lets teams model proposed pricing against real product usage data, compare scenarios side by side, and assess customer and revenue impact before launch. That creates a direct path from observed usage to pricing decisions.
Enterprise-grade accuracy and corrections
Small errors can compound quickly when AI products generate large event volumes at low unit prices. Orb's accuracy tooling is built around retained raw usage events, backfills, and backdated price changes. When historical usage or pricing changes require correction, Orb can backfill missing usage or apply retroactive price changes, recalculate affected invoices, and preserve each adjustment in an audit-ready record through its invoicing workflows. Issued billing can be corrected with mechanisms such as credit notes and voiding as appropriate.
This architecture gives engineering, product, finance, and revenue teams a common data foundation from consumption through invoicing. For AI companies operating hybrid and usage-based models, that combination of granular metering, flexible pricing, simulation, spend controls, and auditable corrections makes Orb a purpose-built revenue design platform for pricing that needs to keep changing with the product.
Frequently asked questions
How is AI pricing typically structured for large language models?
LLM API pricing commonly charges separately for input and output tokens, with rates varying by model, capability tier, context behavior, and other technical characteristics. At the application layer, companies can combine those underlying costs with subscriptions, usage charges, prepaid credits, commitments, or hybrid packages. Orb supports these structures within the same pricing and billing foundation.
Why is hybrid pricing common for AI products?
Hybrid pricing combines predictable recurring revenue with a variable component that can reflect consumption. This is useful for AI products because customer usage and delivery cost can vary significantly. Orb's 2026 study found that 95% of companies analyzed use hybrid pricing, and 94.7% of subscription companies also pair subscriptions with usage-based charges.
Can usage-based billing help optimize AI costs?
Usage-based pricing aligns charges with measured consumption, while optimization depends on the visibility and controls around that consumption. Granular metering, dashboards, threshold alerts, budgets, and spend controls can help teams understand which products, customers, workflows, models, or features are driving cost. Orb connects those capabilities to the same raw usage events used for billing.
What capabilities help control AI spend?
Common capabilities include real-time usage dashboards, configurable spend alerts, threshold controls, detailed breakdowns by dimension, historical usage trends, and forecasting inputs. Orb combines customer-facing visibility through Experience Kit with spend controls and billing-grade metering so usage and financial workflows operate from a consistent dataset.
How do token-based and compute-based pricing differ?
Token-based pricing charges for text or model-processing units, commonly separating input and output tokens. Compute-based pricing charges for infrastructure resources consumed, such as GPU time or another compute measure. Some AI products use one of these units directly, while others translate multiple technical inputs into a customer-facing metric or credit. Orb can model both direct usage units and more complex billable metrics over raw usage events.
Why do AI companies need pricing simulation?
AI unit economics can change quickly as model costs fall, routing changes, customers use more complex workflows, and product value becomes clearer. Pricing simulation lets teams apply candidate rates and structures to historical usage before a live rollout. Orb Simulations uses real product usage to show projected customer and revenue impact, giving pricing teams evidence for iteration before changes reach production.