The main cause of Nvidia's crazy valuation is AMD's unwillingness to invest in making its GPUs as useful as Nvidia's for ML.
Maybe AMD fears antitrust action, or maybe there is something about its underlying hardware approach that would limit competitiveness, but the company seems to have left billions of dollars on the table during the crypto mining GPU demand spike and now during the AI boom demand spike.
I like to watch YouTube retrospectives on old failed tech companies - LGR has some good ones.
When I think of AMD ignoring machine learning, I can't help imagine a future YouTuber's voiceover explaining how this caused their downfall.
There's a tendency sometimes to think "they know what they're doing, they must have good reasons". And sometimes that's right, and sometimes that's wrong. Perhaps there's some great technical, legal, or economic reason I'm just not aware of. But when you actually look into these things, it's surprising how often the answer is indeed just shortsightedness.
They could end up like BlackBerry, Blockbuster, Nokia, and Kodak. I guess it's not quite as severe, since they will still have a market in games and therefore may well continue to exist, but it will still be looked back on as a colossal mistake.
Same with Toyota ignoring electric cars.
I'm not an investor, but I still have stakes in the sense that Nvidia has no significant competition in the machine learning space, and that sucks. GPU prices are sky high and there's nobody else to turn to if there's something about Nvidia you just don't like or if they decide to screw us.
In fairness to AMD, they bet on crypto, and nvidia bet on AI. Crypto was the right short term bet.
Also, ignoring is a strong word: I’m staring at a little << $1000, silent 53 watt mini-PC with an AMD SoC. It has an NPU comparable to an M1. In a few months, with the ryzen 9000 series, NPUs for devices of its class will bump from 16 tops to 50 tops.
I’m pretty sure the linux taint bit is off, and everything just worked out of the box.
Toyota is extremely strong in the hybrid car market, and with ravenous competition for electric cars and slowing demand Toyota may have made the right decision after all
There's also just the idea of endeavour - Nvidia tried something, and it worked. Businesses (or rather their shareholders) take risks with their capital sometimes, and it doesn't always work. But in this case it did.
> I think this could be cultural differences, AMD's software department is underfunded and doing poorly for a long time now.
Rumor is that ML engineers (that AMD really needs) are expensive; and AMD doesn't want to give them more money than the rest of the SWEs they have (for pissing off the existing SWEs). So AMD is caught in a bind: can't pay to get top MLE talent and can't just sit by and watch NVDA eat its lunch.
> So AMD is caught in a bind: can't pay to get top MLE talent and can't just sit by and watch NVDA eat its lunch.
This isn't being caught in a bind. This is, if true, just making a poor decision. Nothing is really preventing them from paying more for specialized work.
I find this strange to believe. Every big company has levels, unless your existing L7+ IC is below market, you can just pull L7+ salaried ML engineers with some secret signing bonus like literally everyone else.
The dirty secret in the tech industry is that most people at AMD or Intel or IBM and historically Nvidia/Oracle (this changed post 2022), were the 2nd-3rd tier tech companies. Staffed heavily by the rejects of the FAANG, they were still happy to have their 100-200K in their MCOL areas, but no free food and a much more boring work culture. Intel's "great place to work" corporate propaganda was known as "great place to leetcode" while I worked there, as Intel was always seen as a stepping stone before you "made it" in a FAANG.
Culturally, none of these companies were happy to pay anyone except the tip, top "distinguished" engineers more than 300K. AMD seems to be stuck in this mentality, just as IBM is.
> AMD seems to be stuck in this mentality, just as IBM is.
And that's why creative destruction is essential for technological progress. It's common for organizations to get stuck in stable-but-suboptimal social equilibria: everyone knows there's a problem but nobody can fix it. The only way out is to make a new organization and let the old one die.
AMD fears anti-collusion action, remember, CEOs of the two are just barely far enough of kinship to not be automatically considered colluding with each other.
The companies' CEO's are related. My conspiracy theory is that they don't want to step on each other's toes. Not sure if that works with fiduciary duty, though.
It does not conflict. Fiduciary duty for a for-profit organisation is not "profit at all costs", it's "you have to care about the company (care), you have to do good business (good faith) and you can't actively waste investors' and shareholders' money to intentionally lose out (loyalty)".
If they are found colluding due to nepotism, both will get a very swift revocation of business licence and a huge prison term. Remember they are just one step of kinship away from presumed collusion.
Maybe AMD fears antitrust action, or maybe there is something about its underlying hardware approach that would limit competitiveness, but the company seems to have left billions of dollars on the table during the crypto mining GPU demand spike and now during the AI boom demand spike.