Tesla Silicon Business

[Mich]

Hi Folks,

one of the biggest takeaways from yesterday's call is the Neural Network chip. On one side it's not a big surprise at all after Dan Keller joined, but on the other side I think it's not priced in at all.

So if Tesla has it's own Neural Network chip, more efficient and way faster than anything else on the market, what do they do with it beside AP3? Will they found a Tesla Silicone devision and sell the chip on its own? Will they take 30% of Nvidia's market?

This could be a big cash cow and a logical step after developing an own chip. But it doesn't fit i to Tesla's Masterplan.. What do you think?

 

[SwTslaGrl]

Disruption of AI chip businesses. A familiar pattern could be a flurry of press releases from other chip makers

 

[dw4ngg]

I have 8008135 reasons this is good for TSLA.

 

[elliott82]

I'm pretty sure they're not going into the silicone business...

Mod: I corrected the title even before I read this. --ggr.

 

[plumazul]

Hi Folks,

one of the biggest takeaways from yesterday's call is the Neural Network chip. On one side it's not a big surprise at all after Dan Keller joined, but on the other side I think it's not priced in at all.

So if Tesla has it's own Neural Network chip, more efficient and way faster than anything else on the market, what do they do with it beside AP3? Will they found a Tesla Silicone devision and sell the chip on its own? Will they take 30% of Nvidia's market?

This could be a big cash cow and a logical step after developing an own chip. But it doesn't fit i to Tesla's Masterplan.. What do you think?

So, Tesla is expanding into home improvement??? (or are we talking RL-AR??)

 

[Mich]

I'm pretty sure they're not going into the silicone business...

What did the book seller Amazon say when they were building their own infrastructure management layer? We are not going into hosting or infrastructure as a service! never!

 

[EVMeister]

What did the book seller Amazon say when they were building their own infrastructure management layer? We are not going into hosting or infrastructure as a service! never!

I think he was having a little joke about the use of silicone instead of silicon. Subtle difference in spelling but a very different meaning.

 

[elliott82]

yeah sorry, thought it was obvious. englisch sprache, schwere sprache

 

[neroden]

There is no "e" in "silicon". Please fix, because "silicone" is something different.

 

[gluu]

The chip architects name is Jim Keller, it seems that the actual main designer is Peter Bannon. Both are legendary in their field.

It will be interesting to see if Tesla will go after market segments outside of FSD with this chip. Could be used for voice controls, drones, manufacturing robot vision and control...

It seems like they’re following the Apple playbook so I don’t see them selling the chip to other OEMs...

We also need to see how it compares with the Google TPU and Intel Nervana when it comes out.

 

[adiggs]

Going after a more general purpose market means creating the rest of the infrastructure around using a chip in a more general purpose fashion. That might mean providing updates to an open source library so it auto-detects presence of the Tesla chip and optimizes software operations so that they use the chip functions.

It might mean building their own general purpose library.

The point is that there's a lot more to making a chip usable for a general purpose than "design chip and get it built" (presumably Tesla will be having it fabbed for them, rather than going into the chip fab business too - even Apple isn't fabbing their own).


If there's enough of a performance delta over alternative offerings, then maybe there's a market there that's worthwhile. Considering how many big fish Tesla's already got on the hook, I'd rather see them stick to making the design hyper optimized for themselves and their uses, and let the rest of the world figure out general purpose AI performance optimization (only because it's a field that has lots of focus, and Tesla has plenty to do already).

Full disclosure - I'd have also said that I'd rather see Tesla be using somebody else's AI enhancement chip, but they've obviously found an angle where there chip design can optimize their neural networks, and they've got enough future volume coming - I can see it. To make the effort worthwhile, they need a lot of performance enhancement AND it needs to be enough more and/or enough easier to implement, than a general purpose chip, to make the investment worthwhile. Assuming a general purpose application, and then investing resources along those lines - I'm dubious that Tesla is going to do it well enough to make it worth the time and money.

(I work in data science, and I work for one of those companies that's investing energy in building chips to optimize AI performance, though not in an area where I'm working on chips to optimize AI performance. I'm more of a consumer of these sorts of optimizations, than a builder of them).

 

[Sean Wagner]

I agree that this has the potential to be very significant, and it seems many people either do not realize what has been revealed, or choose to discount its importance. I was pretty gobsmacked.

Much depends on the competition of course, but Tesla has data, iterates quickly, and evidently assembled a very high quality team with an immediate, practical problem to solve.

During the Financial Times' market chat re latest Tesla results, only one other participant recognized the implications [or put it in writing].

Tesla is not always very efficient [borrowed a Britishism there], but to get this far on a small part of their R&D budget also merits to be mentioned [means the team effort is excellent].