As businesses begin to look at the returns on AI inputs, the “first-to-first-cost” procurement approach in the past is slowing down. Even cheaper open-source models continue to enter the business market, and Microsoft, Amazon and Google are introducing more cost-efficient products, which put OpenAI and Anthropic under new growth pressure.

Enterprise starts compressing AI bills

In the United States, AI Startup, CEO Lindy Flo Crivello, states that this month the company has switched all traffic from the Claude model of Anthropic to the open source weight model of the Chinese company DeepSeek. According to him, this adjustment has led to a marked decline in the cost curve, which could save millions of dollars in the coming months.

Criveello stated that Lindy would still invest a significant amount of its budget on AI, but that previous expenditures were no longer sustainable. He said that if Anthropic were to make subsequent price reductions, the company could also revert to Claude.

This change is not an example. The consulting firm Highspring, President Jeff Henry, stated that some clients were slowing down their expenditures and waiting for clearer returns on investment; there were also companies that were prepared to observe for another 12 to 18 months before deciding whether to expand their inputs.

Uber has also begun to limit the use of some AI tools. This month, the company indicated that a tiered standard of expenditure had been set for some of the tools, with a base slotting of $1,500 per month, and that a higher level of staff could be requested. Uber's Chief Technical Officer revealed in April this year that the company spent the entire AI budget for only four months.

New constraints behind high growth

OpenAI and Anthropic have been the main beneficiaries of this round of AI expansion of corporate spending. The CNBC quoted data stating that the annualized income performance rate disclosed by Anthropic in May was $47 billion; the annualized income performance rate of OpenAI earlier this year was close to $25 billion.

According to D.A. Davidson analyst Gil Luria, the current growth rate of the two companies is likely to be close to high. One reason for this is the natural slowdown in the increase in the base figure, while another layer of pressure comes from large business clients who begin to limit out-of-control token spending.

It was reported that OpenAI and Anthropic had submitted their listing applications in confidence in early June. Against a backdrop of prudent corporate budgets, moving forward with IPOs as soon as possible could help them to complete their listing while income data remain strong.

Open source model and accelerated diversion of large plant products

Businesses control costs while adjusting the way models are used. Some companies have introduced “model paths”, i.e. matching different models to the complexity of the task and entrusting simple tasks to cheaper models rather than always calling on state-of-the-art forward models.

Enterprise AI AISquared CEO Darren Kimura stated that the use of high-end models for simple tasks was nearing the peak of expenditure and was difficult to sustain in the long run. Green CEO Arvind Jain stated that about 95 per cent of the current AI use by enterprises is still on front-line models, indicating that the cost-optimization space is still significant.

In the face of this trend, OpenAI and Anthropic are also completing their business management tools. OpenAI introduced new analysis and control functions this month, allowing managers to split team credit lines, set a ceiling on use and present available budgets to staff. Anthropic has also introduced user configuration, analysis and expenditure limits.

At the same time, Microsoft, Amazon and Google are accelerating the introduction of lower-cost business models. Microsoft released a set of low-price models this month and stressed that GitHub Copilot would automatically match a suitable model for different tasks. The CEO of Microsoft Satya Nadella also stated that the industry should not over-centralize value in the hands of a few model providers.

The Amazon executive Peter DeSantis also expressed this month that the company wants to be able to compete with OpenAI and Anthropic in the front-line model area in the coming year. As firms move from “compelling power” to “costing”, the competition focus in AI model markets is expanding from performance to price and efficiency.