N. Baratte - TechStock01

N. Baratte - TechStock01

Nvidia’s Jensen at Computex: 3 Big Announcements, and Smaller Ones

Nicolas Baratte - TechStock01's avatar
Nicolas Baratte - TechStock01
May 19, 2025
∙ Paid
1
Share
  • Nvidia will provide IP blocks to other semiconductor designers (ie hyperscalers Amazon, Google and their design partners Alchip, Mediatek) so that they can integrate their own CPU or ASIC into Nvidia’s systems.

  • Nvidia launching a small size RTX Pro Server for Enterprise to run their own AI models, AI agents, etc. that’s “on-premise” AI basically. AI Data Platform or, analogy with SQL, a tool to structure and train AI data and models.

  • Mediatek is called a “great partner”. Lots of mentions of Hon Hai / Foxconn, TSMC. No mention of Broadcom.

The big picture take away from Jensen presentation is that Nvidia is morphing into an everything system company. For hyperscalers, compatible IP to custom design chips complementary to Nvidia’s. For Enterprise, on-premise AI servers. For Enterprise and Industrials: data and model training platform. And other smaller announcements for Automobile, Robotics, Industrial digital twins, etc.

Which brings into context these 2 quotes from Jensen:

“it gives me tremendous joy if you just buy something from Nvidia”.

“our partners use whatever they want to use. We provide our technology in a way as easy as possible to integrate. As I said, I love it if you buy everything from me, but please buy something from me”.

This means: I know that you cannot / don’t want to buy everything from Nvidia, so I am becoming a platform company. Very smart.

Share

For the AI doubters, if you want to see examples of AI in the physical word, or how widespread AI applications will reach, watch from 1h50min

Friends and Family stuff:

  • Taiwan is essential

  • Nvidia + Foxconn + Taiwan Government are building a big Data Center in Taiwan

  • 2nd HQ “Nvidia constellation” will be built in Beitou / Shilin (north of Taipei where my in-laws live. You need to know that).

Announcement 1: NVLink Fusion for semi-custom AI infrastructure.

100% Nvidia systems are available: CPU, GPU, InfiniBand, NVLink switch. NVLink can scale up racks

But Nvidia will also support custom ASICs / TPU to be integrated into a large scale-up system with NVLink chiplet. Proprietary CPU integrated into Nvidia Blackwell / Rubin, with NVLink chiplet. Anybody can integrate into Nvidia: bring your own CPU, ASIC, etc.

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2025 Nicolas Baratte
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture