- Ben's Bites
- Nvidia's new team will build custom AI chips for Meta and OpenAI.
Nvidia's new team will build custom AI chips for Meta and OpenAI.
Nvidia, the big dog in AI chips, is branching out into custom chips. The chip market is getting crowded and Nvidia wants to keep the throne by making personalized chips for AI companies.
What's going on here?
Nvidia is opening up new avenues beyond its traditional offerings of pre-designed chips.
What does this mean?
Nvidia holds a whopping 80% of the high-end AI chip market. Even beyond the stock tripling last year, Nvidia is 40% up YTD.
And still, it isn't resting. It is making a new team that’ll focus on making custom chips for AI companies. Its generalized chips (H100s, A100s etc.) do the job well but tech giants like Google, Microsoft and Amazon are shifting towards in-house chip development, aiming to cut costs and boost efficiency.
Reuters report that this team has met with folks at Amazon, Meta, Microsoft, Google and OpenAI to get them to bet on Nvidia for custom chips too instead of going all DIY.
Btw, this isn’t a new development. Nvidia’s CEO Jensen Huang has been wanting to get into offering custom chips since 2022. And this team is also talking to telecom, automotive and video game customers which total up to a $30B market. Dina McKinney is likely to lead this team.
Sidenote: Google gets its TPUs made by Broadcom, a big custom chips player and reports last year said that Google is thinking about ditching a significant chunk of the deal.
Why should I care?
As the demand for more specialized, energy-efficient chips grows, Nvidia's move to tailor-made chips and not just selling off-the-shelf solutions could boost the company even further.
To quote Anissa from The Information: Nvidia is arguably late to the custom chip field. Major cloud providers including Amazon and Microsoft have developed their own custom AI server chips, which they hope will lower their dependence on Nvidia. However, designing and continuously improving a chip that rivals Nvidia’s won’t be easy, so it is possible these customers could still want to work with Nvidia on custom chips.