Even companies cannot afford Nvidia GPU

[PeachGr]

Cruise, a general motors self driving startup started it's own diy Chip, for it's autonomous driving feature like many tech companies doing those days.

Choosing their own path, according to analysts will cost them about 100 million dollars 

Quotes

Quote

The General Motors-owned self-driving startup is following in Tesla’s footsteps by forgoing its partnership with Nvidia to develop its own processors that will be deployed in Cruise vehicles by 2025.

 

Two years ago, we were paying a lot of money for a GPU from a famous vendor,” Carl Jenkins, head of hardware at Cruise, told Reuters.

He added, “There is no negotiation because we’re tiny volume. We couldn’t negotiate at all. So that’s why I said, ‘Okay, then we have to take control of our own destiny.’”

 

By 2025, Gartner predicts, half of the top 10 automotive original equipment manufacturers will move the practice in-house.

 

My thoughts

 Nvidia on it's keynote spent a lot of time talking about their chip on autonomous cars, but since Tesla is too big and doing it's own Chips (and as far as I know machine learning), and then a general motors company is too small to afford them, I wonder who will actually use them. 

It's is strange to me that 100 million is the price that a company will pay to go it's own path, but at the same time it's not enough to even negotiate with Nvidia. 

Then there is the prediction of even more companies to move out to do their own Chips.

I thought that Nvidia will start by losing money to be competitive, to create market share on a relatively new market, but I guess I was wrong on that 

Sources

The quotes are from morning brew emails so I can't Link them 

https://www.reuters.com/business/autos-transportation/upset-by-high-prices-gms-cruise-develops-its-own-chips-self-driving-cars-2022-09-14/

[05032-Mendicant-Bias]

Two years ago GM was competing with mining operation that happily paid 3K$ for a GPU for perhaps 60+% in profit margins. To this day, Nvidia has an enormous amount of unsold high end Ampere dies, I speculate they skimped on making lower end lower margins die.

The automotive industry works on high volumes but low margins. They can't overpay for hardware that prints money elsewhere.

I can see why nvidia ignored them two years ago.

 

As a trivia, a Tesla user reported he was gaining 800$/month mining crypto on his tesla. He didn't pay for Tesla's battery charging electricity, and used that power to mine. Miners are always after "free" energy, their biggest ongoing cost.

 

Of couse, mining demand is boom and bust, demand from car manufacturers is more consistent. I speculate this short term move to please investors already hurt Nvidia. Now Nvidia gets a stock plunge, a production glut AND has lost two years of adoption from the massive automotive industry. Shortsighted short term public traded company decision making at its finest.

[Taf the Ghost]

@PeachGrPer Unit Cost and Extended Support Fees. It's important to remember that, functionally, American car companies have to support cars with available parts for around 15 years. (I forget the actual by-law number, as I believe it's actually variable by State in the USA, but it's something around there as just the industry de facto.) If you amortize the R&D cost over those 15 years, especially when they need to make a run in 10 years and it's costing them 1k per wafer or less, it makes economic sense.

[Quackers101]

makes me hope for when intel brings out more fabs for chip manufacturing + intel Battlemage.