As businesses reflect on their investments and approach to implementing AI, the landscape is changing dramatically. For years, the Silicon Valley doctrine has been that only the most advanced and costly AI models would offer business value to companies looking to stay competitive. But a major change is taking place, with an increasing number of top tech executives saying that alternative technologies that are more cost-effective and efficient could unlock the potential for AI to become more widely used, spanning industries.
This re-evaluation is due to a paradigm shift in the way businesses perceive their AI investments. Until recently, a lot of organisations were actively promoting the use of AI tools at scale and the more they consumed, the more productivity they were gaining. It was dubbed “tokenmaxxing” by tech leaders, referring to the fact that they used to make their decisions based on the mere number of times they could use AI instead of its actual value. However, the start of the realities of these practices is now starting to bite hard and a major strategic shift is beginning to take place.
AI pricing models have changed significantly, making it difficult for businesses to determine the cost of AI services. The price of each token remains on decline, but the price of every task has gone up due to the shift in pricing from flat subscription fees to usage. The price of each token is still declining, but the overall cost of each task has increased due to AI firms’ move towards “usage” pricing models. This change has put a lot of uncertainty in corporate budgets, as companies have a hard time estimating resources needed on specific tasks. This compound impact has shown up especially for companies that were excited about using AI tools and have not taken the long-term financial impacts into account.
One such example is Uber, whose staff enthusiastically embraced AI-coding tools, consuming all the company’s set budget for the year of 2026 in the span of four months. This unexpected increase in usage led to a cap in use of the AI tools, demonstrating the gap between hype and real-world expense. This has become a more common situation across sectors, as companies start to face the financial implications of unchecked use of AI.
Harold Byun, CEO of a start-up called BlueRock, which assists businesses with the deployment of AI systems securely, noted that the price shift came as a surprise to many companies. Byun said that his company has heard from customers that they were in budget overruns by 20 to 30 percent after the price hikes. This feedback is an example of how rapidly AI costs can climb when you move to usage-based billing.
Gartner projects that by 2028, the cost of using AI coding will exceed the average developer’s salary, causing a ripple effect across technology departments.AI coding costs will be higher than the average developer’s salary by 2028, according to research from Gartner, which has sent shock waves through technology departments. Three-quarters of executives are predicting technology budgets will increase this year, with almost half in double-digit numbers, according to a survey by the research firm. The numbers highlight the escalating conflict between the promise and costs of AI in organizations.
Businesses are increasingly thoughtful in their buying of AI, due to these pressures. Many are pursuing routing solutions such as OpenRouter, an AI marketplace where companies can leverage the least expensive routing system for their needs while reserving high-end routing solutions for more complex assignments like coding or advanced analytics. It enables businesses to make the most of their AI investments without compromising on key application performance.
OpenRouter’s data showed that the percentage of open-source tokens used rose from 34 percent in January to 65 percent in June, mirroring this trend toward cost-conscious AI adoption, according to a Citi research note. This dramatic rise indicates that companies are actively searching for other solutions to their high-cost proprietary offerings – especially in routine situations where sophisticated features might not be needed.
This shift has created opportunities for Chinese AI firms, such as DeepSeek, with models that come at much lower cost. Some Chinese models cost as low as eighteen cents per million tokens, compared with an average of four dollars for top-tier Chinese models offered by the Americans, cited Citi analysts. This cost gap has drawn more and more businesses to open-source solutions, especially startups and smaller companies with smaller budgets.
But large companies have been held back by security issues when it comes to open-source models. Though they have become popular among startups and developers, larger corporations have hesitated to use these models in their work because of the potential risks of data privacy and compliance. This is where AI labs with the ability to solve these security issues, and do so competitively, can come in.
Recently, Palo Alto Networks CEO Nikesh Arora spoke about this challenge on social media, saying that the solution is to think differently about pricing models for AI labs to engage enterprise customers: forward pricing. Companies should offer services to their customers at today’s rates, and then impose a higher price next year when the tokens will be worth less, Arora said. This view is indicative of a wider understanding that price flexibility will be critical to the enterprise market.
The advances of AI pricing models are linked to critical factors of the future direction of the industry. On the other hand, superior models come with an unmatched set of features that could be necessary for specific uses. However, the increasing number of affordable alternatives could mean that the market is shifting towards a more segmented strategy that lets businesses pick the model that is best suited to their requirements and cost constraints.
The real challenge for technology leaders is to innovate balancing the cost of doing so. The advanced capabilities of AI models can offer a tremendous competitive edge, but the associated expenses may not always be worth it for specific applications. As businesses keep trying out AI tools, they’re finding that they can meet a significant portion of their demands with cheaper options and keep the more costly ones for truly mission-critical use.
