A CFO’s view of GPUs: not what to buy, but how to fund an asset that loses value while it waits to earn its own return.
Most conversations about AI infrastructure start with the hardware. Which GPUs, how many, how fast can they ship. Those are the engineering questions. The finance questions come right after, and they are the ones that decide whether a company can scale AI on its own terms or ends up boxed in by its own balance sheet.
For a CFO, AI infrastructure is one of the largest capital decisions on the table, and it behaves differently from most of what came before it. Understanding how it behaves is the difference between funding growth and funding a depreciating asset you cannot fully use.
Why AI Hardware Is a Different Kind of Purchase
A traditional server refresh is predictable. You buy, you use the equipment for five to seven years, and you replace it on a schedule. AI hardware does not follow that pattern.
The chips are expensive, they are in high demand, and the performance improves quickly from one generation to the next. That combination changes the math. A large upfront purchase locks a big share of cash into an asset that starts losing value the day it ships and may be outclassed by a new generation before it is paid off. The question is not only whether the company can afford the hardware. It is whether buying it outright is the smartest use of the cash.
The Real Cost Is Cash Tied Up
The sticker price is only part of the story. The bigger cost for most growing companies is what that money could have done somewhere else.
Cash spent on hardware is cash not spent on hiring, product or sales. When a company drops several million dollars on GPUs, it is making a bet that owning the hardware outright beats keeping that money working in the business. For a mature company with plenty of cash, that bet can make sense. For a company still growing, tying up cash in equipment can slow down the very growth the equipment was meant to support.
This is where financing enters the picture, not as a last resort but as a deliberate choice about where the company’s money should sit.
How Financing Changes the Math
Financing does one simple thing: it turns a large one-time cost into smaller, predictable payments over time. Instead of paying for three years of hardware today, the company pays as it uses the equipment and as that equipment produces value.
There are a few common structures, and each fits a different situation. A lease keeps upfront cost low and payments predictable, and works well when a company wants to stay flexible and refresh often. An equipment loan spreads the purchase over time while the company still owns the asset. Asset-backed structures use the hardware itself as collateral, which can make approval easier and terms friendlier. The point is not that one is always right. It is that matching the structure to the situation keeps cash free for the rest of the business.
Financing is also non-dilutive. Raising equity to buy hardware means giving away a piece of the company to fund an asset that will lose value. Financing lets a company scale compute without handing over ownership to do it.
Depreciation and the Refresh Cycle
Because AI hardware improves so quickly, the refresh cycle matters as much as the purchase. Buying to own means the company carries the full risk that a chip loses value faster than expected. Structures that separate using the hardware from owning it forever can shift some of that risk and make the next upgrade easier, because the company is not stuck trying to sell equipment it already paid for in full.
Planning for the refresh from the start, rather than treating each purchase as permanent, keeps a company from being locked into last year’s hardware while competitors move to this year’s.
Power and the Longer Timeline
One more factor changes the economics: the hardware is only useful once it has somewhere to run. Power and data center space now carry long lead times, and that stretches the gap between spending the money and earning a return on it. When the wait between purchase and production grows, paying everything upfront ties cash up even longer before it does any work. Spreading the cost to line up with when the hardware actually goes live keeps the money from sitting idle during that wait.
What CFOs Should Take Away
The companies that scale AI well are not always the ones with the most cash. They are the ones that treat hardware as a financial decision, not just a technical one. That means asking how much cash a purchase ties up, how fast the asset loses value, when it will actually be producing, and whether owning it outright is worth the tradeoff.
The hardware question has a clear answer once the finance question does. A company that runs the financed version of the numbers before committing the cash keeps its options open: it can scale compute, protect its runway and stay ready for the next generation, instead of locking all three into a single upfront purchase.
