Every year, Gartner publishes its Hype Cycle - a graphic representation of the maturity of emerging technologies. The form is always the same: technological trigger, peak of inflated expectations, trough of disillusionment, slope of enlightenment, plateau of productivity.
It’s a simplification. But it captures something real that the 60-year history of tech confirms.
The curve is not new to AI
AI is not in its first hype cycle. It’s been through several. The 1960s: first optimism, first broken promises, first winter. The 1980s: expert systems, LISP, massive investment in Japan, then the collapse. The 1990s-2000s: the Internet steals the show, AI becomes academic again.
Each cycle has its own characteristics: a triggering technology, promises of transformational applications, massive investment, then a confrontation with real-life limitations.
What’s different this time: the data is incomparably more, the computation is incomparably cheaper, and the measurable results are real on some tasks. The bottom of the current hype cycle is probably not as deep as previous ones. But the top is dizzying.
How to read the AI’s current position
In 2025, different AI segments are at different positions on the curve. Consumer LLMs are probably still on the way down from the peak, with a correction of expectations underway in companies that have launched projects without a clear ROI. Industrial computer vision has been at the productivity plateau for years. Autonomous agents are probably at the peak of expectations.
Identifying where your target technology is on the curve helps you calibrate investment timing. Investing in a technology at the trough of disillusionment, with a validated use case, is often the best time. Investing at the peak of expectations on an unproven use case is the best way to contribute to a project that will be cancelled 18 months later.
Signals from the top of the hype
How can you tell when you’re near the summit? Some reliable signals:
Near-unanimous media consensus (contradictory articles are in the minority). Comparisons with past civilizational mutations (“like electricity”, “like the Internet”). Fund-raising at valuations disconnected from any income. Social pressure to adopt or risk being “behind the curve”. Promises on use cases that nobody has yet validated in production.
These signals don’t mean that the technology sucks. They mean that expectations have exceeded proven achievements. Technology can be real and useful, and current valuations can still be irrational.
“90% marketing, 10% reality.”
How not to buy at the top
The practical rule for managers: distinguish between investment in skills and investment in deployment.
Investing in competence (training your teams, experimenting with non-critical internal cases, understanding the mechanisms) is reasonable at any stage of the cycle. The cost is limited, the benefit is lasting.
Investing in a large-scale deployment on an unproven use case in your sector, with a supplier positioned on hype rather than measurable results, is the risky bet. It may pay off. It may also leave you with a cancelled project and a clean slate.
The question to ask any supplier: “Give me three customers in my industry, with their use cases and metrics.” If the answer is vague or redirects to generic benchmarks, you’ve got your answer about real maturity.