Buy or lease GPUs? The honest comparison.
Both sides of this argument are usually made by someone selling one of them. We are on neither side: buyer-side on both structures, paid the same either way. Here is how the decision actually falls, and the numbers that decide it.
The decision in one table.
| Dimension | Buy | Lease |
|---|---|---|
| Upfront capital | Full hardware cost, plus freight, import and deployment | A deposit (typically 20%) secures the allocation |
| Time to live | Weeks to months: procure, ship, clear, rack, burn in | Roughly 8 to 10 weeks to a running cluster in a named facility |
| Balance sheet | Asset plus depreciation on the wrong schedule | A defined rate a finance team can plan around |
| Generation risk | Yours: residual value falls when the next SKU ships | Mostly the lessor's: renew onto the new generation at term |
| Operations | Your team or a hired operator, either way your problem | The facility operator is accountable for the floor |
| Exit | Resale or buyback, at whatever the market says that day | Structured at signature: renew, extend or walk away |
| Where it wins | Long, predictable, hard-running workloads; sovereign control requirements | Uncertain horizons, expensive capital, fast time-to-live |
The rule beneath the table: the hardware question is easy, the capital question is the real one. A cluster that runs at high utilisation for most of its life is cheaper owned. Everything less certain than that is an argument for the lease, and the less certain your roadmap, the stronger the argument gets.
Doing nothing is also a position.
Staying on on-demand cloud while deciding is itself a choice, and usually the most expensive one on the table: at sustained utilisation, on-demand rates run multiples of a reserved lease rate for the same silicon. If the decision is taking more than a quarter, the indecision has already cost more than the diligence would have. The cloud versus bare metal comparison covers that fork in full.
Current verified stock and indicative pricing for the buy side sits on the live inventory; the lease structure is on the GPU leasing desk.
Asked first, answered straight.
It moves with utilisation, horizon and your cost of capital, but the shape is constant: buying wins when the hardware runs hard for most of its useful life and you can absorb the upfront outlay; leasing wins when the horizon is uncertain, capital is expensive, or the roadmap says the next generation matters to you. We run both models on your numbers before you commit either way.
Usually somewhat, in exchange for what it removes: the upfront capital, the import and deployment work, the operations burden and the disposal problem. Whether that premium is worth it is a cost-of-capital question, not a hardware question, which is why the CFO belongs in this decision.
The uncomfortable truth of GPU ownership: the asset depreciates on NVIDIA's release schedule, not your accounting schedule. A lease prices that risk into the rate and hands it to someone else; ownership keeps the risk and the residual value both. Buyback at refresh softens ownership's downside and we structure it up front.
Often the right answer. A stable baseline on owned or long-leased nodes, the variable layer leased shorter or kept on-demand. The mix is a sizing exercise, and buying it as one requirement prices better than buying the parts separately.
Working through this on a real requirement?
Twenty minutes with the desk, no pitch and no quote at the end of it. Tell us roughly what you need and we will come back within one business day.
Sent. We are on it.
Your enquiry has landed with the desk. Acknowledged within one hour.