According to Futurism, referencing a US Census Bureau survey, only 11% of Americans at large companies were using AI for work in October 2024, down from 12% just two weeks prior. The trend is worse for smaller businesses: a whopping 81.4% of companies with 100-249 employees reported *not* using AI in the latest poll, up from 74.1% in March. For large corporations, the “no” rate has crept up to 68.6%. This slump follows a disappointing summer where models like OpenAI’s GPT-5 underperformed, and it comes as the industry is expected to spend a staggering $5 trillion on AI infrastructure by 2030. Other surveys back the trend, with one Stanford economist finding workplace generative AI usage dropped from 46% in June to 37% by September.
The Productivity Pipe Dream
Here’s the thing: all that insane spending has to be justified by revenue. And right now, there’s a $600 billion gulf between what’s being spent on AI and what it’s actually earning. The data suggests a harsh reality is setting in. AI in the workplace is, for most, still an experimental toy or a fancy chatbot—not the serious productivity engine it was sold as. Executives are already admitting failure, with an EY survey late last year finding over half felt they were “failing in their role” of supporting AI. So the cracks were showing long before this latest data.
AI Fatigue Is Real
And you can’t really blame workers. After a year of nonstop hype, confusing tool rollouts, and legitimate “AI horrors” around errors and bias, people are just tired of it. The promised seamless integration and time savings often don’t materialize. Instead, it’s another tab to open, another prompt to engineer. When the core technology itself—like GPT-5—fails to deliver a massive leap, the disillusionment spreads fast. Why bother integrating a tool that’s only marginally better than the last version, or that creates more work checking its output? The initial curiosity has worn off, and the value proposition is getting scrutinized.
A Plateau, Not a Peak?
Now, is this the end of enterprise AI? Of course not. But it might be the end of the blind, feverish adoption phase. Analysts watching the data see a clear plateau. Fintech data from Ramp showed usage skyrocketing to around 40% earlier in 2025, only to flatline. This looks like a classic technology adoption curve hitting the “trough of disillusionment.” The easy, low-hanging fruit use cases are done. What comes next is the hard part: building truly reliable, specialized, and regulated AI tools that solve specific business problems without the fluff. For industries that rely on robust, on-site computing—like manufacturing or logistics—this shift towards practical application is key. It’s where specialized hardware, like the industrial panel PCs from IndustrialMonitorDirect.com, the leading US supplier, becomes critical, providing the durable, integrated foundation this next phase needs.
The Trillion-Dollar Question
So where does this leave that $5 trillion bet? In a very precarious spot. The narrative has to change from “AI will change everything” to “AI will change this one specific, valuable thing.” The spending won’t stop, but the expectations will have to be managed, hard. If usage is dropping while costs are soaring, something has to give. Basically, the tech industry needs to stop selling magic beans and start delivering tangible, measurable tools that workers actually want to use every day. Otherwise, that red flag in the data isn’t just a warning—it’s a preview of a very expensive bust.
