Tokenomics Trend: Global Companies Begin to Abandon Expensive AI Models
The global corporate world is now facing a radical shift in the use of artificial intelligence (AI). After previously being trapped in the trend of using the most advanced AI models at exorbitant costs, many companies are now beginning to realise that they do not need to drain their budgets for such technology.
This phenomenon, referred to as tokenomics, has emerged from corporate fatigue over AI operational costs that continue to swell. As a solution, both large and small companies are starting to adopt more affordable AI models, including open-weight models developed by Chinese startups.
Mike Saeks, Field Chief Technology Officer at Cursor, a startup currently in the process of being acquired by SpaceX for Rp930 trillion ($60 billion), explained that using the most powerful AI models for routine tasks is wasteful. “It is like driving a Lamborghini just to go to the corner shop to buy milk, when the car was designed to be raced on a track,” said Saeks. He now advises companies on how to measure and improve the return on investment (ROI) of their AI spending.
This shift in mindset is drastic. If a few months ago companies boasted about high AI token usage (tokenmaxxing), they are now switching to a strategy of maximum thrift (thrift-maxxing).
This budget shift is not merely a financial matter, but also a geopolitical issue. Prominent US AI models such as OpenAI and Anthropic are generally closed models and tightly controlled. In contrast, China is known for open-weight models that are cheaper and can be customised.
Several Chinese startups that are increasingly being used in the US include DeepSeek, MiniMax, and Moonshot AI. Nevertheless, the use of these models has sparked tension. OpenAI and Anthropic accuse Chinese developers of copying their technology, while US government officials are beginning to voice concerns about data security risks.
On the other hand, technology giants such as Nvidia, Microsoft, and Palantir have signed letters of support for open models, urging policymakers to be cautious in implementing restrictions that could hamper competition.
Many companies are now implementing an à la carte strategy, mixing the use of various AI models according to task requirements. Cursor, for example, conducted an experiment building a web browser from scratch. If using GPT-5.5 fully, the cost would reach more than Rp155 million ($10,000). However, by combining their internal model and Anthropic Opus 4.8, the cost was slashed to only around Rp20.7 million ($1,339).
Facing threats to their valuations, OpenAI and Anthropic have begun to react. OpenAI introduced the GPT 5.6 Sol model, which is claimed to be more efficient in token usage. Meanwhile, Anthropic released a high-powered model at a lower cost to retain customers.
Marty Kausas, CEO of Pylon, described the current situation as a bloodbath for customer loyalty. His company even received incentives in the form of free tokens worth more than Rp24.8 billion ($1.6 million) from AI vendors desperately trying to prevent customers from switching to cheaper competitor models.
With the growing number of available models, the era of cheap AI appears set to persist. This is forcing industry leaders to no longer merely pursue sophistication, but also genuine cost efficiency.