Tech companies are touting new AI technologies that can generate business memos or computer code, but they are still figuring out how these products will generate profits.

Generative AI tools are untested, expensive to run, require powerful servers and expensive chips, and consume a lot of power. Microsoft (MSFT), Google (GOOG), Adobe (ADBE), and other tech companies investing in AI are experimenting with various strategies to develop, market, and generate revenue.

According to a person familiar with the matter, one of Microsoft's first generative AI products has lost money. At present, Microsoft and Google are introducing upgraded versions of software with AI support, and the fees are also increased. Zoom Video Communications (ZM) tries to reduce costs by occasionally using simpler AI that it has developed itself. Adobe and others set monthly usage caps and charge based on usage.

"A lot of customers I talk to are unhappy with the cost of running some of these models," Adam Selipsky, chief executive of Amazon Web Services, Amazon.com's (AMZN) cloud computing unit, said of the industry at large.

Chris Young, Microsoft's head of corporate strategy, said it will take time for businesses and consumers to figure out how they want to use AI and how much they are willing to pay for it.

"Clearly, we now have to translate that enthusiasm and intense interest into practical applications," he said.

Building and training AI products can take years and hundreds of millions of dollars, costing more than other types of software.

Artificial intelligence (AI) can draft emails, presentations, generate images, videos, first drafts of product designs, and perform data analysis and other more complex tasks. A Goldman Sachs study claims that 25 percent of jobs in the United States can be performed by AI. A report in May said nearly 4,000 tech workers have been laid off over AI, while some ceos are changing their hiring plans for the future. WSJ explains the changes taking place.

AI products often don't have the economies of scale of standard software because they may require a lot of new calculations for each question asked. The more customers use AI products, the more expensive it will be for businesses to pay for infrastructure. These running costs expose companies that charge a flat fee for their AI products to potential losses.

Microsoft uses AI technology from partner OpenAI to launch GitHub Copilot, a service that helps programmers write, fix, and transform code. This service is popular with programmers because it can greatly reduce the time and effort required to code. More than 1.5 million people have used the service, and nearly half of Copilot's user code is written by the service.

But the service has been losing money because of its high operating costs.

The AI assistant costs individuals $10 a month to subscribe to. In the first few months of this year, Microsoft spent an average of more than $20 per user per month, according to a person familiar with the data. Some users cost Microsoft as much as $80 a month, the person said.

Microsoft and GitHub did not respond to requests for comment on whether the service is profitable. If computing costs come down, GitHub Copilot and other AI assistants could become more profitable.

Microsoft is raising prices for its next AI software upgrade. In addition to the regular monthly fee (the basic Microsoft 365 office suite for enterprise customers starts at about $13), the company will charge an additional $30 per month for the AI version. This AI-driven feature can independently compose emails, create PowerPoint presentations, and create Excel spreadsheets on instructions.

Google, which is about to launch a similar AI assistant feature for its office software, will also charge $30 a month on top of regular subscriptions starting at $6 a month.

Microsoft, Google and others have adopted a flat monthly fee, betting that this extra, higher fee will cover the average expense of running the technology.

Adobe leaders introduced Adobe Firefly, an AI image generator, at a summit in Las Vegas this year.

One reason such services are so expensive is that some use the most powerful AI models available today, which are more energy intensive and put more strain on computer processors than standard software or cloud services.

For example, Microsoft is using OpenAI's latest software to implement its AI capabilities. The version, called GPT-4, is one of the largest and most expensive AI models available.

Summing up an email with GPT-4 is like asking Lamborghini to deliver pizza.

"These services require a lot of computing power," says Jean-Manuel Izaret, head of the Boston Consulting Group's marketing, sales and pricing practice. "It takes a lot of intelligence."

According to people familiar with the matter, Microsoft has been exploring less powerful and less expensive AI tools for its Bing search engine, including some built using open source AI software from Meta Platforms.

Tech news outlet The Information earlier reported that Microsoft was considering other tools for Bing.

Zoom has developed a smaller, cheaper piece of software for its AI assistant, powered by several models including those developed by OpenAI and Meta. Zoom only uses the most powerful AI tools for the most difficult tasks.

Smita Hashim, Zoom's head of product, says the company doesn't charge extra for the tool, which can summarize meetings and compose chat messages, because it avoids using expensive AI as much as possible. Smita Hashim has also worked at Microsoft and Google.

"We were shocked by the high prices some of our competitors were asking," she said.

Adobe uses a points system to help ensure that its AI image generator, Firefly, doesn't lose money. Once Adobe customers use up their allotted monthly credit, the company slows down the service to prevent overuse.

"We're trying to deliver great value while also protecting ourselves on the cost side," Adobe Chief Executive Shantanu Narayen said.

The business community expects generative AI to become cheaper over time, just like many technologies such as cloud storage and 3D animation. The advent of new chips and other innovative technologies promises to reduce processing costs.

OpenAI earlier this year lowered the fees it charges to use its older version of AI. Its popular free chatbot, ChatGPT, uses an older version of the software. If customers want to use ChatGPT powered by the latest version of the software, they will pay $20 per month.

The lack of clarity about the profit model has deterred investors. Shares of AI-related companies have soared this year. OpenAI is in talks with investors to sell a stake that would value it at as much as $90 billion, about three times its valuation earlier this year.

May Habib, CEO of Writer, which builds generative AI tools for businesses, said the rising valuations reflect enthusiasm for the promise of AI. Executives will soon look more carefully at costs, she said.

Habib said she doesn't think investors will be as enthusiastic about generative AI next year.