Businesses opting for cheaper AI models as token bills climb for premium services

The shift reflects a broader reassessment among companies that had encouraged widespread AI adoption, often treating higher AI usage as a measure of productivity.

Businesses are increasingly shifting towards lower-cost artificial intelligence models as usage-based pricing makes premium AI services significantly more expensive, prompting companies to rethink their AI spending strategies, Reuters reports.Executives, including Microsoft CEO Satya Nadella, Palo Alto Networks CEO Nikesh Arora and Coinbase CEO Brian Armstrong, have argued that smaller, less expensive AI models can meet a large share of enterprise requirements without the high costs associated with frontier models.The shift reflects a broader reassessment among companies that had encouraged widespread AI adoption, often treating higher AI usage as a measure of productivity.

While the cost per AI token has declined, overall spending has increased as AI providers move from flat-fee subscriptions to usage-based pricing, making costs harder to predict.Uber, for example, reportedly exhausted its entire 2026 AI budget within the first four months of the year after employees rapidly adopted AI coding tools, forcing the company to impose usage limits.“Changing the license model caught a lot of people by surprise,” said Harold Byun, CEO of AI infrastructure startup BlueRock.

He said several customers reported AI spending exceeding budgets by 20 per cent to 30 per cent following the pricing changes.AI workloads drive higher infrastructure costsAs enterprises expand AI usage, tasks are becoming more complex, requiring larger datasets, longer prompts and additional processing steps, further driving up costs.Research firm Gartner estimates that spending on AI coding tools will exceed the average software developer’s salary by 2028.

Gartner also found that three-fourths of executives expect technology budgets to increase this year, with nearly half forecasting double-digit growth.The rising costs are encouraging companies to adopt cheaper AI models and use routing platforms such as OpenRouter, which direct workloads to the most cost-effective model while reserving premium systems for more demanding applications such as software development.According to a Citi note cited by Reuters, the share of open-source AI tokens processed through OpenRouter rose to 65 per cent in June from 34 per cent in January.The trend could benefit open-source AI developers, including China’s DeepSeek, whose models have gained popularity among startups despite facing security concerns in enterprise deployments.Nikesh Arora has urged AI developers to rethink pricing strategies, arguing that companies seeking enterprise adoption should lower token prices in anticipation of future cost declines.OpenAI is also reportedly considering substantial price reductions, including lower token costs, amid intensifying competition with rival Anthropic.Analysts, however, cautioned that aggressive price cuts could weigh on revenue growth, particularly as leading AI companies prepare for potential public listings.“There will be a price-war dynamic when it comes to OpenAI and Anthropic as they both duke it out for ‘first to public market’ IPO dates,” said Christopher Brown, a financial adviser at Synovus Securities.Investor concerns over returns on heavy AI spending have also pressured technology stocks, alongside reports that OpenAI may delay its IPO.Chinese open-source models gain groundMeanwhile, lower costs are drawing more businesses toward open-source AI models, particularly those developed in China.

Reuters cited a Citi note showing that the four most-used models on OpenRouter are Chinese, with DeepSeek leading the rankings.Chinese models are narrowing the performance gap with leading US systems while charging as little as 18 cents per million tokens, compared with roughly $4 per million tokens for top-tier models.Byun said the performance gap between open-source and leading proprietary models has narrowed significantly, estimating they now trail by about four months instead of more than a year.Even so, analysts expect security concerns to limit adoption of Chinese AI models in highly regulated industries such as cybersecurity.

Instead, many enterprises are likely to adopt a multi-model strategy, selecting AI providers based on performance, security and cost.Val Bercovici, chief AI officer at WEKA, said that open-source models now deliver “90 per cent as good as at 10 per cent of the price,” reducing the need to rely on expensive premium models for every workload.