A training budget can change even when the model and machine stay the same. A price update needs to be evaluated against the number of paid hours a workload actually consumes, including time spent waiting or retrying.
Jarvislabs’ September 24 notice sets new rates from October 5 at 00:00 IST: H100 SXM moves from $2.69 to $3.49 per GPU-hour, H200 SXM from $3.99 to $4.59, and RTX PRO 6000 from $1.89 to $2.19. The figures are in US dollars, exclude tax and apply across regions. The company says existing committed nodes keep their contracted price. Jarvislabs rates effective October 5, 2026.
Price changes are not performance changes
For an identical H100 workload, the listed hourly increase is approximately 29.7%. That arithmetic says nothing about whether a different GPU, provider or configuration would complete the same task more cheaply. Such a comparison needs measured runtime and the same quality requirement.
Jarvislabs attributes the update to memory-supply costs. That is the provider’s explanation, not an independently established industry-wide cause. It also says storage rates are unchanged and directs buyers to request new reserved-plan pricing, so the notice alone cannot price a reserved deployment.
Make the budget workload-specific
The immediate question is which jobs are exposed to the new rates and which remain covered by existing commitments. Over time, higher hourly charges may make utilisation improvements more valuable, but only if they reduce paid time without increasing failure or operational effort. Watch the actual completed-job cost, rather than assuming the lowest advertised GPU rate is the cheapest useful service.
Image reuse: Akramusns / CC BY-SA 4.0. Original image retained; responsive display may crop the frame.
Sources & evidence
Sources describe the announcements and evidence available when this analysis was published. Company claims are attributed; plans and trials are not treated as sustained operation.
