State of Tokenomics
What the data says about AI spend
Large companies spanning industries
Everyone uses Frontier, a third running edge and private AI
Moonshot, DeepSeek
Azure Foundry
Databricks Genie, Snowflake CoCo
Half of AI consumers spread across 4 of the 6 procurement/hosting channels.
Nearly one in five already uses at least one of Deepseek, Qwen, Kimi or GLM.
High cardinality: lots of models being run in lots of places
(you train or fine-tune)
Share of respondents using each surface. Bubble area is proportional to the number of respondents.
The frontier labs have the workloads today. Enterprises considering shift toward open weights.
models
models
0–10 Scale – Balance of open weight vs closed frontier models in your organization
Yet, three in four enterprises cannot confidently prove AI business outcomes to the CFO
Small companies more likely to see the whole picture.
The biggest Tokenomics challenges are tied to value, visibility, and fragmentation
The overwhelming challenge is connecting AI investments to tangible business outcomes, with most organizations unable to quantify ROI beyond basic token consumption metrics.
Companies lack unified dashboards to track AI spending across multiple vendors and business units, creating governance gaps and budget overruns.
The rapid pace of AI development and lack of common metrics across providers makes it difficult for organizations to establish consistent governance frameworks and forecasting models.
Core capabilities focus on attribution, modeling economic impact, model routing, governance, and demand planning
Respondents emphasized tracking and attributing AI costs to specific users, projects, and business units as fundamental capabilities.
Participants stressed connecting AI spending to business outcomes and calculating return on investment. Measuring AI’s business impact as critical but challenging.
Selecting appropriate models for tasks and optimizing AI resource usage through intelligent routing. They saw this as key for controlling costs while maintaining performance.
Participants emphasized implementing spending limits, alerts, and policy controls for AI usage. They viewed governance as necessary for ensuring responsible AI adoption.
The importance of predicting future AI costs and planning resource capacity. They acknowledged difficulty in accurate AI demand prediction.
Governance of developer productivity spending is immature and budget cap focused
Most organizations are in early stages of AI spending governance, with many lacking formal frameworks to manage developer productivity costs effectively.
While companies invest in monitoring tools and dashboards, they struggle to connect AI spending to measurable productivity outcomes and ROI.
Organizations primarily rely on spending caps and token limits as their main control mechanism, suggesting a need for more sophisticated governance models.
Who owns AI Economics in the enterprise? “Technology” is driving but “Shared” is close behind
Those with defined ownership of Tokenomics are 3.7x more likely to show value to the CFO.
Model routing capabilities being bought and built by enterprises of all sizes
Many model routing tools being evaluated while using multiples most common in large enterprises
Significant fragmentation split between established solutions like OpenRouter and LiteLLM, with home grown builds, and cloud-native options common, suggesting need for clearer vendor differentiation.
Many respondents indicate uncertainty or early-stage evaluation, revealing an opportunity to provide educational content, implementation guides, and proof-of-concept support to accelerate adoption decisions.
Advanced users increasingly adopt hybrid approaches combining multiple routing solutions with homegrown solutions, indicating demand for interoperability standards and integration-friendly architectures in future product development.
The ask to model and token providers is more granular AI spend and usage data, not discounts
Enterprises are not asking providers to charge less. They are asking for a bill they can explain to a CFO. That is the cheapest thing a frontier lab could ever give away, and it directly protects the next budget cycle.
Without being asked about it, 7% wrote FOCUS, the open billing data standard now used across cloud providers, into a free-text answer.
Enterprises hungry for better transparency, standardization and measurement from model and token providers
Organizations desperately need real-time, detailed visibility beyond basic token counts, demanding itemized breakdowns and transparent pricing models.
Lack of consistent billing formats and data structures across AI providers creates significant operational challenges for teams.
Companies struggle to connect AI to business outcomes, highlighting the need for better ROI measurement tooling.
Companies expect their pricing models and margins to be impacted by AI
Most companies have changed or are actively planning shifts toward usage-based and outcome-driven models to handle AI cost variability.
Significant pressure from unpredictable AI costs that threaten traditional profit margins and require new financial planning approaches.
High levels of uncertainty indicate the market is still maturing, suggesting need for flexible pricing frameworks and better cost prediction tools.
Are considering energy consumption in their Tokenomics
Typically those that answered yes own or rent hardware (DC or colo) and train first party models or utilize open-weight models
How has AI costs affected staffing and labor planning?
+ Productivity Enhancement: AI is boosting team efficiency and output, allowing organizations to accomplish more work with existing staff levels.
± Skills Transformation: Organizations shifting focus toward AI-specialized roles while reducing demand for traditional or entry-level positions.
± Budget Reallocation: Companies are redirecting funds from traditional hiring budgets to AI investments, treating it as a staffing alternative.
– Layoffs and Reductions: Companies have implemented workforce reductions, with some experiencing significant layoffs directly attributed to AI adoption.