Investor interest in Artificial Intelligence (AI) is clearly driving stock market returns for the first half of 2026. AI is being built out in North America on a massive scale but at this point long-term winners and losers are hard to predict. In this case study we would like to lay out many of the conflicting issues and challenges for investors. When we use our Avenue discipline of looking for high rate of return companies at reasonable valuations it leads us to data center construction and the inevitable need for power generation.
When we at Avenue ask the question, who is making money now, it is the businesses that supply the infrastructure and the semiconductor chips. When we ask what companies we can invest in at a reasonable valuation, we have found returns in businesses building out data centers like Emcor and Installed Building Products in the US and Toromont in Canada. Once these data centers are up and running, we believe an important part of the power supply will be natural gas and here we have investments in Tourmaline and Topaz.
This is an image of Meta’s data center superimposed over Manhattan to give you a sense of scale. Photo Source: Bloomberg
All stock market investments come with risk, but we can choose which type of risk we want to take. Right now, the large technology companies and semiconductor manufacturers have been in a stock market frenzy and risk appears elevated. However, calling something a bubble is usually only possible in retrospect. The term being used halfway through 2026 is that we are in a unique earnings bubble.
US Big Tech vs China
The elephant in the room is acknowledging that the US big tech companies’ strategy for AI is the polar opposite of their Chinese competitors. We are seeing Chinese AI increasingly being used more than US AI, because ironically the Chinese strategy is to democratize the use of open-source code at much lower cost. In contrast the US tech strategy is to keep AI private, restrict competition and eventually charge high prices. Before we get to our Chinese AI usage chart, we will lay out a series of future profit challenges which we can call an accounting paradox for US tech companies.
AI in North America is being built out on a massive scale, and some businesses are making a lot of money: $1 trillion of direct spending and almost $2 trillion when we include indirect spending estimates for 2026. That number is about 6% of US annual gross domestic product this year alone, which dwarfs the capital spending boom of the original internet rollout in the late 1990s.
Nvidia earnings bubble
The best chips for data centers are made by Nvidia and the company is making a lot of money. Even with the stock market valuation of $5 trillion dollars, earnings are expected to be over $300 billion dollars next year so the stock would be trading a normal 17 times earnings. Where more nuance is needed is that Nvidia currently has a 70% profit margin, which is high in any historical period, especially for a technology hardware business. The reason given for why profit margins are so high is due to the interwoven nature of chip financing. A Nvidia customer like CoreWeave needs all the chips they can buy yesterday but doesn’t have the money to pay for them. Nvidia lends CoreWeave the money and as you would imagine, the buyer isn’t driving a hard bargain on price given they need a loan from the vendor. The question is, how long can Nvidia rely on this business model as competition increases over the next few years. A realistic outcome is that Nvidia’s sales of chips could still double. However, with a more normal profit margin of say 20%, earnings would be cut in half over the next few years.
Capital spending exaggerates earnings
In a capital spending boom, the reported profits are elevated due to typical accounting practices. For example, Nvidia makes and sells semiconductor chips. Sales revenue minus the cost of producing the chips equals the company’s profits. However, for the data center buyer of the chips, the cost of these chips is written off over the 5-year useful life of the chips. Nvidia books profits annually but the buyer has elevated profits as the costs can be smoothed over 5 years. In the growth phase, profits appear higher than when the business matures a few years from now. The stock market is excited about earnings now, but these earnings will decelerate in the future.
Free cash flow switches to no cash flow and debt accumulation
The big US tech companies like Google, Amazon, Microsoft and Meta were effectively debt free and making an enormous amount of cash that was then recycled into the financial markets. Now within a very short period, all these companies have gone on a capital spending spree even to the extent of taking on debt. The mantra of the day is that AI is so important these companies fear they cannot be left out. Earnings for these big tech companies might boom after the capital spending spree or the business of AI might become commoditized by China and US tech profits will not materialize.
Earnings increase from unrealized capital gains
Google, Amazon and Nvidia have reported almost $70 billion dollars in unrealized capital gains in 2026 from their respective ownership of private companies of OpenAI and Anthropic. Both private OpenAI and Anthropic are expected to go public in the next year with $1 trillion dollar market capitalizations respectively. Both these private AI companies have very expensive valuations as both have revenue below $40 billion dollars and both don’t even make a profit. If these trillion-dollar valuations are not realized, our big US tech companies will be reporting an unrealized capital loss.
Advertising spending, really?
The leading AI companies, OpenAI and Anthropic, are giving away their products for free or at a greatly reduced price to incentivize adoption. This is an understandable business strategy but again it is the scale that is so impressive. OpenAI has spent $30 billion dollars on ‘advertising’ which is really just giving away its service for free instead of admitting that they don’t want to charge for it so the customer doesn’t know what the AI service really costs. Not unlike the drug dealer who gives you free drugs to get you addicted then charges full price when you come back for more. The problem is that if US tech companies raise their prices to charge what their service really costs, plus a profit margin, customers have a choice of a much cheaper Chinese AI model. The Chinese model might not be as good, but it might be good enough for the needs of most users.
The stock market math starts to look stretched
We were intrigued to read an article a couple weeks ago that argued that for US AI related stocks to justify their current market capitalization, they will need to make all the money in the economy and not leave anything for anyone else. So, let’s play with some numbers and see if there is merit to this idea.
Currently the US stock market capitalization is $70 trillion dollars. If we add up the 12 biggest US tech companies starting with Nvidia at $5 trillion dollars, newly listed SpaceX at $2 trillion dollars and give an estimated value for OpenAI and Anthropic of $1 trillion dollars each, this adds up to $30 trillion dollars of market capitalization. AI related tech companies represent over 40% of the US stock market capitalization. Let’s now assume that these companies are mature businesses and trade at the historical stock market multiple of 16 times earnings.
Our 12 companies, with a market capitalization of $30 trillion divided by 16 times earnings multiple, would equal $1.9 trillion of hypothetical earnings. This compares to the total S&P 500 current earnings of just under $2.8 trillion dollars. So, our hypothetical AI profits are not all the earnings, but these 12 companies would make over two thirds of the earnings in the US stock market which is certainly a big chunk of the economy. Anything is possible and a lot of new earnings might be created with AI, but we understand the point that AI alone can’t make all the money because the customers still need to make money to pay for it.
Chinese AI is open-sourced code and cheaper
It appears that the Chinese strategy for AI adoption is to commoditize AI and have the user decide what to do with it. An analogy being used is that Anthropic is making a Ferrari and giving it away to get the customer hooked. Chinese AI is more akin to a Toyota which is functional and durable for the average user. The US might try and block Chinese AI, but the rest of the planet is incentivized to adopt Chinese tech over the US products. That is a big risk for future US tech stock profits.
Let us restate our previous conjecture that Chinese AI models might not be as good, but they might be good enough for the needs of most users.
Bill Harris, July 2026
