The Scaling Paradox of AI Services

Jan 23, 2025

“Insane thing: we are currently losing money on openai pro subscriptions! People use it much more than we expected”

This is what Sam Altman, CEO of OpenAI, recently posted on X about their $200-per-month Pro subscription plan. According to Altman, users are utilizing the service so extensively that the cost of running the AI models exceeds the revenue generated from subscriptions.

Clearly, people love using AI services—perhaps even more than AI companies anticipated. But does Altman's comment also imply that making money from AI services, especially in the consumer mass market, is far harder than anyone expected?

Scaling - Why Subscription Model is a Challenge for AI

Most software companies are highly valuable because the marginal cost of adding a new subscriber is virtually zero. This is particularly true for platform-based businesses that sell access to their software. Beyond minor costs for data storage and network usage, most revenue from additional subscribers flows directly to the bottom line. Unlike manufacturing physical products, digital products can scale almost infinitely without significant investments in supply chains, logistics, or production capacity. You simply add a few more servers, and you can handle thousands of new customers.

However, Sam Altman’s post suggests that the "build it once and sell it many times" principle may not apply in the same way for AI services. The reason likely lies in the energy-intensive nature of running AI models. So much energy, in fact, that OpenAI’s Pro subscribers consume more than $200 worth of it each month! Of course, the situation is more nuanced. OpenAI’s cost base is subject to many variables, but Altman’s comments clearly point to usage, rather than fixed costs, as the reason Pro users are unprofitable.

The issue is probably more compounded by the freemium model used by many AI companies. They offer limited versions of their services for free, with subscription tiers based on the level of AI technology offered. But if most mass-market consumers stick to the free version and OpenAI is losing money on its highest-paying tier, how can the AI companies in general generate the profits needed to justify their current valuations? Perhaps the real problem isn’t the cost of running AI services but the way they’re monetized—through fixed-price subscriptions.

Subscriptions work well for services with low variable costs. That’s why they’re popular for digital products like newspapers, magazines, video streaming platforms, telecom providers, and SaaS (Software-as-a-Service) companies. But subscriptions break down when variable costs are high because heavy users can end up costing the provider money. AI companies, as Altman seems to indicate, may fall into this category.

Introducing Pay-Per-Response by UNITT

A simple solution is to move away from the freemium and subscription model—at least for those users who either don’t pay or who use the service excessively.

UNITT is the solution. It is a new payment tool designed for all types of digital things. Using its instant chat interface, users can communicate with each other and send or request UNITT tokens in exchange for accessing content. AI companies can easily integrate their models with the UNITT chat system and charge users for each response generated. In fact, ChatGPT and Claude.ai are already integrated with UNITT.

With UNITT, AI companies can charge variable amounts depending on the complexity of the response. For example, a simple fact-check might cost a fraction of a cent, while generating a video might cost several dollars. This "pay-per-response" model eliminates the need for expensive subscriptions while ensuring AI companies are paid proportionally to usage.

Pay-per-response also lowers the barrier for users to try advanced AI tools by removing the upfront commitment of a subscription. At the same time, AI companies can ensure that no customer consumes more than they pay for.

Often, the main barrier to adopting new technologies is a simple practical problem: how to pay for the service. With UNITT’s pay-per-response tool, this barrier is removed, paving the way for mass-market adoption of all types of AI services.