I Need Your Clothes, Your Boots, and Your Motorcycle...

In the iconic line from The Terminator, the cyborg bluntly demands everything he needs for survival from a random biker. The modern approach to artificial intelligence often follows the same logic. Companies aggressively seize vast resources — money, energy, data, and skilled personnel — to launch the next wave of cloud-based models and "revolution." However, in practice, this path frequently proves unprofitable. The hype bubble is deflating, and the market is gradually shifting toward more balanced and economically sound solutions.

The Scale of the Problem and the Gap Between Expectations and Reality

According to McKinsey's 2025 research, about 88% of companies have already implemented AI in at least one business function. Yet only around 6% derive significant profit from it, exceeding 5% of EBIT. Most enterprises implemented the tools "for show," reported to shareholders or the board, and continued working as before. This is a classic example of hype-driven investments, where the fear of falling behind competitors overrides sober calculation.

The shift by providers from fixed subscriptions to pay-per-token usage became a moment of truth. Powerful models like the latest versions of Claude or GPT consume resources at an enormous rate. The financial group Citi completely restricted employee access to top models, citing them as the main cause of sharp corporate expense growth. Atlassian, the developer of the popular Jira system, saw its AI cloud service costs rise from $5 million in August 2025 to over $15 million in May 2026. Forecasts suggest the annual figure could exceed $120 million. The company was forced to introduce strict controls and abandon unlimited access. Similar measures are being taken by Adobe, Amazon, Accenture, and other major players. Employees complain in corporate chats: tokens run out in two or three days, with little noticeable return on investment.

Firing Employees Because of AI Is Becoming Unprofitable

The paradox is that fully replacing employees with AI is often economically unjustified. AI, especially in the form of complex agents, requires such volumes of computing resources that its costs begin to exceed those of human labor. Nvidia's Vice President of Applied Deep Learning, Bryan Catanzaro, openly admitted that in his division, computing costs significantly exceed employee salaries. Companies that actively encouraged AI use (introducing departmental rankings by token usage or leaderboards like "Claudeconomics" at Meta) are now forced to impose limits and even cancel licenses.

Cloud Computing: Power vs. Price

Cloud computing undoubtedly provides access to the most advanced and powerful models. However, its cost under intensive use often turns out to be unjustifiably high. Dependence on external providers creates additional risks: rising rates, limits, and data confidentiality issues. This is why more and more large companies are considering a shift to their own solutions.

A positive signal came from OpenAI: the company's engineers developed software optimizations that more than halve the cost of processing requests to existing models — purely through software methods, without new hardware. Such breakthroughs show that the era of mindless infrastructure expansion is gradually giving way to smart optimization and improved efficiency.

Real Successes: Targeted Application Delivers Results

Despite the general problems, AI demonstrates impressive results when applied purposefully to specific tasks. Samsung implemented digital twins and reduced product testing time by seven times — from 15 days to just two. Virtual simulations replace hundreds of physical tests and significantly cut costs.

Toyota plans to use AI to bring order to its documentation and terminology. This is a classic example of useful application: neural networks will help standardize terms, speed up information retrieval, and ensure consistency across vast volumes of technical documents. Such tasks are ideally suited for AI and deliver quick, measurable results.

The Future Belongs to In-House Solutions

In the coming years, large companies will increasingly develop their own local AI, finely tuned to their data and processes. Such systems avoid expensive cloud token costs, increase security, and achieve maximum relevance. Cloud services will remain for peak loads or access to the newest models, but day-to-day work will shift to controlled internal platforms.

The Main Lesson

The unprofitability lies not in the technology itself, but in the mass "implement for the sake of implementing" approach driven by hype. Explosive costs, the need for constant oversight, and inflated expectations of complete human replacement lead to restrictions and disappointment.

"I need your clothes, your boots, and your motorcycle" is a vivid metaphor, but a poor long-term strategy. AI has already proven its value in specific tasks, such as Samsung's testing optimization or Toyota's documentation work. However, for widespread success, a shift from hype-driven resource grabs to conscious, economically justified use is needed. Optimization, local models, targeted implementations, and constant ROI calculation are what will allow the technology to deliver stable benefits rather than become another item of uncontrolled expenses.

Companies that manage to adapt and separate real value from hype will gain a serious competitive advantage. The rest risk spending huge sums and being left with nothing when the bubble finally bursts.

    A. FIGURNOV

Link to the article: //msn.kg/en/news/44661/