Joey on SQL Server

Inside the Accounting Math Driving Wall Street’s AI Anxiety

A closer look at AI-driven capital spending, GPU depreciation and shrinking margins helps explain why massive hyperscaler investments -- and Microsoft’s AI disclosures -- rattled tech stocks and raised new questions about the economics of artificial intelligence.

• By Joey D'Antoni
• 02/09/2026

You may have been wondering why this week in tech stocks looked like the profile chart of the Women's Downhill at the Winter Olympics. I wanted to take a deeper look at the economics of artificial intelligence that I touched on in my December column, how the market is judging it and how I'm evaluating it. While this will delve into some technical accounting details, I wanted to explain them in a technical context so you can better understand the situation. Let's start with a few key definitions:

Capital Expense (CapEx) is the term used to describe funds companies use to acquire, upgrade or maintain physical assets like buildings, technology or equipment, with the goal of increasing operational scope or future economic benefits. CapEx is listed as an investment on a company's balance sheet. Also, importantly, CapEx resources, such as GPUs and datacenters, must be depreciated over their useful lives. Traditionally, computer equipment, such as servers and PCs, has been depreciated over a three-year period. However, recently, the hyperscalers have all aligned on depreciating their GPU investments over a six-year period. This is important, because CapEx is not directly deductible from taxes, but the annual depreciation on the capital assets is.

Operating Expense (OpEx) is an expenditure that a business incurs for the cost of its normal operation. For a perfect example of OpEx, look no further than your cloud bill or office rent. For tax purposes, OpEx is directly deductible from income.

Depreciation is an accounting concept to capture the reduction in the fair value of an asset as it is used, and the allocation of the costs of those assets to the periods in which the assets are used. This concept is increasingly important as hyperscalers are predicting and account for how long the useful life of the GPU assets will remain useful and retain their value.

Profit Margin (Margin) is the measure of the degree to which a company, or a particular line of business, makes money compared to expenses. Simply put, profit margin is what's left after you sell a good and pay all of the related expenses associated with selling that good. If you sell a piece of software for a $1,000 and your total expenses for selling that software was $100, your net profit margin would be 90%. While that 90% number is very good (and not uncommon in a purely software business), "traditional" cloud computing isn't far behind -- last quarter, Microsoft reported a gross margin of 69% for the cloud business overall.

Ok, now that we've done our accounting homework, let's talk about how this applies to the tech industry. The concept of cloud computing is very CapEx-heavy. If in 2026, you wanted to start your own hyperscale cloud company, you would need to buy or lease land all around the world, build data centers, fill those datacenters with computers and storage and then network them all together, and then you could sell services. Microsoft and AWS didn't really have their traditional cloud businesses built out this way -- as they both started out small, with high margins and rapid growth, added capacity organically, and didn't have to make massive capital investments or take massive loans to expand those offerings. In other words, cloud computing has been a tremendous business for both Amazon and Microsoft.

Now, let's talk about the AI model. I talked a bit about this in my December column, but AI datacenters are expensive. Because LLMs are inefficient, they require much more power and cooling than traditional cloud datacenters. In many cases, power requirements require both new datacenters and additional power capacity beyond what already exists. As you can imagine, this leads to a lot of CapEx to build out that datacenter capacity and fill it with high-end GPUs. The big difference between this AI scenario and the way the current cloud infrastructure was built is that the sales were there to drive expansion almost immediately. While broad cloud adoption took time, AWS and Azure were successful businesses very quickly. AI, as it exists currently and for the foreseeable future, is at best a very low-margin business and more likely a negative-margin business. For those of you who haven't been following closely, that means it's losing money.

Public companies have to report the total of their capital expenditures on their balance sheets. There are several financial metrics that use those capital expenditures as an input. Investment analysts use those metrics to assess the company's health. One of the reasons for this week's market drop is the massive collective CapEx announced by Amazon ($200B), Microsoft ($150B) and Google ($180B), all of which are investing in AI infrastructure. The other factor here was that Microsoft reported that OpenAI (a firm that many analysts, including myself, question the financial viability of) is responsible for 45% of their future cloud contracts.

There were a couple of quotes from the conference call after Microsoft's recent earnings report that raised my eyebrows, both from Chief Financial Officer Amy Hood. The first: "Company gross margin percentage was 69%, down slightly year-over-year driven by investments in AI, including the impact of scaling our AI infrastructure and the growing usage of our AI product features."

In reporting Azure earnings, Hood also mentioned: "and this quarter, demand again exceeded supply across workloads even as we brought more capacity online."

This is strictly my opinion, but it feels like Microsoft is sacrificing current revenue, by increasing datacenter capacity, but only focused on AI, rather than expanding "regular" Azure capacity that customers are consistently demanding. Anyone who works in the Azure space, particularly in Europe, but also in the U.S., is very familiar with the current capacity constraints that limit choices and often force difficult decisions. Back to our accounting class, that's trading high margin guaranteed business (cloud) for a very speculative business, with questionable margins (AI).

The job of public company leadership is at its core to drive revenue growth. The entire industry has convinced itself that the only path forward to growth is AI. Even if AI is the only source of future IT earnings, you must question whether or not a very expensive, low-margin business, that is quite possibly a commodity business is ever going to achieve the success to justify the massive capital investment.  

So if you're an IT admin or manager, what do you need to know about the current state of the technology world? Well, cloud capacity in popular regions is going to continue to be challenging. I would also be on the lookout for either direct price increases or subtle upcharges in various cloud services -- profit goals still need to be met. There is always the option go back on-premises, but if you haven't you should check the price of memory -- AI investments have driven the demand for relatively fixed supply of memory through the roof. Stay tuned to the markets and us at Redmondmag, where I'll do my best to explain the industry's turbulence.