As companies begin to scrutinize the return on their AI investments, the previous "scale first, worry about cost later" procurement approach is cooling down. Cheaper open-source models continue to enter the enterprise market, while Microsoft, Amazon, and Google are launching products that prioritize cost efficiency, placing new growth pressures on OpenAI and Anthropic.
Companies are cutting their AI bills.
Flo Crivello, CEO of U.S.-based AI startup Lindy, said the company has shifted all its traffic this month from Anthropic’s Claude model to DeepSeek’s open-source weight model, a Chinese company. According to him, this change has significantly lowered the cost curve, saving millions of dollars over the coming months.
Crivello said Lindy will still allocate a significant budget to AI, but previous spending has become unsustainable. He added that if Anthropic lowers its prices in the future, the company might reconsider using Claude.
This shift is not isolated. Jeff Henry, President of consulting firm Highspring, said some clients are slowing spending while waiting for clearer returns on investment; others are preparing to observe for another 12 to 18 months before deciding whether to increase their investments.
Uber has also begun limiting usage quotas for certain AI tools. The company stated this month that it has implemented tiered spending limits for some tools, with a base tier of $1,500 per month; employees requiring higher limits may apply separately. In April, Uber’s Chief Technology Officer revealed that the company had exhausted its entire annual AI budget in just four months.
New constraints are emerging behind high growth
OpenAI and Anthropic have been the primary beneficiaries of this wave of corporate AI spending expansion. According to CNBC, citing data, Anthropic disclosed an annualized revenue run rate of $47 billion in May; OpenAI’s annualized revenue run rate earlier this year approached $25 billion.
D.A. Davidson analyst Gil Luria believes that the current growth rates of both companies are likely nearing their peak. One reason is the natural slowdown due to a larger base, while additional pressure comes from enterprise clients beginning to curb runaway token spending.
The report noted that OpenAI and Anthropic secretly filed their IPO documents in early June. Amid cautious corporate budgets, moving forward with an IPO promptly could help them complete the offering while their revenue figures remain strong.
Open-source models and large-company products are accelerating market segmentation.
While controlling costs, companies are also adjusting how they use models. Some organizations are adopting "model routing," matching tasks to models based on complexity—assigning simpler tasks to more cost-effective models rather than routinely invoking the most advanced state-of-the-art models.
Darren Kimura, CEO of enterprise AI company AISquared, said that using high-end models for simple tasks has nearly reached spending peak and is unsustainable in the long term. Arvind Jain, CEO of Glean, noted that approximately 95% of current enterprise AI usage still runs on frontier models, indicating significant room for cost optimization.
In response to this trend, OpenAI and Anthropic are also enhancing their enterprise management tools. This month, OpenAI introduced new analytics and control features that allow administrators to split team credit spending, set usage limits, and display available budgets to employees. Anthropic has also rolled out features such as user configuration, analytics, and spending limits.
Meanwhile, Microsoft, Amazon, and Google are accelerating the launch of lower-cost enterprise models. Microsoft released a set of low-cost models this month and emphasized that GitHub Copilot will automatically match the appropriate model for different tasks. Microsoft CEO Satya Nadella also stated that the industry should not allow excessive value to concentrate in the hands of a few model providers.
Amazon executive Peter DeSantis also stated this month that the company aims to compete with OpenAI and Anthropic in the frontier models space over the next year. As enterprise procurement shifts from “competing on compute” to “optimizing cost,” the focus of competition in the AI model market is expanding from performance to price and efficiency.