AI Subscriptions Pile Up as Users Face Monthly Bills of Hundreds of Dollars
People and businesses are finding that using artificial intelligence (AI) costs more than just paying for a ChatGPT subscription. AI tools are now built into everyday software, including Zoom, Adobe, Slack, Microsoft 365, Notion and Dropbox. Many of these tools require a more expensive plan or an extra fee. These charges can range from a few hundred yen to tens of thousands of yen each month.
When each service costs about ¥3,000 (around $19) a month, the total can grow quickly. One ¥3,000 monthly add-on costs ¥36,000 (about $230) over a year and ¥360,000 (around $2,300) over ten years. Add a main chatbot subscription and several other AI services, and monthly spending can reach tens of thousands of yen. This is according to an analysis published by Toyo Keizai Online.
This situation is similar to the cloud storage boom more than ten years ago. At that time, Dropbox, Evernote, Box and iCloud competed by offering free plans and paid upgrades. But there is an important difference with AI: every time someone sends an AI a request, the company running it must pay the ongoing cost of processing that request.
Why Every App Wants a Share of the AI Spending
Most software companies do not build their own large language models, the systems that power many AI tools. Instead, they rent computing services through APIs from OpenAI, Anthropic, Google or xAI. They pay based on how much data is processed and which model they use. These changing costs are why companies charge subscription fees, limit how much customers can use their services or charge based on usage.
API fees are not the only expense. Companies must also pay to build and maintain AI features, store and retrieve data, and protect their systems. Companies that run their own AI models must also pay for the computing power needed to operate them.
OpenAI's latest move shows how hard AI companies are working to find a reliable way to make money. Its new Dots product is an AI agent service that works inside ChatGPT. It has a plan costing $200 a month and another costing $500 a month for heavy business users. Rachel Warren, an analyst at Motley Fool, called it "an advanced workflow infrastructure layer" rather than a chatbot. She said it is aimed at data analysts working with large data sets and software engineers who run code tests continuously.
Even with these high prices, making the service profitable is difficult. Warren noted that OpenAI is already reducing computing limits on its $200 plan because running AI agents in the background is expensive. Lou Whiteman, another Motley Fool analyst, questioned how many people would pay for it. He said, "There's no chance normals are going to say, let me sign up to pay $200 a month to find out if this is any good."
The Price War Has Two Sides
Competition is making the cost of AI tokens fall, even as customers' total bills rise. Anthropic cut the price of Claude Opus 5.5 to $4 per million input tokens and $20 per million output tokens, a 20% reduction. About 90 minutes later, OpenAI followed by offering GPT-6 Sol and Luna at roughly half the price of its previous model.
Cheaper tokens do not always lead to lower bills. Gartner predicts that the cost of running AI agents through a workflow will increase more than fivefold by 2028. This is because better and cheaper models encourage companies to automate more tasks and let AI agents run for longer.
Apple hopes some customers will choose to buy computer hardware instead of paying for ongoing AI use. On Sept. 22, it began shipping Mac mini and Mac Studio desktop computers that it promotes based on their running costs, not just their speed. Apple says buying a computer once lets users run AI models on their own devices "without counting tokens or worrying about rising cloud costs." The Mac mini starts at $899. The Mac Studio with the M5 Ultra chip costs $5,499, while fully upgraded versions can cost close to $20,000. Apple says it connected four Mac Studio computers to run a model with one trillion parameters using power from a single wall socket.
However, not everyone can easily switch to Apple computers. According to IDC data cited by Reuters, Apple has only 4.6% of the business desktop computer market, compared with 91.3% for Windows. Microsoft is also trying to attract these customers. Chief Executive Satya Nadella has described AI that runs directly on a device as "unmetered intelligence." Microsoft has launched a Surface Laptop Ultra priced at $2,599 and a Surface RTX spark development computer with 128GB of memory priced at $5,999.
Apple is also using its supply-chain strategy to manage the cost of AI itself. It has set daily limits on features that rely on models running on remote servers and plans to offer paid access to extra capacity. It has also arranged partnerships with model providers, including Google for cloud-based models and Alibaba and Baidu for the Chinese market. Most of Apple's own AI development work focuses on models that run directly on its devices.
A Market That Is Testing People's Patience
The debate over spending is happening as people take a closer look at the business costs of AI. S&P expects six major U.S. hyperscalers — Amazon, Microsoft, Alphabet, Oracle, Meta Platforms (META) and SpaceX — to spend more than $7 trillion on data centers and other AI-related investments from 2025 to 2030. It also expects these companies to have negative free cash flow at least through next year, meaning they will spend more cash than they bring in after operating costs and investments.
Apple's yearly spending related to AI is estimated at $11 billion to $13 billion. Some investors see the company as a safer investment choice. Its shares have risen about 25% in 2026, bringing its market value to roughly $5 trillion.
For individual users, the key question is whether each new AI fee provides enough value to justify its cost compared with a general-purpose chatbot. Tasks such as summarizing a PDF or turning a meeting recording into text can often be done by giving the file to an AI service someone already pays for. However, special-purpose tools can save time and offer businesses extra controls to manage how the tools are used.
Analysts expect many AI services to disappear as the market changes. As happened with cloud storage, many paid AI add-ons may be discontinued, while features that cost extra today may become standard and free within a few years. The analysis concludes that the most important factor is not how advanced an AI model is, but how much useful value it provides for its price.