Tech • AI • Robotics • Game
What Happens When ChatGPT Drives the Economy
9.4/10TL;DR
The leading AI companies are increasingly betting that long-term value will come less from frontier models themselves than from becoming the indispensable operating layer for regulated professional work.
KEY POINTS
A costly product with a short exclusivity window
Training a frontier model requires billions of dollars in compute, energy, research and talent, yet the commercial lead can be brief. Mozilla’s State of Open Source AI estimated the gap between top open-weight models and proprietary frontier systems at about 4.4 months. Open models were described as improving every 4 months, versus 5.5 months for closed ones, compressing the period during which premium pricing can be defended.
Why model superiority alone is a weak moat
Closed models still lead on some complex tasks, including long expert workflows, where the best proprietary systems can handle around 12 hours of work against roughly 7 hours for top open models. But that advantage moves quickly as open rivals catch up. In practice, a task that justifies paying for a proprietary model today may become feasible on open infrastructure a quarter later.
The new target is workflow, not just intelligence
The commercial opportunity lies in embedding AI into the actual machinery of work: data access, permissions, templates, traceability, audit logs and compliance controls. In fields like investment banking, answering a question is less valuable than operating inside the bank’s environment with the right market data, internal rules and document formats. That shifts scarcity away from the model and toward the surrounding system that makes intelligence usable.
OpenAI’s finance push illustrates the strategy
On September 10, OpenAI launched ChatGPT for Financial Services, built around ChatGPT 6 Astra and developed with Morgan Stanley and Evercore. The product integrated sources such as Dealogic, PitchBook, LSEG News and Crunchbase, while also supporting role-based access, audit exports and governance controls. It was designed to produce outputs directly in a firm’s own Excel, Word and PowerPoint formats, turning the offering into infrastructure rather than a chatbot.
Adoption is already strongest in enterprise
Consumer willingness to pay remains limited: in the first quarter of 2026, only about 3% of Americans were paying out of pocket for AI tools. Meanwhile, enterprise revenue has overtaken consumer revenue at OpenAI, after finance chief Sarah Friar had said in October 2024 that about 75% of revenue still came from consumer subscriptions. At Anthropic, enterprises already accounted for 80% of revenue in 2025.
The same playbook is spreading across regulated sectors
The pattern now repeats in law and healthcare. Harvey, an AI legal platform used by 80% of the largest US law firms by revenue, gives OpenAI an indirect route into legal work alongside direct products such as Astra for Law. In healthcare, ChatGPT for Healthcare adds medical sources, institutional controls and access to authorized patient information through Epic, extending AI from general assistance into hospital systems.
Integration creates switching costs
A rival model can be downloaded, but replacing a working enterprise setup is harder. Firms must reconnect licensed datasets, internal histories, risk models, templates, procedures and validation layers, then prove the new setup works on sensitive cases. That makes the model increasingly interchangeable while the product environment becomes sticky.
The business model may move beyond subscriptions
Charging per seat or usage may undersell tools that help produce enormous economic value. A researcher earning $300,000 a year might contribute to a drug worth $20 billion; a trading workflow might generate hundreds of millions. Both Anthropic and OpenAI have moved toward services and deployment arms to capture integration revenue, but an even larger prize would be value-sharing through licenses, royalties or outcome-linked contracts.
Royalties could turn AI vendors into economy-wide toll collectors
Executives have floated models in which the AI provider shares directly in upside from discoveries or transactions enabled by its systems. That could mean helping fund AI use upfront, then collecting royalties if a drug succeeds or another commercially valuable result follows. If applied across finance, law, healthcare and research, the provider would not need to own those firms to participate in their economics.
The bigger risk is concentration of power
If banks, hospitals, law firms and public agencies all organize work around the same AI layer, that provider gains leverage over pricing, capabilities and access conditions. OpenAI for Government already groups work with bodies including NASA, the US Treasury and national laboratories. The resulting dependence could boost productivity while reducing resilience, because switching away in a crisis would be operationally difficult.
CONCLUSION
The central bet behind today’s towering AI valuations is not that any one model will stay unique for long, but that a company can use a short-lived technical lead to become embedded in how major institutions work. If that succeeds, the real asset investors are buying is not a model with a four-month edge, but a position at the center of large parts of the economy.
Ask a question