Many Companies Begin to Question Whether Large-Scale AI Deployment Truly Delivers Returns
Many companies are starting to question whether large-scale AI deployment truly delivers returns. Microsoft has canceled most Claude Code licenses, primarily due to high costs; Uber's COO has publicly stated that AI spending is becoming increasingly difficult to justify.
An AI consultant revealed that one client spent as much as $500 million in a single month due to unrestricted employee use of Claude. The industry refers to many companies falling into "tokenmaxxing," where blind AI adoption by all employees leads to skyrocketing IT costs.
In terms of market mechanisms, corporate IT budget decision-makers are accelerating cuts to non-core AI token consumption and shifting towards strict ROI assessments; event-driven funding is flowing from generalized AI applications to more mature scenarios like programming; companies focusing on enterprise-level AI optimization and programming tools are benefiting, while generalized AI subscription service providers and high token consumption platforms are under pressure.
Source: Public Information
ABAB AI Insight
Microsoft previously made large-scale purchases of OpenAI and Anthropic services to drive corporate AI transformation, but quickly discovered insufficient ROI in non-programming scenarios during actual deployment. Similar adjustments to the Azure OpenAI budget were made multiple times from 2023 to 2025, reflecting its consistent cautious approach to cost control.
In terms of capital pathways, companies are shifting AI budgets from generalized use across the board to high-value scenarios like programming, releasing cash flow through layoffs to cover AI bills, while demanding suppliers provide more transparent token consumption control and tiered pricing, reallocating resources from high-cost experimental deployments to quantifiable productivity gains.
This mirrors the "tool fatigue" and cost reflections seen after large-scale SaaS deployments in the 2010s, as well as the optimization wave following early excessive cloud computing subscriptions; current corporate AI adoption is transitioning from fervent expansion to rational ROI validation.
Essentially, this represents capital concentration, as tightening budgets and layoffs focus AI benefits from dispersed organizational consumption to a few high-return scenarios (mainly programming). The mechanism is that under the token economic model, simple queries can quickly accumulate substantial costs, forcing companies to reassess AI as a productivity tool rather than a universal solution.
ABAB News · Cognitive Law
The more powerful the technology, the more companies must learn to account for every token.
When AI costs are covered by layoffs, truly mature applications are actually few and far between.
The outcome of blind tokenmaxxing is always high bills and a return to rationality.