Goldman Sachs TMT Conference: OpenAI's Top Clients Consume 8x More Than Average, Billing to Shift Toward Outcome-Based Pricing
nashnova research
OpenAI CFO Sarah Friar revealed at Goldman Sachs' Communacopia+ conference that top-tier enterprise clients now consume 8× the tokens of average customers. The company is steering pricing from per-token billing toward outcome-based revenue sharing — a signal that AI monetization is entering the phase where depth of use, not breadth, drives growth.
What does an 8× usage gap tell us?
OpenAI's top 10% of enterprise clients now consume 8× the tokens of average customers; previously that ratio was just 3×.
OpenAI's own internal usage runs even higher — 33× — and Friar framed it as "a preview of where existing customers are headed."
This means → the growth engine has shifted: revenue is no longer driven by signing new logos, but by each customer going deeper.
Why is enterprise outpacing consumer?
July annualized run-rate revenue grew 20% month-on-month; the enterprise slice grew 32%.
Consumer-to-enterprise revenue split moved from 60:40 at the start of the year to 50:50 by mid-year — ahead of schedule.
Codex — OpenAI's coding assistant — jumped from 100,000 to 25 million users; Canva's code is now 100% generated by OpenAI.
In plain terms = enterprise clients are not just experimenting — they have embedded AI into core workflows, and the usage curve is steepening.
How would "outcome-based pricing" actually work?
Friar laid out a clear pricing roadmap: subscription → usage-based → outcome-based, with the end goal being vertical revenue-sharing.
The case for it: OpenAI says it needs 68% fewer output tokens than competitors to achieve the same result.
This means → OpenAI wants customers to stop asking "how much per token?" and start asking "how much per task completed?" — the pricing anchor shifts from compute to business outcome.
Can open-source models still compete?
Friar pushed back firmly: Luna's deployment cost on Cloudflare is already lower than GLM 5.3.
In plain terms = frontier labs like OpenAI can price inference below what it costs enterprises to self-host open-weight models — the open-source cost advantage is eroding.
Is there enough compute?
Friar acknowledged that capacity remains very tight; the company makes weekly trade-offs between training, research, and inference.
The path to margin expansion depends on two forces: rising revenue per gigawatt + a declining cost curve.
Asked about criticism of over-investment, Friar said the spending "truly paid off this year."
This reflects a bottleneck that is not about demand — it is about supply failing to keep pace with demand.
What is the next key proof point?
Whether outcome-based pricing can move from roadmap to scalable contract structure is the next milestone for OpenAI's commercial credibility.
This means → investors should watch not user counts, but whether OpenAI can close deals that pay out on business results — that is the real proof of pricing power.
市场有风险，内容仅供研究参考，不构成投资建议。