On October 14th, an independent research institute, MacroStrategy Partners warned in its latest report that the current situation was not the sameAIInvestment booms have surpassed many famous bubbles in history. According to the analyst Julien Garran, AI foam is 17 times larger than the "Internet bubble" or even four times more than the "subprime crisis"。
THE REPORT STATES THAT THE FEDERAL RESERVE ' S QUANTITATIVE EASING POLICY AND LOW INTEREST RATE ENVIRONMENT OVER THE PAST DECADE OR SO HAVE LED TO LARGE-SCALE MISALLOCATION OF CAPITAL TO AREAS SUCH AS AI, REAL ESTATE, NFT AND VENTURE CAPITAL. AS THE INTEREST RATE ENVIRONMENT RETURNS TO NORMAL, MARKET RISK GRADUALLY EMERGES。
Garran also states that large language models have reached the "scale limits". For example, the training costs of GPT-3 were about $50 million, GPT-4 went up to $500 million, while GPT-5 cost $5 billion, but performance improvements were not significant。
IN HIS VIEW, THIS TREND OF "INCREASED INPUT INDEX LEVELS AND DECLINING RETURNS" MEANT THAT THE AI INDUSTRY LACKED A LONG-TERM MOAT。
IN ADDITION, THE REPORT INDICATES THAT AI HAS EXPERIENCED A DECLINE IN THE RATE OF ADOPTION IN THE ENTERPRISE AND THAT SOME TEST CASES INDICATE THAT THE MODEL HAS A LOW COMPLETION RATE OF 341 TP3T FOR COMPLEX TASKS。
In Garran, it appears that as data centre investments and wealth effects peak and reverse, the United States economy is likely to repeat the recession that followed the collapse of the Internet bubble in 2001。
He warned that this could lead to a "zone 4 recession/deflationary bust" and that even if the Fed or the government tried to adopt stimulus policies, it would be difficult to reverse the trend easily. In such cases, the overall market valuation could fall significantly, with technology/AI being the first to suffer。