Oracle just stopped selling AI by the token. In its Q4 earnings call, the company revealed a pilot program with 33 organizations where customers pay for business outcomes — not for the API meter spinning in the background. If that sounds like a small billing tweak, look again: it’s the first serious attempt by a hyperscaler to fix the budgeting nightmare that has quietly broken enterprise AI procurement.
Why Oracle Is Walking Away From Token-Based Billing
Oracle’s Q4 numbers told two stories at once. According to newly appointed CFO Hilary Mason, cloud infrastructure revenue jumped 93% year over year, driven by AI workloads and database demand. Oracle is now planning to push capital expenditure above $70 billion (roughly 105 trillion won) next year to keep up. But the more interesting line came from CEO Mike Sicilia, who said the company is running a pilot that ties price directly to the value customers create.
This matters because token-based pricing is hostile to how enterprises plan. CFOs build annual budgets. Token meters do not respect annual budgets. As Greyhound Research’s chief analyst Sanchit Vir Gogia put it, no enterprise wants to build a strategic AI roadmap on top of a cost model that behaves like a taxi meter. He cited research showing agentic workloads can consume up to 1,000x more tokens than simple tasks, and even the same task can vary by up to 30x depending on how it’s executed.
Imagine you’re a mid-sized insurance company piloting an AI-integrated claims processing system. Under token pricing, your finance team has no way to forecast next quarter’s bill within an order of magnitude. Under outcome pricing — say, a flat fee per processed claim — that same team can finally model ROI. Our take: this is the pricing model that finally lets CIOs sell AI projects to skeptical CFOs.
The 1,000-Agent Pivot Hiding Inside the Earnings Report
Sicilia disclosed that Oracle has shipped more than 1,000 AI agents across its application suite over the past year. These agents reason, decide, and execute work across business processes — not just answer questions. That’s the operational foundation that makes outcome pricing even possible: if you control the data, the database, and the agent that acts on both, you can define and measure what “done” looks like.
That’s where Oracle’s advantage gets concrete. Gogia argued that because Oracle directly manages the core operational data and databases of its enterprise customers, it can define outcomes far more reliably than vendors who only see prompts and responses going over an API. That’s a genuine moat. A standalone LLM provider has no idea whether the answer it gave actually closed a support ticket, processed an invoice, or onboarded a customer. Oracle does, because the system of record is sitting in its own cloud.
If you’re a logistics company running on Oracle’s stack, that means your AI agent doesn’t just suggest a routing change — Oracle can verify the route was executed, the delivery completed, and bill you accordingly. Our prediction: every hyperscaler with a foothold in enterprise data — Microsoft via Dynamics, Salesforce via Data Cloud, SAP via S/4HANA — will announce some version of outcome-based pricing within the next 18 months. The ones without that data layer will be stuck selling tokens.
The Conflict-of-Interest Problem Nobody Is Talking About
Gogia flagged the obvious risk, and it deserves more attention than it’s getting: when the vendor is also the referee, outcome-based pricing can quietly turn into a self-grading exam. If Oracle defines what counts as a successful outcome, measures whether it happened, and then bills you for it, the incentives get murky fast.
This is the same governance problem that has plagued KYC and identity verification software for years — whoever defines “verified” controls the economics. Enterprise buyers will need contractual definitions of success metrics, third-party audit rights, and clear dispute mechanisms before signing outcome-based deals. Gogia also pointed out that tokens aren’t actually disappearing; they’re being repackaged behind cleaner commercial interfaces. The meter still spins. The question, in his words, is no longer whether token complexity has gone away — it’s where that complexity has been hidden.
If you’re a procurement leader evaluating one of these new outcome-priced contracts, the practical move is to insist on visibility into the underlying consumption data, even if you’re not being billed on it. Our take: the contracts that win in 2027 will look more like managed services agreements than software licenses, with SLAs tied to business KPIs rather than uptime.
What This Means for the Token Abstraction Race
Scott Bickley, advisory fellow at Info-Tech Research Group, framed it bluntly: tokens are essentially a black box. Buyers struggle to understand what they’re actually purchasing, models consume tokens differently, and the goalposts keep moving. SaaS vendors have responded with abstractions — “AI Work Units,” “AI Credits,” “AI Actions,” Salesforce’s “Agentforce Actions” — but Bickley argues those wrappers don’t go far enough. You still need to understand the underlying mechanics to budget responsibly.
The vendor that fully hides this complexity behind a fixed price for a guaranteed outcome, Bickley said, will pull ahead of competitors. If you’re building enterprise SaaS or AI-powered B2B products, take note. The pricing model is becoming a feature.
If you’re a fintech building AI credit scoring, for example, your enterprise buyers don’t want to pay per inference. They want to pay per approved loan that performs above a default threshold. The infrastructure to deliver that pricing model — measurement, attribution, dispute handling — is now part of the product, not just the contract. Our prediction: by 2027, “price per outcome” will be a checkbox on enterprise AI RFPs, and vendors who can’t answer it will be filtered out before the first demo.
FAQ
Q: What is outcome-based AI pricing? A: It’s a billing model where customers pay for measurable business results — a processed claim, a closed ticket, a verified identity — rather than for compute usage like tokens or API calls. Oracle’s pilot, currently limited to 33 organizations, is one of the first major hyperscaler implementations of this approach.
Q: Does outcome-based pricing mean tokens are going away? A: No. Per Greyhound Research’s Sanchit Vir Gogia, tokens aren’t disappearing — they’re being hidden behind cleaner commercial interfaces. The consumption meter still runs internally; vendors are just absorbing the volatility so customers see predictable prices.
Q: Why is Oracle expanding CAPEX to over $70 billion? A: According to CFO Hilary Mason, cloud infrastructure revenue grew 93% year over year on AI and database demand, and Oracle is building out capacity to match committed customer demand. Analysts caution, however, that the real bottleneck for AI infrastructure is shifting from GPUs to power supply, permitting, and political conditions.
Key Takeaways
- Enterprise buyers should start asking every AI vendor for an outcome-based pricing option in their next RFP cycle — vendors without an answer will be at a structural disadvantage within 18 months.
- Negotiate audit rights and independent measurement clauses into any outcome-based AI contract; the vendor cannot be both supplier and sole judge of success.
- Hyperscalers that own enterprise system-of-record data have a structural advantage in outcome pricing; pure-play LLM vendors will struggle to compete on this dimension.
- The real AI infrastructure constraint is moving from GPU availability to power, permits, and political environment — factor this into multi-year vendor commitments.
- Treat token abstractions like “AI Credits” or “AI Actions” with skepticism; they often repackage the same volatility under a friendlier label without fixing the budgeting problem underneath.