Elon outs $16.5B Samsung chip deal Tesla asked to keep secret Samsung Electronics has scored a $16.5 billion contract to make the silicon to power Tesla's next-gen self-driving computer hardware. The firm is set to produce this from a new fab it is building in Texas, according to Tesla CEO Elon Musk. The Korean technology giant published a disclosure on its investor site today detailing …
The hardware that was meant to be ready for "full self driving"/"true autonomy" (the real thing, not the marketing name) is at least two generations away from being available. Unless I'm mistaking the meaning of strategic importance?
Seems to me we're stuck in the "we just need to build it bigger to make it work" loop common to all the other AI startups that are struggling to deliver robust generalisable real world functionality.
Yes, the envisioned self-driving car will require a trailing lorry filled with equipment to process the data and then send back the driving instructions. The lorry must have a qualified driver and will be pulling a small nuclear reactor to power the computers that are filling the lorry's loading area completely. Unfortunately, the lorry will be requiring diesel fuel because an electric lorry in that weight class does not work yet. The energy requirements of the combined computers and lorry are too much for any small nuclear reactor. An optional umbilical between the small nuclear reactor and the self-driving car has been envisioned to recharge the batteries while driving.
Seems to me we're stuck in the "we just need to build it bigger to make it work" loop common to all the other AI startups that are struggling to deliver robust generalisable real world functionality.
Actually, we're stuck in the much larger "we just need to feed the shared delusion that our company's valuation will keep going up and up and up" loop.
Corporate CEOs like Musk know that stock valuation is disconnected from (and typically runs contrary to) such real-world indicators as productivity, environmental impact, worker satisfaction, customer need, or technical viability. They have learned that valuation can be easily driven by vast, showy yet utterly wasteful expenditures. Hence the prevalence of over-priced acquisitions and mergers, as well as the farcical over-hyping of immature technologies like AI and 'self-driving.' Such moves are all money in the bank for clowns like Elon, even as they ultimately hollow out the actual worth of a company and wreak collateral damage in every direction.
What is really crazy is Nvidea I think has a P/E of 31, which is high, but they at least are growing qtr/qtr. Tesla on the other hand is 180 and has shrunk the past 2 qtrs. Worse, with trump in office, the milk of carbon credits is drying up and the Chinese EV makers are eating tesla's lunch abroad, so the future looks grim. So I have no doubt that musk wanted sammy to keep it secret until elon thought it was the best possible timing for a stock boost. And it appears to have worked. Stock is up even with the dreadful quarterly report. I give the guy credit, he plays the masses like a fiddle.
Nvidia is only growing because of the AI boom.
When that finally reaches the Singularity of Instability and craters, Nvidia goes with it.
At least Nvidia produces an actual viable product and it can pivot back to gaming accelerators. As for the rest, as Richard Feynman put it in a far different context, well, "there's plenty of room at the bottom."
"Such moves are all money in the bank for clowns like Elon, even as they ultimately hollow out the actual worth of a company and wreak collateral damage in every direction."
Considering Teslas recent performance and Musks admission that they are in for "some rough quarters" as Trumps anti-green policies remove the subsidies and carbon credits Tesla relies on, and how long it takes to get a new fab up and running, one wonders if Samsungs "exclusive" customer will still be in business by the time the first chips roll off the production line.
.. other car companies already have the chips but also use the far wider spread of sensors that make it easier to avoid children, emergency vehicles and lorries lying across a motorway.
Add to this Musk announcing to involve himself in the process and staffing that factory with competent people will prove a major challenge.
If Musk was able to run his vehicles on BS they'd be useful as there's no shortage of that, not just at Tesla..
Musk's problem is that whether or not his initial opposition to LIDAR made sense at the time- or was at least understandable- he's made his opposition so vociferously personal, and done that so publicly, that his ego will never let him admit he was wrong, and even if it did, he couldn't change track without it being seen by everyone else as backing down.
And we all know that's the last thing a raging-but-insecure narcissist like Musk would let happen.
Declining sales, shrinking profit margins, increased competition, and a share price that Elon seems to be maintaining by sheer force of will alone rather than any bright future - the question therefore is whether Tesla will still be around in its current form to take delivery of Samsung's chips by the time Samsung is able to deliver them...
We do seem to have reached the state where the share price is that much purely because that's what the people who invested want it to be worth.
Despite the apocalyptic company reports, there's no reason for it to drop until some other external force comes along that requires people to actually think about what their investments are worth, such as a recession or market crash. It's a good job America isn't currently being run by a mercurial, innumerate, narcissistic felon who holds grudges..... oh.
Previously you had to work one of Musk's companies for him to be able to act out his poorhouse manager fantasies on you but now in this exciting new development this bureaucratic requirement has been removed and Musk could turn up to your workplace one day to bully, push you around, and fascist salute in your face.
Doesn't mean Samsung will let him "walk the line". I'm sure they'll give someone spending $16 billion a nice tour, but unannounced visits to "accelerate progress" are not going to happen. I don't care how much you might spend at your favorite restaurant, they aren't going to let you into the kitchen to hassle the chef to get your food out faster.
With just a couple of (high end cameras).
Y'all think doing this stuff may be doing it a bit wrong?
Just a thought.
Mine's the one with a copy of Analog VLSI and Neural Systems in the oversize side pocket.
"With just a couple of (high end cameras)."
We don't "see" with our eyes. We see with our brains and eyes are just the photon collectors to feed data to the brain. We are also wired for pattern matching in a highly optimized way. It's very bizarre how much we don't really see since once we've taken in the data previously. The brain often uses the stored data rather than unprocessed fresh input. This is part of the explanation of why we may not see our keys on the table and pass right by a few times until our brain is forced to assimilate new input and, wham, there's the keys.
We also use our ears, our inner ears and tactile inputs when we drive. Once we've been driving for some time, a lot of what we do is on "autopilot" and it's only when something is off that we notice.
Programming a computer to do something as complex as driving at speed is more than throwing more processing power at it. Our grey matter runs at how many watts? There may be no brute force approach that makes any sense.
Not only don't we see with our eyes, per se, there's an approximately 300 millisecond delay on most sensory inputs1, so we're all living so very slightly in the past, a fact that I found very disturbing when I was told that by a neuroscience professor for whom I once worked.
Driving is literally about trying to predict the future and hoping you get it right.
It's a wonder anyone can ever hit a baseball.2
____________________
1 My understanding is that there are bunch of caveats and exceptions to that but for the most part, it's more or less correct.
2 Well, I can't. The Calvin and Hobbes baseball episode is, as they say in the movies, "based upon a true story."
I read a lot of human vision physiology stuff when I was doing my post-grad (in image processing) and IIRC some of the basic processing is done in the back of the eye. Your brain doesn't just get a feed of pixels it gets movement vectors and contrast information. I could be remembering it wrong, it was a non-trivial number of decades ago, but seems like that would be necessary for the "shit, something scary just jumped out of the shadows" reaction.
> shit, something scary just jumped out of the shadows
I recall driving down a motorway, in very heavy traffic, extremely monotonous, and I must have zoned-out. First I know about it is having come to an emergency stop inches behind the car in front.
Whatever Apps come pre-installed, they are pretty impressive.
> Your brain doesn't just get a feed of pixels it gets movement vectors and contrast information
So the eye does some pre-processing of the image before sending it to the brain
Also don't forget how many years of data learning the system has to undertake before it is capable of safely driving a car
Not many.
That was rather my point. Yet they do what a Petaflop of processing, based on modern estimates?
And it was Carver Mead's point in 1989. The book is the result of his teams work digging into how the brain does so much with so little power, and how to mimic it using a conventional CMOS fab process.
And it's been known since the 70's that in fact our eyes do substantial pre processing (David Marr, sadly no longer with us). So yes high end camera is a fair description of our eyes.
I don't know who voted me down. Probably the usual chorus of SEL's. I strongly doubt it's anyone who's read Mead's book.
This post has been deleted by its author
Is this one of those times where everone says big deals are happening but they aren't? What is the follow-through rate on multi-billion dollar PR announcements?
My bet is that Tesla doesn't have the money and Samsung doesn't have the building plans, but the announcement sounds good. Maybe it's for the senile orange guy to hear.
Musk has no grasp of the difficulty of the problem he's trying to solve with full-self driving. Any AI which is trusted with the lives of human beings will be held to excruciatingly high standards. Higher standards even than any human being. This is an almost impossible to climb hill.
If he thinks he can solve this by throwing more GPU or NPU power at it he's sorely mistaken.
I find this strange.. Samsung claiming that the name and other contract terms are being withheld "in accordance with the counterparty's request to maintain business confidentiality" makes no sense. After the Elongates Muskrat apparently blabbed about something that Sumsung was to keep secret at Tesla's request, CNN writes that "Samsung’s shares jumped as much as 6.8% to their highest since September last year after news of the deal. Tesla shares were up 1.9% in U.S. premarket trading."
Which suggests that keeping this secret would have benefited none of the parties and that Musk going public with it did. Or am I missing something here?
