AI Costs Explode: Companies Block Access or Advise Older Models
Just a few months ago, some corporations were pushing their employees to have AI models do as much as possible. Now, a reversal is underway.
(Image: Proxima Studio/Shutterstock.com)
Several major US corporations are trying to curb their employees' use of AI or at least steer them towards less powerful models for certain tasks to reduce costs. This is reported by the US magazine 404 Media, citing internal documents and reports from various companies, including Atlassian, Adobe, and Amazon. For example, the financial group Citi temporarily blocked access to the newer AI models from Anthropic and OpenAI for several days and recommended via email which model should be used for which tasks. Citi has since denied this step to 404 Media, “although the email and other screenshots clearly show Citi blocking access to certain models”.
“Crazy” Reversal
According to the report, Atlassian recently stopped unlimited AI use by employees and instead launched an internal overview where they can see how much their AI usage costs the company. According to internal data, the costs for AI tools at the company have tripled from 5 million US dollars, and 120 million are expected to be accumulated throughout the year. Atlassian has denied this, but did not specify which of the figures originating from internal documents are incorrect. At Adobe, unlimited use was reportedly discontinued at the end of June, with “employees essentially being told to try to get everything they can done before that date instead”.
At Amazon, shortly after the deactivation of an internal ranking for AI usage – which encouraged maximum AI utilization – a token limit was introduced: “Crazy, we go from no more leaderboard to actual usage limits in two weeks,” 404 Media quotes from an internal exchange on Slack. However, Amazon claimed upon inquiry that the internal guidelines for AI usage had not changed. Meanwhile, the magazine quotes an anonymous person from an unnamed media company, where the monthly AI limit for ChatGPT was reached for the first time. Almost half of it was attributed to a single person who did not achieve any obvious return on investment with it.
Videos by heise
Overall, the report indicates that more and more companies are scrutinizing how employees use AI technology. Instead of pushing for more AI usage, the aim is now to reduce costs. At Citi, employees are explicitly encouraged to always consider whether others within the company could better utilize AI resources. Just a few days ago, 404 Media reported that the consulting firm Accenture found that AI tools are more often used for simple tasks like converting PDF files into presentation slides than for demanding coding tasks. For the boom in AI usage, “non-technical staff burning through tokens for non-specialized tasks” would be responsible to a not insignificant extent.
(mho)
