Terry Tanaka says investors should consider the economics of AI Is it different this time, or are we in the mother of all bubbles?
More people are raising concerns about AI's dominance of the stock market than before. In an article published on October 27, Wired argued that "AI may not simply be a bubble or even an enormous bubble." Perhaps it's the ultimate bubble. Brent Goldfarb, co-author of Bubbles and Crashes: The Boom and Bust of Technological Innovation, commented on the piece. According to Goldfarb, the AI boom satisfies all of his criteria for a technology-driven bubble, including a reliance on narrative, a focus on "pure play" companies, uncertainty about the ultimate end use, and novice investor participation.
The AI bubble is also on shakier ground than many other technology-driven bubbles, including the .com bubble, the "Roaring Twenties," and the US railroad boom, all of which were followed by significant economic depressions, as financial journalist and former hedge-fund strategist Edward Chancellor has noted in these pages. It's also more hypothetical. While the internet, automobiles, and railroads were proven technologies during their bubble periods, self-teaching computers are not. The AI bubble is more accurately described as "a multi-trillion-dollar experiment" to determine whether artificial general intelligence (AGI) technology can effectively match human intelligence. If that experiment fails, we will have millions of outdated computer chips and dormant, debt-funded data centers instead of canals, railroads, or fiber-optic cables.
The study of artificial intelligence (AI) began at least with Alan Turing and encompasses well-known disciplines like computer vision and machine learning; generative AI is a more recent subfield that has emerged in the last 15 years. When ChatGPT was introduced in November 2022, it quickly gained popularity and expanded the field. However, it's important to note that if you are content to purchase Nvidia stock at the current price, you are essentially placing a wager on the long-term financial success of generative (and its more recent subset, "agentic") AI rather than the industry as a whole. And the possibility of that makes even the bulls anxious. "AI,". is generating trillions of dollars in spending over the coming years, which will sustain this tech bull market for at least another two years, according to Dan Ives, Wedbush Securities' head of global technology research. Ives is one of the great tech bulls, so that is important. Doubt is beginning to creep in if even he is subtly acknowledging that the current bull market may end and offering a potential timeframe.
The characteristics of the AI bubble.
Although "this time its different" is considered one of the riskiest investment phrases, proponents of artificial intelligence are using it more and more. They claim that, in contrast to the .com era's abundance of unprofitable internet start-ups, the businesses driving AI today are extremely profitable. This also applies to Nvidia and the "hyperscalers" (Alphabet, Amazon, and Microsoft), but none of them are profitable due to the money they make from generative AI products. Before ChatGPT arrived, they were already very profitable (and, generally speaking, less capital-intensive). Nobody disputes that selling computer chips and cloud computing can be profitable. However, AI is a different matter.
When generative AI companies are viewed in isolation, the current structure resembles the .com bubble. Venture capital is pouring into speculative companies that spend a lot of money and have no realistic plans to turn a profit anytime soon. The Wall Street Journal's James Mackintosh notes that despite media mentions of the .com bubble growing annually between 1995 and 2000, the bubble continued to expand. Even if everyone is aware that they are bubbles, they can continue to grow.
Typically, a bubble bursts when it comes into contact with a pin. An energy-driven inflation crisis is one possibility, but nobody can predict what that will be for AI. The energy required by AI is enormous. Energy prices cannot be controlled by policymakers, but they can make life easier for AI developers. AI models are likely to use more energy as they grow more sophisticated and gain more users. Additionally, there are indications that AI is already driving up energy costs for US consumers. In the third quarter of this year, average gas and electricity payments rose 3.6 percent year over year, according to Bank of America deposit data. It doesn't take much imagination to see energy-driven inflation quickly becoming an issue for AI developers themselves, regardless of whether it reaches a point where it causes problems for US politicians.
By 2033, OpenAI CEO Sam Altman hopes to have 250 gigawatts (GW) of data center capacity. That is more than four times the peak electricity demand for the state of California and roughly one-third of the peak demand on the US grid, according to Liam Denning of Bloomberg. According to Jensen Huang, CEO of Nvidia, building 1GW of data center capacity costs between £50 billion and £60 billion (of which £35 billion or so goes on Nvidia chips). Building this could cost OpenAI more than £12 trillion. This is simply not going to happen, at least not in the next eight years, but the figures demonstrate how much energy AI's largest players are planning to consume.
AI developers are struggling financially, even with the current state of energy prices. In the first half of 2025, OpenAI reported an operating loss of £7.8 billion, according to tech news website The Information. Although OpenAI's own estimates indicate that it won't be cash flow positive until 2029, when it projects revenue of £125 billion, annual recurring revenue is expected to exceed £20 billion this year. OpenAI's semiconductor-spending binge may abruptly stop if the economics of expanding its capacity at this rate ever start to look bad.
You only need to look at AMD (Nasdaq: AMD) to get a sense of how exaggerated the stock market's response to this binge is. AMD executives estimated that OpenAI's announcement on October 6th that it would purchase up to 6GW of GPUs from AMD could generate an additional £100 billion in revenue after accounting for ripple effects. AMD's market capitalization rose by about £115 billion in just two days following the announcement, exceeding the anticipated revenue from the deal. That's not to mention that AMD will be compensated for the GPUs with its own stock rather than cash.
Where do AI's advantages lie?
Professional productivity does seem to be increasing thanks to generative AI. According to Joseph Amato, president and chief investment officer at Neuberger, "an unusual combination that points to productivity doing more of the heavy lifting" is what the data we currently have from the US indicates: a trend of reasonably healthy GDP growth along with muted job creation. Over the next ten years, AI is predicted to increase productivity in the US by 1.5 percent and in the G7 by between 0.2 and 1.3 percent. However, Amato warns that these benefits are not evenly distributed and that there are risks associated with them. He claims that "lower-end white-collar roles performing routine analysis or administrative tasks frequently filled by recent college graduates face significant displacement risk." The policy ramifications of that are significant.
The revolution in AI might consume itself. It is impossible to unemploy a whole generation of middle-class people worldwide without anticipating significant economic repercussions, possibly enough to offset any potential GDP gains. There is proof that this is already taking place. Since the housing bubble that caused the 2008 crash, average real spending on retail goods relative to total employment has stagnated, according to Ron Hetrick, principal economist at workforce consulting firm Lightcast. This trend was briefly disrupted by COVID and the ensuing stimuli; retail spend per employee is currently declining once more.
Hetrick refers to AI as "a jobs-destroying, money devouring technology" that poses a threat to hasten this downfall. AI firms predict that "large enterprise clients will also see their buyers stagnate" as retail spending keeps declining. The core business pillars of Amazon, Google, and Meta would suffer greatly if the world went into a recession because advertising and e-commerce revenues are ultimately dependent on a sizable population of middle-class consumers who are content to spend money. None of these businesses have an AI division that can sustain the larger business, let alone be even remotely profitable on its own.
Improved medical research is a clear example of the truly revolutionary social benefits that AI could bring about. With significant implications for disease research and treatment development, Google DeepMinds AlphaFold is a program that can predict a protein's structure based on the sequence of amino acids that make it up.
However, there are two things to keep in mind. Firstly, this isn't all that new; DeepMind introduced AlphaFold in 2018, so its potential should have been factored in prior to ChatGPT. Additionally, these methods are not generative AI; top biotech researchers do not ask ChatGPT to generate new amino acid sequences for them because that is not how big language models operate. More importantly for investors, the profitability of the medical applications is not a given.
How to use AI to protect your bets.
Given these trends, how can investors hedge their bets? Choosing energy stocks wisely is one way to take advantage of the rising demand for power that AI companies will fuel in the years to come. However, this window may already be over. For instance, Vistra (NYSE: VST) has increased by 60% over the last 12 months and is currently trading at 22 times forward earnings, which is a fair price for a tech stock but appears high based on conventional utility valuations. Nevertheless, the suppliers should benefit if the AI bubble is ultimately burst by energy inflation. Data-center energy suppliers can be found through an investment trust that has exposure to businesses that power and construct data centers, like Pantheon Infrastructure (LSE: PINT).
You may also search for cheap AI plays. For as long as it takes for the bubble to burst, some semiconductor stocks, like Taiwan Semiconductor Manufacturing Company (NYSE: TSM), stand to profit from AI infrastructure spending without the inflated valuations of the major US names.
TSMC is well-positioned to profit from any technological advancements that may follow AI if and when the bubble bursts because of its effective monopoly on the production of high-end chips.
In order to hedge portfolios, Chris Beauchamp, chief market analyst at IG, recommends a few conventional defensive plays, such as gold, government bonds, defensive shares in industries like healthcare and consumer staples, and multi-asset funds. "Finally, holding cash-like assets is no longer punitive, with policy rates still elevated," he claims. "Diversification is crucial; no single hedge is effective in every situation, but a combination can protect portfolios in the event that the euphoria surrounding AI wanes. The "
Leave a comment on: AI investing is the ultimate bubble