Stop Looking at Earnings: The Real Threat to AI Stocks Is the Collapse of Compute Prices

Barchart Barchart Open on Barchart
Stop Looking at Earnings: The Real Threat to AI Stocks Is the Collapse of Compute Prices

Today, Wall Street is obsessed with a question: “Where is the money from AI?” Analysts are meticulously calculating return on investment (ROI), studying the revenue from subscriptions, and doubting whether software companies will be able to monetize the gigantic capital expenses that Big Tech is pouring into infrastructure.

In my opinion, though, there is another question to ask. The real threat to the tech sector — and the whole U.S. stock market — is not hiding in a deficit of ideas for AI monetization but in a classic economic shock: a crisis of computing power overproduction. Investors are used to thinking that the GPU shortage will last forever. But the flywheel of the investment cycle is already launched, and the reverse countdown has begun.

More Top Stocks Daily: Go behind Wall Street’s hottest headlines with Barchart’s Active Investor newsletter.

 

The Anatomy of Demand and Supply

In 2023 and 2024, this industry faced a harsh bottleneck. Chips were in short supply, cloud giants stood in queues, and rental rates for flagship Nvidia (NVDA) H100 GPUs peaked near $8 an hour. The logic of businesses back then was simple: Grab any accessible capacity no matter the price. An unprecedented building boom began in response to this shortage. Hyperscalers — Microsoft (MSFT), Amazon (AMZN), Alphabet (GOOGL) — and hundreds of other companies began to build data centers in huge volumes.

This is where the time lag plays out. Data centers, which began to be built in 2024 and 2025, are coming online right now in a big way. They're also hitting the market simultaneously with chips from the latest generations, such as the Blackwell (B200) architecture, which is multiple times more efficient than its predecessors.

Oversupply has already begun to corrode unit economics. As fresh academic research on the dynamics of the cloud market shows, spot rates for once-scarce H100 chips at specialized providers have already collapsed to $1.49 to $1.99 an hour, while AWS has been forced to conduct an aggressive lowering of prices.

Paradox of Demand: Teraflops Against Dollars

We should clarify that there is a difference between the physical and monetary volume of demand. Demand in teraflops — the volume of calculations — will continue to grow exponentially since humanity needs more and more neural networks, video generation, and data processing. The demand in dollars, however — the revenue of providers — could differ and show another dynamic.

Why? Because new chips make the cost of one computing operation (token) cheaper. While in 2024 you may have paid $8 for an hour of work from a slower chip, you can now receive capacity surpassing it by three to four times for $5 to $6. A seller's market, where the provider dictates conditions, can in the future turn into a buyer's market.

The physical volume of calculations grows. But because of harsh price competition, the total revenue pool circulating in the cloud rental market may begin to shrink — and it could hit both the margins and incomes of Big Tech.

Bullwhip Effect: Who Will Suffer Down the Chain?

If the supply of computing power exceeds demand, a chain reaction will hit the market.

The first to be impacted would be the cloud segment and “neo-clouds.” Specialized GPU providers — CoreWeave (CRWV), Lambda Labs, and others — will be the first to take the hit. Many of these firms bought equipment on credit using chips themselves as collateral. As cloud rental prices fall, their business models will lose viability. Accordingly, they will not be able to service debts, and the collateral hardware rapidly depreciates on the secondary market. Following these firms, cloud giants will suffer, especially those most heavily dependent on the sale of infrastructure as a service — primarily Amazon (AWS) and Microsoft (Azure). While Meta (META) or Google use AI predominantly inside their ecosystems for ad targeting and user retention, AWS and Azure sell raw compute to corporate clients. The fall in the margins of cloud divisions will cause a harsh revaluation of the shares of these trillion-dollar companies.

The second impact will be a freeze of capital expenditures. As soon as the management teams of Amazon and Microsoft see that an oversupply has emerged instead of a shortage of capacities, the expediency of large capex budgets will come under question. This could lead to a sharp reduction of investment programs for the ensuing periods. Why buy new servers if the old ones are underutilized?

Finally, the next to be impacted in this chain would the chipmakers. The market capitalizations of Nvidia, Taiwan Semiconductor (TSM), AMD (AMD) and Broadcom (AVGO) price in endless order growth. For shares of Nvidia, a simple decline in revenue isn't just dangerous, but also a slowdown in the pace of its growth. If forward orders from hyperscalers for 2027 and 2028 decrease, chipmakers' mulitples will collapse, pulling the whole S&P 500 Index ($SPX) down with them.

The Macro Risk for Investors

This scenario is not just a risk for the next three months but an absolutely tangible possibility for the foreseeable future. Investment cycles are inexorable; a phase of cooling and overproduction always follows a building boom.

For U.S. financial markets, this infrastructural imbalance appears to be the top threat right now. It's much more dangerous than the dynamics of GDP or the trajectory of interest rates from the Federal Reserve. A structural crisis of overproduction in the cloud computing sector, which has provided a significant portion of the entire stock market's growth in recent years, is capable of causing a systemic shock.

With that said, I think that investors right now need to look less at company reports and more at spot prices for cloud compute. If prices continue to fall, it could have far-reaching consequences. Take this risk into account and price it into your portfolios while the market still lives by the illusion of an eternal shortage.


On the date of publication, Mikhail Fedorov did not have (either directly or indirectly) positions in any of the securities mentioned in this article. All information and data in this article is solely for informational purposes. For more information please view the Barchart Disclosure Policy here.

 

More news from Barchart

Stop Looking at Earnings: The Real Threat to AI Stocks Is the Collapse of Compute Prices The Trump-Intel Deal Stole Headlines, But the President Also Bought Nvidia Stock in Q1 Qualcomm Is Teaming Up With ByteDance. What It Means for QCOM Stock Datadog Benefits When AI Goes Into Production. Here’s Why You Should Buy DDOG Stock Now.