Rent it from a hyperscaler. Build a server room. Rack it in a colo.
Or run it in a NERV Centre on premises you control — bought outright,
or leased monthly with support included. Same capability, five ways to
pay for it. Every assumption is visible below, and you can change it.
every
yearsre-buys the GPUs on the three owned options — cloud & HaaS already bundle refresh into their rates
Total cost of ownership
Cumulative cost by year
Where the cloud line crosses the owned lines is where renting stops making sense.
Model assumptions (editable)
Data table — every year, every option
Want this run properly for your site? We'll prepare a full
engineering breakdown for your region, power tariff, workload, and
floor plan — including what happens to each option when the uplink
is cut.