Useful Life
Useful life is the period over which a company expects to use a purchased asset and spreads its cost in its accounting. For artificial intelligence data centers, this figure also helps determine how high a corporation's reported profit turns out to be.
When a company buys something expensive that it will use for many years, it doesn’t deduct the full purchase price from profit all at once. Instead, it spreads the cost over the years the item is expected to be in use. This exact period is called the useful life. A server costing 60,000 euros with an assumed useful life of six years thus reduces profit by 10,000 euros per year. This annual allocation is called depreciation. The useful life is an estimate here, not a measured fact.
Why a few years' difference moves billions
The useful life acts like a lever on reported profit. If a company extends it from four to six years, the same purchase price is spread over more years. Less is deducted per year, and profit immediately looks higher. Nothing changes about the money actually spent.
For the major cloud providers, enormous sums are at stake. Corporations like Microsoft, Alphabet, and Amazon buy data center equipment worth double-digit billions of dollars every year. All three have extended the assumed useful life of their servers in recent years, usually from three or four to five or six years. The effect on annual profit was in the billions each time.
That’s why investors watch such changes closely. A longer useful life can be objectively justified, because older chips can still perform meaningful work. But it can also flatter profits that wouldn’t exist without this accounting rule. Critics warn: if AI chips become obsolete faster than assumed, depreciation will catch up with the company abruptly later on.
How the figure comes about
Companies are not free to simply invent the useful life. They must justify how long an item is expected to remain economically usable. For many capital assets, official tables serve as guidance—in Germany, for instance, the AfA tables issued by the Federal Ministry of Finance. For financial statements under international rules, however, the company’s own plausibly justified estimate is what counts.
It’s important to distinguish between technical and economic useful life. Technically, a server can run for ten years without breaking down. Economically, it may only be worthwhile for five years, because newer hardware delivers significantly more performance per watt of electricity. The shorter of the two periods is always the decisive one.
A common misconception: that useful life indicates when a device will be disposed of. That’s not true. Once the useful life has elapsed, the device is carried on the books at nearly zero euros, but it can keep running. Large providers often shift older servers from AI computation to simpler tasks like data storage.
Where the figure shows up in quarterly reports
The issue is most visible in tech companies' quarterly figures. There’s a line item called depreciation. If a company changes the useful life, it must explicitly disclose this and quantify the effect in figures. Business media regularly pick up on such adjustments because they boost profit without any new business.
Useful life also plays a central role in the debate over a possible AI bubble. Experts disagree about how long graphics processors for AI actually remain profitable to operate. Some investors consider two to three years realistic, while the corporations plan for five to six. Who is right determines whether current profits are sustainable.
However, the term also comes up far outside the world of AI. A craft business depreciates its van over six years, an office depreciates its laptops over three. The principle is always the same; only the amounts differ by several orders of magnitude.