WSJ: Companies Shift from Encouraging AI Token Maximization to Thrift and Efficiency Optimization Due to Rising Costs
The Wall Street Journal reports that the era of AI "tokenmaxxing" in U.S. companies has ended, and firms are now embracing "thrift-maxxing." Previously, companies encouraged employees to maximize the use of AI tokens to drive adoption, but now they are shifting towards cost-saving and efficiency optimization due to rising costs.
AI usage is transitioning from unrestricted expansion to cost control, with funding moving from high token consumption to more economical models and tools; the driving force is billing pressure, benefiting low-cost AI providers while high-consumption companies face internal budget constraints.
Source: Public Information
ABAB AI Insight
Early on, companies promoted extensive use of tokens for models like OpenAI and Anthropic through leaderboards and incentives, creating a "tokenmaxxing" culture to accelerate AI integration into workflows; however, actual bills exceeded budgets within months, leading Meta, Uber, Walmart, and others to set usage limits and eliminate leaderboards.
The shift to thrift-maxxing emphasizes unit economics and value output, motivated by the need to control exponential cost growth; this mirrors the early path of cloud computing from unlimited use to refined management, with funding returning from subsidized adoption to ROI validation.
This transition is similar to other technological waves moving from hype to practical implementation; currently, enterprise AI is at a juncture of transitioning from experimental expansion to sustainable operations.
Essentially, this is a capital path adjustment: high-consumption models are unsustainable, driving a reconstruction from "the more you use, the better" to "the more value you get, the better," accelerating the rise of low-cost models and internal optimization tools.
ABAB News · Cognitive Laws
- There’s always a bill for a free lunch.
- Maximizing usage ultimately hits a cost wall.
- Thrift is the real threshold for AI implementation.