The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →There is no established, industry-wide “capital half-life” for AI hardware. “How long does it last?” can mean how long it physically operates, how long a company depreciates it for accounting, or how long it remains competitive and earns a return. Those clocks differ—and the buildings and power infrastructure supporting AI workloads can have very different lives from servers, accelerators and networking equipment.
What does “capital half-life” mean?
Here, capital half-life is a metaphor for the changing economic usefulness and competitive value of AI infrastructure—not an accounting measure, a measured physical lifespan or a claim that an asset becomes worthless at a particular age. There is no single sourced figure for how quickly AI hardware loses half its value or capability.
As an Amazon Associate I earn from qualifying purchases.
- Physical life: how long equipment can keep operating, which depends on its design, condition, maintenance and environment.
- Accounting life: the period over which a company estimates an asset will provide benefits and recognizes its cost through depreciation.
- Economic or competitive life: how long an asset can be used productively and earn enough to justify its costs relative to alternatives. This depends on workload, demand, utilization, power and operating costs, and newer technology.
A server can remain operational after its accounting schedule ends; an accelerator can lose its relative appeal while still running; and a data-center building can remain useful across several generations of equipment. One clock cannot stand in for the others.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesWhy does AI infrastructure cost money before it earns revenue?
AI capacity is built in stages. A company may commit cash to land, electricity supply, construction, servers, accelerators and networking before a facility is ready, before customers can use its capacity, or before a product using that capacity is monetized. Amazon CEO Andy Jassy described the sequence this way: “So the way it works is that we have to lay out capital and cash in advance of when we can monetize it. This is for land for the data centers, power, the buildings themselves, the hardware, the chips, networking gear.” The statement describes Amazon’s experience, not a universal schedule for every provider.
#1 Best Overall
Jassy said some infrastructure outlays precede monetization by about six months and others by about two years. The interval varies with what is being built and when the resulting capacity can be used. That makes timing a capital-planning problem as well as a technology question: cash can leave the business well before the asset supports revenue.
- Commit and build: acquire or prepare a site, secure power and construct facilities. These steps may take place before computing equipment is installed.
- Install and prepare equipment: bring servers, accelerators and network gear into the facility and make them ready for intended use.
- Put capacity to work: use it for internal services or make it available to customers, then seek enough utilization and revenue to support the investment.
The stages can overlap, but they should not be collapsed into one date. A spending announcement, a completed building, an asset entering service and monetization are different milestones.
Rank #2
How large are the disclosed spending plans?
The figures below describe different companies, periods and measures. They illustrate the scale and timing of investment; they are not directly comparable totals and should not be added together as though they shared one definition.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →| Company and source | Period and measure | Reported amount | Status and scope |
|---|---|---|---|
| Meta Platforms, Inc., 2025 Form 10-K (2026) | 2025 purchases of property and equipment | $69.69 billion | Reported purchases during 2025. |
| Meta Platforms, Inc., 2025 Form 10-K (2026) | 2026 capital expenditures | Approximately $115 billion to $135 billion | Company forecast, not realized expenditure. |
| Alphabet Inc., 2025 Form 10-K (2026) | 2025 capital expenditures | $91.4 billion | Reported for 2025. |
| Alphabet Inc., 2025 Form 10-K (2026) | 2025 property-and-equipment depreciation | $21.1 billion | Reported depreciation expense, not capital spending or a measure of physical wear. |
| Microsoft, FY2026 Q3 earnings call (2026) | Calendar-year 2026 capital expenditures | Approximately $190 billion | Forecast stated on the call; Microsoft attributed approximately $25 billion of it to higher component pricing. |
The labels matter. Purchases of property and equipment, capital expenditures and depreciation are related but distinct financial measures. The periods also differ: a calendar-year forecast is not the same thing as a fiscal-year result. Microsoft’s call described its 2026 figure as a forecast, while the 2025 figures above are reported results.
How long do companies estimate the equipment and facilities will be useful?
Company disclosures describe useful-life assumptions for particular asset categories, not a common expiry date for all AI hardware. In Amazon’s case, the figures are an executive’s description; Meta’s figure is an accounting estimate for most server and network assets.
| Company and source | Asset category | Stated period | What the figure represents |
|---|---|---|---|
| Meta Platforms, Inc., 2025 Form 10-K (2026) | Most server and network assets | 5.5 years, effective January 1, 2025 | Estimated accounting useful lives; not a GPU-only estimate. |
| Amazon CEO Andy Jassy, company-published interview summary (2026) | Hardware and networking | About six years | Jassy’s description of Amazon’s assets, not an industry standard or a GPU-specific result. |
| Amazon CEO Andy Jassy, company-published interview summary (2026) | Data-center assets | 30-plus years | Jassy’s description of Amazon’s data-center assets; not the life of the servers or accelerators inside them. |
The separation between equipment and facilities is consequential. Replacing servers does not necessarily mean replacing the data-center building, and a building’s long service life does not imply that its installed accelerators remain competitive for the same period.
When does depreciation start, and why is that not the same as spending?
Alphabet says depreciation begins when property and equipment are ready for their intended use. A cash payment or construction commitment can therefore precede depreciation: Alphabet notes that data-center projects can take multiple years, with assets remaining under construction or assembly before they enter service.
This creates several dates worth keeping separate in any analysis: when cash is spent or committed, when construction or installation is completed, when an asset is ready for use, when depreciation begins, and when the capacity contributes to revenue. These dates need not coincide. A company’s reported annual capital spending cannot, by itself, show how much capacity was already earning revenue during that year.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Does an accounting useful life tell you when a GPU becomes obsolete?
No. A useful-life estimate is an accounting judgment about expected benefit, not proof that an asset will fail physically or stop being useful at the end of that period. Nor does it establish how quickly a particular accelerator generation loses competitive value.
Alphabet says estimates can change with asset performance, expected technology advances and future network deployment plans, and that changes may affect its financial condition and operating results. Those factors explain why a useful-life estimate can be revised; they do not translate it into a measured GPU obsolescence date. The company disclosures discussed here report aggregate categories such as server and network assets, not a harmonized, GPU-only measure of economic life across providers.
In general, estimating a shorter accounting life recognizes an asset’s cost over a shorter period. That accounting treatment should not be confused with an immediate cash outflow, physical failure or a quantified change in earnings: the effect depends on the relevant company’s circumstances and disclosures.
What should readers look at instead of a single lifespan number?
To judge whether an AI infrastructure investment is keeping pace, separate the asset’s age from the business case for continuing to use it. The relevant evidence is different for a data-center shell, a power system, a network and a server fleet.
- Asset category: check whether a disclosed life applies to a building, power equipment, servers, networking or a narrower class. An aggregate server figure should not be presented as the lifespan of a specific GPU.
- Deployment status: distinguish announced spending and construction from equipment ready for intended use and capacity actually deployed.
- Utilization and demand: ask whether installed capacity is being used and whether the services or customer workloads it supports generate revenue.
- Relative economics: consider performance, power and operating costs alongside newer alternatives. Operational survival alone does not show that continued use is the best economic choice.
- Accounting assumptions: read useful-life estimates as company-specific judgments, noting their effective dates and the asset categories they cover.
- Forecast versus result: keep guidance separate from completed spending and keep financial measures with different definitions separate.
The public figures available here illuminate investment scale, the gap between spending and monetization, and the different accounting assumptions companies apply. They do not establish a universal half-life for AI hardware or show exactly how long a particular generation remains profitable under different workloads.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
