How Companies Are Testing Token Economics in Corporate AI
Photo: Wired
Quick answer
Companies are adopting generative AI but encounter high token costs—the units of data processing. Expenses are rising faster than anticipated, forcing a reevaluation of AI strategies and the search for cost-saving…
Corporate adoption of generative artificial intelligence has encountered an unexpected challenge: the rapid rise in token costs. Representatives from a Silicon Valley software company and a major e-commerce retailer revealed that data processing expenses have exceeded initial forecasts, raising questions about the profitability of AI projects.
The issue is compounded by the fact that AI models, such as GPT or similar, are priced based on the number of tokens—the smallest units of text the system processes. The more data fed into the system, the higher the cost. With widespread AI adoption, this leads to exponential cost growth, particularly for companies handling large data volumes.
Experts note that businesses have yet to develop effective strategies for managing "tokenomics." Some companies are experimenting with reducing query volumes, while others seek alternative models with lower token costs. However, such measures may compromise result quality, which is unacceptable for mission-critical processes.
In the long term, optimizing token costs will be key to the sustainable development of corporate AI. Without addressing this issue, even tech leaders risk facing financial constraints that could slow innovation.
Common questions
- What are tokens in the context of AI?
- Tokens are the smallest units of text processed by an AI model. They can be words, symbols, or parts of words. The cost of using AI often depends on the number of tokens.
- Why are tokens becoming a problem for businesses?
- The cost of processing tokens increases with data volume, making AI usage expensive. Companies must balance result quality with costs.
- How are companies optimizing token costs?
- Optimization includes reducing input data volumes, using more efficient models, caching results, and revising business processes to lower AI workload.
Dzen feed: /feed/dzen.xml · RSS: /feed.xml