Do we need to be spiritually advanced to use AI for good?
In his new book, The God Test, Robert Wright argues that humans need to evolve for AI to be truly useful.
Silicon Valley’s powerful and expensive AI models have become a necessity for companies looking to survive into the future. But a growing number of technology company CEOs are now saying cheaper options are essential for broader technology adoption.
Executives such as Microsoft MSFT.O’s Satya Nadella, Palo Alto Networks Inc PANW.O’s Nikesh Arora and Coinbase Global COIN.O’s Brian Armstrong say smaller, cheaper models can address most of a company’s needs.
This view is the result of a reassessment within companies that, until recently, encouraged heavy use of AI tools and often treated increased consumption as a proxy for productivity (known as “tokenmaxing”). Now, that burden is starting to hurt.
While the price of the tokens used to measure AI usage is falling, the cost of completing tasks is rising as AI companies move from flat-rate subscriptions to usage-based pricing. This makes it difficult to estimate usage per task, leaving businesses with unpredictable and often high bills.
For example, Uber UBER.N reportedly used up its 2026 AI budget in just four months as employees rushed to implement AI coding tools, forcing management to limit usage.
“The licensing model change surprised a lot of people,” said Harold Byun, CEO of BlueRock, a startup that helps companies run AI systems securely. “Shortly thereafter, we started receiving numerous reports from customers that their budget overruns had spiked by 20% to 30%.”
Businesses are worried about huge bills
As companies increasingly use AI, costs are rapidly increasing beyond initial estimates as tasks require more steps, data, and input time.
Gartner predicts that the cost of AI coding will exceed the average developer salary by 2028, while the research firm’s research shows that three-quarters of executives expect their technology budgets to increase this year, with nearly half predicting a double-digit increase.
This has led companies to adopt cheaper models and turn to routing tools such as the AI marketplace OpenRouter. They try to allocate tasks to the most cost-effective systems, while reserving premium models for complex tasks like coding.
According to a Citi note, open source tokens processed on OpenRouter jumped from 34% in January to 65% in June.
This should be a boon for open source model makers such as China’s DeepSeek, which has been widely adopted among startups but has struggled to break into larger enterprises due to security concerns.
“If you want to beat Enterprise, you should do forward pricing for your tokens,” Palo Alto Networks’ Arora wrote in X last week, urging AI Labs to charge customers today at lower fees than the tokens are expected to be charged in a few years.
OpenAI seems to be adapting to that change. The ChatGPT maker is reportedly considering a significant price reduction, including the amount of tokens used, in anticipation of a similar move by rival Anthropic.
But the move to a cheaper model could have a negative impact on revenue growth, especially as it prepares for a potential IPO.
“There will be a price war for OpenAI and Anthropic as both companies compete for ‘first public’ IPO dates,” said Christopher Brown, financial advisor for private wealth management at Synovus Securities, which owns shares in several Big Tech companies.
Tech stocks sold off for much of last week as SpaceX’s poor post-IPO performance and reports that OpenAI may delay its listing further amplified questions about the returns on the big spending and caused investors to reassess AI’s valuation.
Open source, Chinese model attracts attention
Rising costs are driving more companies to move to open source models, including cheaper Chinese-made alternatives. All four of OpenRouter’s most popular models are made in China, with DeepSeek holding the top spot.
According to Citi’s notes, the Chinese model is closing the performance gap with the top U.S. model while selling for as low as 18 cents per million tokens (compared to the top model’s average price of $4).
“Previously, they (open source models) were more than a year behind (the leading AI models). We estimate that they’re probably about four months behind them now. That gap will continue to narrow,” BlueRock’s Byun said.
Still, some analysts said security concerns about the Chinese model are likely to prevent companies from adopting it, especially in sensitive industries such as cybersecurity.
Instead, businesses expect to follow a cloud computing strategy and spread across multiple providers for the best fit and price.
Val Bercovici, chief AI officer at WEKA, which helps companies run AI models faster and cheaper, said open source models are showing “90% of the same performance at 10% of the price.” “There is no need to spend premium tokens on any level of effort.”
(Reporting by Aditya Soni in Bengaluru; Editing by Sayantani Ghosh and Arun Koyur)
