At one point in the past, the phrase “When the GM is sneezed, the United States was cold” was banned around to indicate how much the economy was dominated by the car manufacturers. This phrase was edited to an old version, “When America is sneezed, the world gets cold.” For now, keeping the “cold catching” aside, the phrase was acknowledged that acknowledged the inferior economic influence that the United States has done on the rest of the world.
At this point of the current AI cycle (there are other people in the past), Generative AI (Genai) has certainly engulfed a significant part of the technology cycle. Ai -ai companies remarkable priceLike, Nvidia, from hitting $ 4 trillion) gives a break and invites someone to ask what the original Geneai is upside down. The news cycle is full of numerous AI successes and failures, and nevertheless, when we approach the third anniversary of the release of the Chattagpat (November 30, 2022), most of the “AI promise” still seems unacceptable. There have been successes.
Opposite now The notorious MIT report, which cited 95 % of the Geni failureAnother study of MIT Sloan Management Review. Practical AI Implementation: Success Stories Hends light on areas where AI is gaining success. From the study:
- Task Ordinary To employees in many roles: Large models of language are popular Task Such as information synthesis and summary and documentation documents.
- Special use for specific roles and tasks: High risk tolerance businesses are ready to use Generative AI for business processes. Famous issues of use include coding, supporting customer service, guide the creative process, and scale content. For example, Carmex uses Generative AI to summarize customer studies, which are then posted on reservation research pages for users use.
- Products and users facing applications: E -commerce companies are introducing chat boats and increasing personal purchase experiences. Companies like Adobe and Canva, make both graphic design software, embedded Generative AI tools in their products.
In addition, recently TCP25 Conference Highlight the use of AI for science. Given some of the challenges for the General-II (such as deception, prejudice and discrimination, intellectual property), there is clearly progressing with the General-A solution.
And how much does it cost?
Generative A has been surprised. Has been market in the last three years Too much Buying all GPUs nvidia can make (and someone else) in this case. The models of models, however, It seems like Reaching a decreasing point of return.
(Pratmish t/shutter stock)
Chat GPT5’s recent hypid release has been under influence in many cases (it is clear that some people have seen improvement. Something Problem Domains). Industry analysts and critics/realistic Gary Marcus For years, AI has been tracking research. Its updated Chat GPT5’s analysis paints disappointing photo. According to Marx, who supports AI,
“GPT -5 can be a moderate quantitative improvement (and it may be cheap), but it still fails in the same way as its predecessors, chess, in reasoning, in vision, even sometimes counting and basic mathematics.
In addition, now a computational wall appears in front of the LLM scaling. In pre -print paper The wall is the major language modelsAuthors PV COCY and SSCOS State:
“We show that scaling laws that determine the performance of large language models (LLM) strictly limit their ability to improve the uncertainty of their predictions. As a result, increasing their reliability to meet the standard of scientific inquiry is to increase their reliability.”
Some suggestions have been made by Marx and others that poor performance of GPT5 may be a move to save cost. In fact, the financial image of many living companies works well over the next several years. And the whole LLM-AGP discussion (artificial general intelligence) It seems like Off the table
I in a recent interview Slate, Ai Critical Ed Zetran Talking about the financing of Open and Anthropic:
“Openi, which mainly funds Microsoft, operates all its infrastructure. Microsoft Chat owners all GPUs needed to run GPT. When Open is making more infrastructure, they are relying on Venture Capital. Between expecting a $ 44 billion burning, which is basically financing through Amazon and Google, it is going to be more like $ 5 to $ 10 billion.
Based on these and other public estimates, many AI companies need to know a way to break the chariot sooner Shortly
Recently, as reported in the Wall Street Journal, Openi announced a contract with Oracle It is unclear where the Open AI will receive these funds from the Billion 300 billion in the cloud infrastructure, starting in 2027, or how Oracle will secure the GPU and power necessary for such a deal.
Can you guess?
At this point, AI companies need to ask, “Can we evaluate the money selling?” A additional Analysis by Ed ZetranBased on, based on Data from informationIndicates that in 2024, the opening revenue was likely in a region of $ 4 billion, and operational loss after calculating the income was $ 5 billion. Thus, the rate of 2024 burning rate was about $ 9 billion. In addition, the information also reported that the Open AI had 15.5 million paying users in 2024, though it is unclear what level they use by Open AI’s premium products. This is equivalent to spending $ 580 per customer per customer.
Open will need to either take more compensation, reduce costs, or sell chat GPT in conference and other major markets will need to be profitable. Basically, the “killer app” for AI has not been found.
The computational cost of diagnosis varies. It is speculated that the GPT5 was designed to make more powerful than the previous version, thus helping to reduce costs. Even a simple inquiry illuminates enough GPU in the data center. The new and more powerful GPU will help reduce costs, but the amount of income will not be sufficient to stop the cost gap. There is also a research of specific models related to small domain that can perform with larger basic models, but use very little strength. If the AI ​​is to be profitable, it will be very important to overcome the costs of individuality. At this point, there Does Not It seems like A clear way of profit. One will assume AI investors are strictly aware of this situation.
Of bubbles, black laughter and danger
Naseem Nicholas Talib has popularized the Black Swan Theory (or dealing with an unknown risk). Explaining the scope of general expectations in history, science, finance and technology, high -level, harsh forecasts and rare events. According to the definition, “Black Swan events” (BSE) are unexpected and may be especially effective in some cases.
In subsequent treatment, the student also offers ideas that can help reduce BSE by implementing strong structures that can maintain negative effects and even Be stronger. These ideas are not new. Investment is recommended for diversity, and the ancient idiom “Do not put all your eggs in a basket” provides a strong strategy. The difficulty is trying to predict things that cannot be predicted (BSE). Due to external conditions and events, the bubbles are not strong and will eventually end or fall.
Consider these methods, consider the data below Take a look at the market by Michael SimblestChairman of the Market and Investment Strategy for JP Morgan asset and wealth management (scroll down on page 6).
US Real GDP Growth Partnership with Tech Capex Expenses, Source: Michael Sambals, JP Morgan, September 2, 2025 (p. 6) Market on the Market (p. 6)
According to the data and the source report, the capital’s capital costs have participated in the previous three constituencies about 35 % -45 % of the overall growth of US GDP. Corporate AI’s investment reached 252.3 billion dollars in 2024. A large portion of the tech sector caps (or almost all) has been fueling through AI costs. Everything from GPUS Data centers Premium is being sold at rates.
There are concerns that the AI ​​market is a bubble. Like in the past, like dot com and real estate bubbles Deep Economic impact in industries. As reported in StuffyOpeni CEO Sam Altman indicates that he believes there is AI’s bubble:
“When bubbles are encountered, smart people grow more about the kernel of truth,“Altman told a group of reporters last week.
“Are we at a stage where the overall investor is about the highest limit about AI? My opinion is yes. Is AI going to be the most important thing in a very long time? My opinion is also yes.
Based on the aforementioned data, even a partial decline in AI Capex costs will have a significant impact on the US economy. The summary is that the current American economic AI basket has seven eggs: alphabet, Amazon, Apple, Meta, Microsoft, NVIDIA, and Tesla. Given the state of adolescents and expensive genius markets, the current situation offers a high level of threat when seen in the context of the US economy. If the AI ​​market sneezes the basket, we can all get cold.
This article first appeared on our sister’s publication, HPCWIRE.
Relevant
