From Per-Device to Per-Token: A Hybrid Approach to AI-Era Software Agreements
In this blog
Remember when public cloud took off more than a decade ago? Enterprises did not fully abandon their on-premises infrastructure. Instead, hybrid cloud strategies emerged and endured, letting businesses balance established investments with new capabilities.
Artificial intelligence (AI) echoes that era. It adds a new way of transacting software alongside traditional enterprise agreements and calls for a hybrid approach. Terms vary widely by vendor. Software for AI factory and on-premises deployments is often licensed per graphics processing unit (GPU), so it behaves much like a traditional enterprise agreement. However, instead of per-seat, per-device or per-node licenses, many cloud-hosted AI products are consumed as a pool of credits or committed spend that you draw down as you use the product. Some also pair an annual or per-seat fee with a consumption component. Expiry dates, overage charges and the size of the commitment can all change what you pay.
The industry calls this tokenomics, the economics of AI, from token production to consumption, payment and value, and the financial operations (FinOps) community has made it a priority. In fact, WWT is a founding member of the Linux Foundation's Tokenomics Foundation. Much tokenomics advice focuses on what happens after the sale, such as tracking burn-down and optimizing usage, and that work stays essential. But the commitment size, rate structure, overage pricing, expiry rules and early renewal terms are set in the agreement, so you have the most leverage pre-signature.
For software buyers, the takeaway is simpler: Your portfolio now holds agreements that behave very differently, and each needs managing from signature to renewal. You may now be managing per-device license agreements, public cloud commitments and AI token consumption together. That takes more than watching how fast credits burn down.
WWT helps you navigate your software agreement portfolio by bringing together software advisory expertise, vendor relationships, lifecycle management and customer success through our EA+ program.
| $2.7 trillion | 40%+ | 64% | 98% |
|---|---|---|---|
| worldwide AI spending forecast for 2026, up 49.5% (Gartner) | of software as a service (SaaS) spend expected to shift to usage-, agent- or outcome-based pricing by 2030 (Deliotte) | of FinOps practitioners now manage software licensing, up 15 points in a year (FinOps Foundation) | of FinOps practitioners now manage AI spend, up from 31% two years earlier (FinOps Foundation) |
Three buying models in most hybrid portfolios
Most software agreement portfolios now mix buying models, and each needs a different kind of planning.
| Model | What is counted | What you commit to | Examples |
|---|---|---|---|
| Traditional enterprise agreement | Devices, users or product suites | A fixed annual fee over a three- or five-year term | Familiar and predictable. This example of Cisco's Enterprise Agreement includes a 15% growth allowance on eligible suites and charges for growth beyond it at the next True Forward review. |
| Committed spend pool | Usage across products, drawn down against a committed dollar amount, as with public cloud | An annual spend commitment for the term | Flexible across products, but the spend level needs planning, as with Grafana Cloud. |
| Hybrid AI agreement | A platform or seat fee, plus tokens, credits, tasks or compute units | Seats or a platform fee, plus a consumption pool often sized before real usage is known | Scales with value as adoption grows, so sizing deserves more care, as with Glean Enterprise Flex. |
How AI vendors transact
Here is how several AI software vendors publicly describe their commercial models.
| Vendor and offering | How it is sold (published terms) |
|---|---|
| Glean Enterprise Flex | Per-user seats plus organization-wide FlexCredits consumed by premium-model reasoning, agent runs and application programming interfaces (APIs). Credit packs can be added at any time. |
| Grafana Cloud | Enterprise starts at a $25,000 annual spend commit, with unit prices that scale with the commitment. |
| Anthropic Claude Enterprise | A per-seat access fee, billed annually, plus all usage at standard API rates; seats include no token allowance. Sales-assisted plans are invoiced monthly in arrears. |
These summaries reflect published terms. Vendors sell other products and billing models, and enterprise contracts are often negotiated away from these defaults.
Many AI vendors also sell through cloud marketplaces. As an example, eligible Microsoft Marketplace offers can count toward an Azure consumption commitment. Eligibility, caps and how much counts vary by agreement and offer, so confirm before pursuing a purchase. WWT can help with marketplace transactions, too.
A real scenario: Consumption outpaced commitment
An enterprise* bought a multi-year committed-spend pool for a SaaS platform. Usage ran at almost twice the pace that would last the full term. Even with optimization, the enterprise faced 35% of the commitment remaining and 78% of applications still to be onboarded. At this rate, the funds were projected to run out about 14 months before the term ends. The customer's choices were to stay on current terms and pay list-rate overages, or renew early at a locked discount.
| Measure | Result |
|---|---|
| Pace that lasts the term | 2.8% a month |
| Actual pace | 5.4% a month, almost twice the pace that lasts the term |
| Commitment | 65% used, 35% left |
| Applications onboarded | 22% onboarded, 78% still to onboard |
| Option A: do nothing | $2.40M a year, overage billed at list rates |
| Option B: early renewal | $1.73M a year, 28% off locked for three years; saves about $670K a year |
*Engagement has been anonymized and figures rounded/adjusted for confidentiality.
The takeaway: Before you sign is the time to understand the terms and negotiate flexibility (e.g., the right to resize or renew early), so you can adjust mid-term if you need to. Consumption forecasts made at signing are hard to get right. Plan for a range of usage, set a checkpoint and renegotiate before you over-consume.
Questions to consider before you sign
- What exactly is being counted in your AI software spend, and how will you see your usage?
- How do you size the commitment before real usage is known, and can you adjust it mid-term if adoption runs ahead of or behind plan?
- Who checks how fast you are using the commitment compared with your forecast, and when do you revisit the deal?
- What happens to credits you do not use, and can they carry over to the next term?
- Is your rate fixed for the full term, or is it a discount off a list price that can go up? Does it also cover usage above your commitment, plus new models and products?
- What protects you at renewal, such as a price cap, a locked discount or the right to resize?
- Can this agreement end on the same date as your other agreements, and be managed alongside them?
One hybrid portfolio, one partner
WWT helps you answer these questions and manage every agreement through the full term. We partner with AI vendors to understand their buying programs and terms, then tailor each agreement to your roadmap. We also manage your traditional agreements, such as Cisco and Palo Alto Networks, so you get one renewal calendar and one view of commitments, whether an agreement counts devices, users, credits or tokens.
Before signature: Software advisory
- Assess and forecast: Historical spend, renewal timelines and a low-to-high usage range
- Compare and model: Term and commitment options against each vendor's tier, expiry and overage rules, with total cost of ownership (TCO) and return on investment (ROI)
- Structure and align: What is counted, shared credit pools, one renewal date and renewal protection in the order form
After signature: Customer success
- Activate and adopt: Licensing setup, success criteria, usage reporting where applicable
- Track and renew: Regular check-ins on consumption, renewals tracked across your portfolio and access to other WWT services
Other WWT services are available to support your AI journey, including our AI Proving Ground for validating platforms before committing, our FinOps for AI Workshop to assess your spend maturity, and training and implementation services to help teams adopt what they buy.
If you want help structuring and managing your next AI software agreement, we are ready to talk. Learn more about WWT EA+.
Sources
- The AI infrastructure reckoning: Optimizing compute strategy in the age of inference economics (Deloitte, Tech Trends 2026)
- Linux Foundation Launches the Tokenomics Foundation to Define the Economics and ROI of AI Value (Linux Foundation, August 4, 2026)
- State of FinOps Survey: AI Value and Skills Top Priorities as FinOps Matures (Linux Foundation and FinOps Foundation, n=1,192, February 19, 2026)
- EA+: Turn enterprise software into a strategic advantage (WWT)
- Gartner Forecasts Worldwide AI Spending to Grow 49.5% in 2026 (Gartner, September 16, 2026)
- SaaS meets AI agents: Transforming budgets, customer experience, and workforce dynamics (Deloitte, 2026 Technology, Media and Telecom Predictions, citing Gartner)
- Cisco Enterprise Agreement security portfolio features (Cisco, April 2026)
- Grafana Cloud pricing (Grafana Labs)
- Glean Enterprise Flex pricing documentation (Glean)
- Token Economics: The Atomic Unit of AI Value (FinOps Foundation, J.R. Storment, May 10, 2026)
- How am I billed for my Enterprise plan? (Anthropic, Claude Help Center)
- Azure consumption commitment benefit (Microsoft Learn)
- AI Proving Ground overview (WWT)
- FinOps for AI Workshop (WWT)