Generative AI suffers from low economic and engineering efficiency, with 95% of corporate adoption attempts failing to produce meaningful results.
Imagine a product where the more you mass-produce it, the more expensive it becomes, and every time you run the factory, it consumes enough electricity to power a city on the other side of the world for days. Most would say, “This is poorly engineered!” Yet, critics argue that generative AI—which the global IT industry is currently pouring trillions of dollars into—is following exactly that path.
Recently, the industry has seen extreme assessments of generative AI, with some calling it a “trillion-dollar engineering disaster” (Source: The Atlantic). What on earth is happening?
Why does this matter?
We have long painted a rosy future where generative AI automates all tasks and brings massive productivity gains. However, the reality is colder than expected. While companies are investing heavily to adopt AI, they often fail to see noticeable profits or productivity improvements in real-world business environments.
This is not simply a matter of “the technology being immature.” It is because indicators suggest that the costs required to maintain and advance AI are so inefficient that they outweigh the technical value. There are now serious questions about whether AI can be integrated into our daily lives—like the smartphone apps or web services we use every day—in a highly affordable manner (Source: The Atlantic, Source: Gary Marcus).
Simplified: Why is it inefficient?
To put it simply, the training method for generative AI is like creating a highly sophisticated “giant filter.” Every time you select a pretty filter in a photo app, data in the photo is modified. AI models perform this filtering process with extreme precision using billions of parameters (internal settings that adjust as AI processes data) (Source: MIT News).
The problem is the cost of training this filter. For manufactured goods like light bulbs, cars, or clothing, production becomes cheaper the more you make, thanks to “economies of scale.” Once a factory is properly set up, subsequent units are much cheaper to produce. Generative AI is the opposite. As the model scales up, the electricity and computational power required increase exponentially (Source: The Atlantic).
One AI researcher noted, “I don’t know anyone who can name any other real software product that scales this inefficiently” (Source: AI Weekly).
Current State: 95% Failure
The situation is more serious than expected. According to a 2025 study by MIT, a staggering 95% of generative AI projects conducted by companies failed to generate meaningful profit or results relative to investment (Source: The Economic Times, Source: AI Commission).
Despite this, the investment frenzy has not stopped. Over $44 billion (approx. 60 trillion KRW) was poured into AI startups in the first half of 2025 alone (Source: The Economic Times).
Some experts defend it by saying there is no alternative, claiming that Large Language Models (LLMs—AI models that learn from vast amounts of data to understand and generate language) are currently the “best working” way we have discovered (Source: Digg). However, the vast majority of companies are struggling with performance verification, ethical risks, and massive energy consumption (Source: The Economic Times).
What happens next?
What is needed for generative AI to shed the burden of being the “worst technology” by economic and engineering metrics? The industry is already realizing that the era of simply increasing model size is over.
Efficiency will become the most significant topic for AI moving forward. Instead of maintaining giant, dinosaur-like models, there will be fierce competition to find technologies that specialize in specific tasks, use less energy, and still achieve high accuracy. Companies must now focus on the engineering considerations of the remaining 5% of cases that actually create profit, rather than unconditional adoption (Source: Forbes).
MindTickleBytes AI Reporter’s View
A technology being evaluated as a “disaster” is not a failure; it is the process of the bubble bursting and finding its essence. Now is the time to break away from vague fantasies and coldly consider the real value of AI.
References
- Generative AI Is an Engineering Disaster - The Atlantic: https://www.theatlantic.com/technology/2026/07/generative-ai-engineering-disaster/687901/
- Generative AI Is an Engineering Disaster - Gary Marcus: https://x.com/GaryMarcus/status/2077275136701481375
-
Reisner: Generative AI Is a Trillion-Dollar Engineering Disaster AI Weekly: https://aiweekly.co/alerts/reisner-generative-ai-is-a-trillion-dollar-engineering-disaster -
Explained: Generative AI’s environmental impact MIT News Massachusetts Institute of Technology: https://news.mit.edu/2025/explained-generative-ai-environmental-impact-0117 - Generative AI Is an Engineering Disaster (shared) - Phil Stock World: https://www.philstockworld.com/2026/07/15/generative-ai-is-an-engineering-disaster/
- Generative AI Is an Engineering Disaster. A shockingly inefficient trillion-dollar project. - Communick News: https://communick.news/post/6570648
- Shehzad Younis on X: “Generative AI Is an Engineering Disaster A shockingly inefficient trillion-dollar project”: https://x.com/shehzadyounis/status/2077317219420434790
- Generative AI Is an Engineering Disaster. A shockingly inefficient trillion-dollar project. - Baraza: https://baraza.africa/post/4219008
- Atlantic Article Calls Generative AI an Engineering Disaster - Digg: https://digg.com/tech/aomq04kd
- MIT study shatters AI hype: 95% of generative AI projects are failing: https://economictimes.indiatimes.com/magazines/panache/mit-study-shatters-ai-hype-95-of-generative-ai-projects-are-failing-sparking-tech-bubble-jitters/articleshow/123428252.cms
- MIT report: 95% of generative AI pilots at companies are failing: https://aicommission.org/2025/08/mit-report-95-of-generative-ai-pilots-at-companies-are-failing/
- MIT Says 95% Of Enterprise AI Fail- Here’s What The 5% Are Doing Right: https://www.forbes.com/sites/jaimecatmull/2025/08/22/mit-says-95-of-enterprise-ai-failsheres-what-the-5-are-doing-right/
- MIT Finds 95% Of GenAI Pilots Fail Because Companies Avoid Friction: https://www.forbes.com/sites/jasonsnyder/2025/08/26/mit-finds-95-of-genai-pilots-fail-because-companies-avoid-friction/
- Power consumption and poor scalability
- Lack of internet speed
- Shortage of developers
- 5%
- 50%
- 95%
- Technical complexity
- Achievement of economies of scale
- Number of users