After two years of urging employees to adopt generative AI at every opportunity, a growing roster of major companies is now pulling back. The reason, according to a Financial Times investigation published June 19, 2026, is bluntly summarized in a quote from one corporate insider: "We created a monster."
The FT report, which quickly became one of the most-discussed business stories of the week, documents how runaway AI spending, much of it metered in tokens billed by AI providers, is straining corporate budgets and forcing executives to impose discipline on tools they had only recently mandated. For more context on this story, see our ongoing more AI stories.
From Mandate to Restraint
The shift is striking because it reverses course on policies many large employers had championed. As CBC reported on June 17, the stated goal at numerous companies had been to "use as much AI as possible." Now, some of those same organizations are pulling back, capping usage, restricting access to premium models, and auditing which AI workflows actually deliver value.
Crypto Briefing, citing the FT reporting, named specific companies curbing employee AI use as costs surge, including Amazon, Walmart, and Uber. The headline capture, that three of the most sophisticated technology employers in the world are tightening the tap on AI spending, signals that the cost problem is not confined to cash-strapped startups. It reaches the largest enterprises on the planet.
The Token Bill Problem
At the heart of the pullback is the pricing model underpinning most generative AI. Rather than a flat subscription, many AI services charge by the token, the chunks of text a model reads and produces. When tens of thousands of employees paste lengthy documents into chatbots, run AI coding assistants all day, or build internal agents that loop through many model calls, the token meter spins fast.
A separate analysis from Let's Data Science, published June 17, focused on the phenomenon of companies reining in AI token spending specifically for engineers. Developer tools, which can issue dozens of model requests per task, are among the most expensive AI deployments inside large companies. As more code-generation and agentic tools move into production, the cost of keeping them running has climbed sharply.
The challenge is compounded by the difficulty of measuring return on investment. While individual employees often report productivity gains, attributing those gains to specific dollars of token spend, and proving they exceed the bill, has proven harder than many executives expected.
Even the AI Builders Feel the Pinch
The cost squeeze is not limited to AI consumers. It is touching the companies building the technology. On June 12, Stocktwits reported that Meta's AI costs had "blown up," prompting the company to outline plans to rein in spending. For a firm that has positioned AI as central to its future, acknowledging cost pressure underscores how widely the discipline has spread.
The pattern points to a maturing market. In 2024 and 2025, the dominant corporate instinct was to experiment broadly and worry about the bill later, on the assumption that early AI adoption would confer a durable competitive advantage. In 2026, with budgets tightening and the novelty fading, finance departments are demanding the same rigor applied to any other line item.
A New Market for Cost Controls
The pullback is itself creating commercial opportunities. On June 4, CNBC reported that expense-management startup Ramp reached a $44 billion valuation as companies look to rein in AI spending. Ramp's ascent illustrates how quickly the pain point has matured into a product category: tools that monitor, allocate, and cap AI expenditures are now in demand alongside the AI tools themselves.
That dynamic, providers racing to sell AI, and a parallel industry racing to help buyers control what they spend on it, captures the awkward middle phase the enterprise AI market has entered. The technology has cleared the bar of usefulness for many tasks, but not yet the bar of predictable, controllable cost.
What It Means for the Industry
For AI providers, the trend carries a warning. If large customers move from unlimited experimentation to rationed, ROI-tested usage, revenue growth tied to raw token consumption could slow. That prospect is particularly consequential for companies whose financial models assume ever-rising inference volumes.
For enterprises, the lesson is more nuanced. The pullback does not necessarily mean AI is failing inside big companies. Several of the organizations tightening oversight are simultaneously expanding targeted, high-value AI deployments. The shift is from breadth to selectivity, from "use AI everywhere" to "use AI where it pays."
The quote that anchors the FT report, "We created a monster," may ultimately describe a temporary excess rather than a permanent verdict. Companies raced to embed generative AI broadly, generated large bills in the process, and are now correcting course. Whether that correction slows the AI rollout or simply makes it more sustainable is the question the industry will spend the rest of 2026 answering.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
