Chamath Says AI Pricing Is About to Flip From Tokens to Results
As seen on the 24/7 Wall St. homepage on October 3, 2026.
If AI vendors get paid for finished work instead of tokens consumed, the revenue models behind today's AI valuations get rewritten, and OpenAI and Salesforce are already testing outcome-based pricing.
Current State of the Union of AI/SI: Most everyone is selling CPM. Cost per thousand tokens + markup. Future State of the Union of AI/SI: Everyone will sell CPA. Accomplish the task, participate in the ROI. In the transition from CPM to CPA, we will debunk much of the current
- Replies33
- Reposts1
- Likes97
Continue ReadingShow less
Chamath Palihapitiya argues that the AI industry is currently built on a simple metering model: vendors charge a cost per thousand tokens plus a markup, a structure he labels CPM. That model is familiar and easy to invoice, and it ties revenue to consumption instead of to the value a customer actually receives.
His forecast is that the industry will shift to CPA, or cost per action, where vendors get paid only when a task is accomplished and share in the return on investment that their work generates. That is a fundamentally different commercial relationship, one where the AI provider has skin in the game alongside the customer.
The implications for how AI companies are valued today are significant. If revenue eventually depends on outcomes rather than raw compute throughput, the assumptions embedded in current AI pricing models, and by extension the valuations built on top of them, need to be reconsidered. The commentary around his post notes that OpenAI and Salesforce are already testing outcome-based pricing, suggesting the transition is not purely theoretical.
Palihapitiya's post was cut off mid-sentence, leaving his full argument unpublished in this version, but the core thesis is clear enough to provoke debate. The post drew more than 21,000 impressions, indicating the idea is landing with an audience that has strong opinions on it either way.