yeah that's about right....gross margin at Nvidia is usually about 65%. They ordered a load of extra Ampere chips from Samsung in the first half of 2022 and Russia/Ukraine/inflation/interest rates disrupted consumer demand, so they're looking at about 60% gross margin for the year to 31 Jan 2023 (with a massive dip in Q2 and probably Q3) and about 66-67% gross margin in the year to 31 Jan 2024, which is when most of the Ada Lovelace chips will be sold...before coming back to 65%-ish in the year to 31 Jan 2025 as they clear Ada stock and get ready for the next gen (all subject to how harsh the coming recession is). Hopefully Intel coming back in as a leading edge fab will create a bit more competition post 2025, but with US/China tension building at the moment all the new fabs are breaking ground in the US, where people cost wayyyy more than in Taiwan. Maybe RDNA3's MCM/chiplet designs do for GPU what AMD did in CPU with Zen and we see way better yields, which could lower manufacturing prices? Also Raja Koduri (the guy who lead the Navi/RDNA project at AMD) is now over at Intel and their next gen (Battlemage?) could push more competition at the top. It's not just the cost to manufacture, market and distribute though (i.e. the costs reflected in gross margin).....Nvidia and AMD are spending $billions in R&D for each generation as well. And there are only so many gaming nerds and datacentres that are willing to pay up for the chips they're designing.
Don't get me wrong...on a personal level I'm ****** that top end GPUs cost more than a lot of people's cars nowadays. The engineering is amazing though....I'd venture Ada/Hopper chips are the most advanced thing mankind has ever built and a lot of the spend in improving performance is justified by AI/ML applications. It's great that the same tech can be used to chase higher FPS in video games. What a time to be alive