Corporations are looking to offload AI tasks to cheaper models as usage blows out IT budgets and returns on investment haven't solidified.Why it matters: The search for cheaper subscriptions could threaten the three biggest AI labs' near trillion-dollar valuations right as they near record IPOs.Driving the news: Anthropic raised $65 billion in its latest funding round, pushing its valuation to $965 billion Thursday and eclipsing OpenAI's latest valuation of $730 billion. This comes as executives are telling Axios they're increasingly concerned about their AI bills. Some are closely monitoring usage or even switching to cheaper models to keep costs down. That tension could pressure the revenue projections of these AI companies just as they're expected to go public. What they're saying: "There are many tasks you don't need [Claude's] Opus for," Matan Grinberg, CEO at Factory, whose proprietary router selects the most cost-effective AI model for each query and task, told Axios. Clients are "being much more surgical" about AI usage, said Eric Yunag, EVP at Convergint, a tech security firm.Others are switching to open-source models or agents built for specific use cases, which are often cheaper and better performing, Ali Ansari, CEO of model training firm Micro1, told Axios.Follow the money: Grinberg said his customers are "really scared" of committing to a single vendor. "No one wants to standardize on OpenAI, or just Anthropic or just Google," because they don't want to be subject to "price gouging" down the line.Open model use at Factory has tripled in the last month relative to closed models, like those offered by OpenAI and Anthropic, which can be more expensive. Yes, but: Once you have skills and tasks set up within one AI platform, it can be hard to switch to another one, even if the monthly IT bills are painful. And enterprise revenue for the frontier AI labs continues to tick higher. The bottom line: Just as Anthropic is on pace to top $47 billion in annual revenue, its customers are searching for cheaper alternatives.