I think Anthropic and OpenAI have found product-market fit

The Shift: From Burning Cash for Growth to Finding a Cash Cow

For a long time, a core contradiction in the AI field was how to translate products with hundreds of millions of users, like ChatGPT, into sustainable, massive revenue. OpenAI boasts over 900 million weekly active users for ChatGPT, but only 50 million—5.6%—are paying consumer subscribers. Charging $10-$20 per month per user is a business, but covering an estimated $1 trillion in infrastructure costs with that model is a tall order. However, industry observer Simon Willison recently pointed out that the winds have changed. The two leading labs—OpenAI and Anthropic—seem to have found the answer: coding and general-purpose AI agents. They are no longer just chasing user counts; they have found ‘whales’ willing to pay a premium for immense value: enterprises, especially those employing highly-paid knowledge workers like software engineers.

Unpacking the Value Logic Behind the Pricing Shift

The most tangible change is a complete overhaul of enterprise pricing models. Almost in sync, Anthropic and OpenAI have shifted their enterprise offerings (like Claude Code/Cowork and Codex) from fixed seat fees with generous allowances to a ‘base seat fee + pay-per-API-use’ model. This means enterprise customers now pay the same per-token rates as the public API.

Underlying this is a crucial product insight: the value of coding agent tools is directly proportional to the resources they consume. Willison himself, as a heavy user, consumes over $2000 worth of API capacity per month from both services combined, while paying only $200 in personal subscriptions. For an enterprise, an agent that automates parts of a software engineer’s work—even at a cost of hundreds of dollars per month—is a bargain if it boosts the productivity of top-tier talent. This is no longer a ‘fun chatbot’; it’s a productivity tool embedded directly into core workflows with a calculable ROI. Thus, the labs have the leverage to say, ‘Since the tool is so useful, please pay according to actual usage.’

Trend Insight: AI Agents as the New ‘Enterprise Software’

This shift reveals several deeper trends:

1. AI Commercialization Enters ‘Deep Waters’: Value capture is moving from a ‘subscription economy’ targeting consumers to a ‘usage economy’ targeting enterprises. Enterprise customers focus more on value than price, and their long-term contracts provide stable cash flow—critical for labs planning an IPO.
2. ‘Agents’ are the Real Game-Changer: The popularity of chatbots was phenomenal, but the business model for agents is sustainable. Agents upgrade AI from an ‘information assistant’ to a ‘task executor,’ integrating directly into workflows and creating an indisputable reason to pay. Coding is the first beachhead, but as Willison notes, any work that can be done by typing commands into a computer can theoretically be automated by agents, opening a vast market.
3. Accelerated Market Polarization: Companies capable of building truly useful agents (currently OpenAI and Anthropic) are gaining pricing power. By releasing stronger models (like GPT-5.5, Opus 4.7) and simultaneously raising API prices, they are forming a positive business cycle: ‘stronger capabilities -> higher value -> higher prices -> more revenue -> R&D investment.’ This could further widen the gap with competitors.

Practical Value and a Counter-Intuitive Point

For IT and internet professionals, this means:

• Re-evaluating AI Budgets: If your company is using or planning to use AI agents at scale, be aware that the cost structure has changed. Pay-per-use can lead to unexpectedly high bills, necessitating monitoring and optimization mechanisms.
• Focusing on ‘Agent Engineering’ over Pure Models: Future competition will center not just on how powerful the base model is, but on how to build efficient, reliable, and deeply integrated agent systems around it. This points developers, product managers, and architects toward new skill sets.
• Changes in Investment Logic: The AI investment narrative is shifting from ‘user growth’ to ‘revenue quality.’ Companies that can prove their tools create measurable value for enterprises will command higher valuations in the capital markets.

A potentially overlooked counter-intuitive point is that individual heavy users are among the ‘biggest winners’ in this shift. They enjoy thousands of dollars’ worth of cutting-edge model service for a very low fixed fee. Essentially, the labs are subsidizing early adopters with individual subscription revenue to cultivate usage habits and gather feedback, with the ultimate goal of leveraging the much larger enterprise goldmine. When enterprises start paying full price, whether the ‘golden age’ for individual users will last becomes an interesting question.