The era of using generative artificial intelligence (AI) virtually "without limits" is fading. As the old model—near-unlimited use for a flat monthly fee of about $20 (about 29,000 won)—hits its limits, companies are simultaneously moving to raise prices, curb usage, and introduce ads.
On the 14th, according to related industries, OpenAI, Anthropic, Google, and xAI are rapidly overhauling their pricing. Previously, major services such as ChatGPT, Claude, and Gemini ran mainly on flat plans around $20 a month, but lately they have introduced high-end plans above $200 or have shifted to restricting access to high-performance features separately.
The core backdrop to this shift is a collapse in profitability. Generative AI consumes massive computing resources each time it generates an answer, and the recently emerged "inference AI" requires far more computation than before, driving expense up sharply.
OpenAI's internal financial outlook projects the company will post a loss of about $14 billion (about 21 trillion won) this year. ChatGPT users total about 900 million, but paying users account for less than 5%. With most users staying on the free service, the structure generates expense without revenue.
OpenAI Chief Executive Officer Sam Altman directly raised this structural problem. Altman said, "Even at a $200 monthly plan, some heavy users still cause losses." For users who concentrate on high-performance inference models, a flat fee makes it hard to cover the expense.
This phenomenon is appearing across the industry. As of April, Anthropic's annual recurring revenue (ARR) is said to have grown to around $30 billion, yet it still bears the burden of large-scale training and compute expense. By Anthropic's internal estimates, AI model training expense runs about $12 billion a year, and inference expense about $7 billion.
Overall AI industry expense is also rising quickly. Summarizing major big tech corporations' investment plans, global AI infrastructure investment in 2026 is estimated to have surpassed $500 billion. Microsoft (MS), Google, Amazon, and Meta are pouring tens of billions of dollars annually into data centers and semiconductors.
The problem is that while unit computing expense has fallen with technological advances, total expense is instead increasing. In a recent report, market research firm Gartner predicted that "AI inference expense will decline by more than 90% by 2030," but also analyzed that as usage surges, total expense is rising.
This is explained by the "Jevons paradox" in economics: the more efficient a technology becomes, the more usage grows, increasing total consumption. In fact, AI token unit prices have plunged in a few years, but the growth in usage is outpacing that decline.
Because of this, corporations see it as difficult to maintain an "unlimited flat-rate" model. Instead, they are moving to segment plans, impose extra charges for high-performance features, or set limits once usage exceeds a certain level.
Policies for free users are also changing. Early this year, OpenAI launched the "ChatGPT Go" ($8 per month) plan and is testing ad introductions for free and low-cost plan users. The company has called ads a "last resort" as it seeks to diversify its revenue structure.
There is also a strengthening shift toward enterprise customers. Anthropic secures 70%–80% of its total sales from corporate clients, and Google and Microsoft are likewise integrating AI into cloud and workplace software, focusing on the enterprise (B2B) market.
In the industry, the view is that low-cost, near-unlimited AI services were possible thanks to investment funds and big tech support, but it is now time to recoup expense. Analysts say pricing will likely be reorganized into usage-based billing, like electricity or cloud services.
Choi Byoung-ho, research professor at the Human-Inspired AI Institute at Korea University, said, "In the current Generative AI market, revenue is shifting toward B2B services," adding, "Consumer plans that are close to unlimited are structurally hard to sustain." Choi added, "Free services are largely a user acquisition strategy and are likely to operate in forms such as expanding premium plans or tightening limits going forward."