The profit margin for the Nvidia H100 accelerator is said to be around 820 percent. However, it is unclear exactly which factors this estimate takes into account.

An Nvidia H100 accelerator would cost Nvidia about $3,320 to produce. The chips are then sold to customers at a suggested retail price of €25,000 to €30,000. Depending on the effective selling price, the profit margin for Nvidia increases to over 900 percent. That’s according to a tweet (X?) from Tae Kim, senior writer for American finance week Barron’s.

Kim, in turn, refers to a calculation by Raymond James’ consultants, who carried out an analysis of the production costs. Unfortunately, we lack the details of this analysis. In particular, it is unclear whether the production costs only take into account the actual investments currently involved in the production of an H100, or whether the entire life cycle of the accelerator is taken into account.

R&D costs

After all, the Hopper H100 didn’t appear out of nowhere. Years of research and development by engineers preceded production. A single engineer at Nvidia would cost an average of $202,000 per year. In other words, Nvidia not only has to cover the production costs with the sale of the H100 accelerators, but also the entire research and development apparatus that enables the development of accelerators.

Anyway, this seems to be working fine. Last quarter, Nvidia posted $7 billion in revenue and made about $2 billion in profit from it. The numbers also give an interesting insight into the true cost of a Hopper H100. Additionally, the high demand for components to support complex AI workloads means that Nvidia can easily charge extortionate prices and still sell out its entire inventory.