The software industry is grappling with soaring costs associated with artificial intelligence (AI) tokens. Major AI service providers have released high-performance models, leading some companies to spend nearly 100 million won per month on tokens.
As of June 25, the IT sector reports that businesses are rapidly increasing their token expenditures as they integrate AI agents into both development and non-development tasks.
Company A, a software development firm, revealed that its monthly AI usage costs exceed its initial budget by 30%. Although the budget was set conservatively, a surge in exchange rates and higher-than-expected token usage have contributed to the increase.
On June 24, the exchange rate for the won to the dollar was 1,546 won, up about 7% from the mid-1,400s six months ago. This means companies must spend an additional 7% for the same tasks. Furthermore, developers are required to read, infer, modify, and repeatedly validate extensive codebases, leading to a significant rise in token consumption.
A representative from Company A stated, "Despite not setting a conservative monthly budget, we are using more tokens than anticipated. The rapid pace of AI agent version updates complicates budgeting. We plan to review methods to minimize and optimize token usage, starting with prompt strategies, and will issue a company-wide notice afterward."
Game developer Company B reported daily token costs ranging from $150 to $400 per developer. While this amount currently falls short of individual labor costs, there is a possibility that it could soon exceed them, prompting the company to seek solutions.
Discussions are underway to distribute guidelines aimed at reducing token usage and to implement departmental limits on token consumption. However, the varying AI models and performance levels used for different tasks make it challenging to predict costs accurately, complicating the creation of specific guidelines.
Market research firm IDC forecasts that by 2027, the usage of AI agents will increase tenfold, with token call volumes expected to rise by 100 times. The total annual demand for tokens is projected to grow from approximately 100 quintillion this year to about 4 quattuordecillion by 2030, reflecting an average annual growth rate of around 150%. This translates to a yearly increase in token costs by a factor of 2.5.
The surge in per-person token usage is becoming a significant concern for companies. Users accustomed to conversational AI are now entering longer and more complex prompts than before, leading to explosive increases in token consumption. Additionally, as tasks such as complex reasoning, collaboration, and coding are delegated to AI, per-person token usage continues to rise.
The ICT sector observes that corporate interest in AI has shifted to the stage of 'cost management.' A representative from Company B remarked, "While AI undoubtedly enhances development productivity, the cost structure does not allow for unlimited usage. We are now considering feasible cost-saving measures at the corporate level."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.