Real Alpha
Watch modeSubscribed

All-In · 2 min · Jul 23, 2025

Nvidia GPUs retain value as their software improves

Jensen Huang says Hopper GPUs retained much of their value after launch, with cloud supply still sold out. Software updates also lifted Hopper performance fourfold.

Jensen Huang

Founder & CEO, NVIDIA

Co-founded NVIDIA in 1993 and has led it as president and CEO, steering its GPUs from gaming into AI.

  • Chamath Palihapitiya, Host · Founder & CEO, Social Capital

    Led Facebook's growth team; his firm backed Slack and AI chipmaker Groq.

  • Jason Calacanis, Host · Founder & CEO, LAUNCH

    Early Uber angel; sold Weblogs Inc. to AOL; hosts This Week in Startups.

  • David Sacks, Host · Co-founder & partner, Craft Ventures

    PayPal's first COO, founded Yammer; White House AI & crypto czar 2025–26.

  • David Friedberg, Host · CEO, Ohalo

    Founded The Climate Corporation, sold to Monsanto for about $1.1B in 2013.

About this segment

Jensen Huang says Hopper GPUs retained much of their value after launch, with cloud supply still sold out. Software updates also lifted Hopper performance fourfold.

From “Winning the AI Race Part 3: Jensen Huang, Lisa Su, James Litinsky, Chase Lochmiller”, Jul 23, 2025. Watch the full episode

Answers “Is there still a GPU shortage?” See all 12 answers

“Hopper improved in performance by us and others by a factor of four in the time that we shipped it.”
Jensen Huangat 50:57 · All-In50:57

Takeaways

  1. Hopper held about 75–80% of its value after one year.
  2. Cloud supply of Hopper GPUs was still sold out.
  3. Software improvements increased Hopper performance by four times.

In this episode

Full episode
  1. 13:36Arizona fabs are reaching Taiwan-level yields at a modest premium4:41
  2. 33:59AI data centers are making energy and labor bottlenecks3:53
  3. 59:05DeepSeek showed why the U.S. technology stack matters2:39
  4. 3:12The DoD deal protects MP from below-cost magnet pricing3:26

  5. 48:57Nvidia GPUs retain value as their software improves2:22

  6. 10:18MP and DoD split risks they can and cannot control3:12Up next
  7. 19:46Lisa Su sees physical AI chips as a five-year-plus market3:39
  8. 7:21MP plans to train workers as it adds thousands of jobs3:03

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

  1. 49:02

    these days what's the lifespan now you you know i was looking into how they're amortizing you know these units four or five years what happens to this massive build out in your six seven and eight what will be the use of those computers if you keep building such great products that replace them at two three four times what do we do with all that concepts are happening right now the first thing first thing is every generation we increase the performance by x factors If the perf per watt goes up by X factors, whatever your data center power is, we just increase your revenues by X factors.

  2. 49:36

    So perf per watt is equal to revenues. Perf per dollar equals the cost. And so when we increase your perf per dollar by X factors, we reduce your cost by X factors. Does that make sense? That's the first idea. And so the reason why we're moving so fast is we're trying to increase everybody's revenues. We're trying to decrease everybody's cost so that we have the benefit of driving AI cost down as far as possible so that we can have thinking AI.

  3. 50:02

    It's not that we're trying to make AI so that it generates a thousand tokens and that's it. In the future, you're going to be generating millions of tokens, and that generates an answer as a result of that. You've got to think a long time, and so you've got to get that cost down. The second idea is, if you look at the residual value of NVIDIA gear right now, Hopper, for example, one year later, it's probably about 80%, 75% to 80% of the value, of the original value.

  4. 50:33

    And then one year later, it's another kind of like 65%. And then one year later, it's like 50%. And right now, if you try to get Hoppers in the cloud, it's all sold out. The reason for that is because CUDA is so programmable and we're constantly, the whole world, not just us, the whole world is doing open source development, improving its effectiveness. And so what's amazing is the performance of Hopper increases over time because we're improving the software stack.

  5. 51:02

    Hopper improved in performance by us and others by a factor of four in the time that we shipped it. Now, you can't get that out of a CPU. Right. Jensen, can you explain to us Elon's tweet and the impact to your industry?