AI pricing models under pressure as operators rethink subscriptions and usage limits

AI companies are adjusting subscription models and tightening usage as operating costs rise. Services that were previously offered mainly through monthly plans—such as ChatGPT, Claude, and Gemini—are increasingly being restructured to better match the cost of delivering compute-intensive responses.

Why pricing is changing

The core driver is profitability. Each time an AI model generates an answer, it consumes substantial computing resources. New “reasoning” AI models require deeper processing and longer compute times, which increases operating costs.

The issue is amplified by high-frequency usage. Even with a $200/month subscription, some heavy users can still cause the provider to incur losses.

Vietnam pricing example and enterprise shift

In Vietnam, the “ChatGPT Go” package is priced from 132,000 VND per month for individual users up to 2,849,000 VND per month.

Enterprises are also considering separating high-performance features into additional paid services, reflecting a move toward pricing that better aligns with usage intensity and compute demand.

OpenAI and Anthropic cost pressures

Internal financial forecasts suggest OpenAI could record losses of up to $14 billion this year. While ChatGPT has attracted about 900 million users, the paying-user share remains under 5%, meaning most users rely on the free version—raising costs without proportional revenue growth.

OpenAI CEO Sam Altman has acknowledged the pressure, noting that even the $200/month plan does not prevent losses from some high-usage customers, implying that a fixed-fee model may become less sustainable as demand for high-performance models increases.

Anthropic faces similar economics. The company is reported to have annual recurring revenue (ARR) of about $30 billion as of April, yet it still carries large training and operating costs. Internal estimates place Anthropic’s AI-model training costs at about $12 billion per year, while inference costs—the cost of generating answers—are around $7 billion.

Jevons paradox and rising total costs

The Jevons paradox is cited to explain why total costs may not fall even as efficiency improves and per-unit costs decrease: usage can increase faster than efficiency gains.

Move toward B2B and tighter compute management

Providers are expected to push more strategies targeting enterprise customers. Anthropic derives about 70–80% of its revenue from enterprise customers. Google and Microsoft are also focused on expanding B2B by integrating AI into cloud platforms and productivity software.

Infrastructure constraints are further influencing pricing and service terms. CoreWeave, one of the largest listed AI cloud infrastructure providers, raised service prices by more than 20% and tightened contract terms, including requiring small customers to commit to at least three years of service instead of one.

Hourly GPU rental prices have risen sharply since last autumn, reflecting tighter availability across the industry. Anthropic has also faced service disruptions and has begun implementing compute-resource allocation during peak hours to control demand.

OpenAI’s capacity reallocation

OpenAI has similarly adjusted resource allocation. The Wall Street Journal reports that the company paused the video-creation app Sora to free compute capacity for core products, especially programming tools and enterprise customers.

Potential shift to usage-based pricing

AI services historically maintained low prices and near-unlimited usage, supported by abundant funding from major technology groups. As the industry moves toward financial self-sufficiency, this model is less likely to persist. Many analysts expect a shift toward usage-based pricing, similar to utilities or cloud computing.