Last month at Moatt Capital we brought you our first Moatt Black analysis on the future of compute economics, predicting a change in how big tech would control and price compute. This week our thesis was confirmed as OpenAI drastically changed the pricing models and compute allowance for users, halving the compute allowance of their pro plan, and introducing a new $500 tier called PRO 500.
In our report “How compute economics, throttling and variable intelligence could reshape the AI business model” we found evidence that big tech was keen to throttle compute allowances given the high cost and subsidised plan pricing that were given to users in the current market.
In the report our conclusion stated:
The subsidy phase sets an expectation before the underlying economics have been fully tested. When OpenAI and Anthropic enter the public markets, the pressure to convert growth into durable margins will intensify. That transition will begin to reveal the true cost of delivering AI at scale, and by then millions of consumers and businesses will already be deeply embedded within their chosen ecosystems.
OpenAI’s pricing and compute allowances have now been published. Stating that the PRO 200 plan for codex use reduces from 20x the Plus plan to 10x the Plus plan, adding that GPT-6 Pro messages in chat fall from 200 to 100 a week. This 50% reduction will impact the majority of small businesses that rely on the $200 plan, and will inevitably force a proportion of those into the new $500 plan.
In leaked financials from June 2025 OpenAI is reported to have a $13 Billion revenue, whilst total costs reach $34 Billion. The leaked documents also show that OpenAI’s R&D exceeded $19 Billion in 2025, more than the companies entire revenue.
As the company barrels towards an IPO and is now on the back foot from Anthropic’s IPO announcement, these are considered steps to combat severe compute bottlenecks and stabilize margins before a public offering.
With agents becoming a large part of the AI ecosystem and their ‘never-off’ technical model the token cost is currently unmanageable. This is only highlighted by the recent Navier-Stokes math problem OpenAI solved in 88 hours. For this, whilst the headlines may have been impressive, solving a 90 year old maths problem and showcasing the ability of AI. But the usage shows an interesting issue with agents. At currently priced token costs the company spent $10 Million to solve the problem. Using 10,000 agents, exchanging just under 3 million messages and using up 130 billion output tokens.
As token subsidisation begins to be removed, and the true cost of intelligence usage comes into view investors at every stage of the cycle need to be more aware of the underlying business model and the external socioeconomic pressures in the USA. Most pertinently for those retail investors looking at listing-day purchasing.
