Jensen Huang: Even if competitors' AI products were free, Nvidia's chips would still be...

midian182

Posts: 9,745   +121
Staff member
A hot potato: Nvidia has become the dominant force in the AI hardware industry, and CEO Jensen Huang knows it. The leather jacket-loving boss is so confident in the ability of his products, he says that even if the competitors' chips were free, they would still be a worse option than Nvidia's expensive alternatives.

Speaking during a keynote at the 2024 SIEPR Economic Summit, Huang was asked by John Shoven, the Charles R. Schwab Professor Emeritus of Economics at Stanford University, about the prospect of rivals that could offer competing chips that were "good enough" to replace Nvidia's and at a much cheaper price.

Huang started his answer by complaining that Nvidia has more competition than anyone on the planet, having to deal not only with direct competitors but also customers who buy its AI products in order to design competing versions. He also said that Nvidia shows these customers the current and future designs of its chips, partly due to Nvidia's completely "open-book" policy that sees it working with almost everyone in the industry.

Huang's claim of openness and working with competitors will likely raise some eyebrows. According to a report from last month, hardware startup Groq, which creates AI chips for LLM inference, said Nvidia customers have to be secretive about acquiring or ordering AI acceleration tech from rival firms in case Nvidia decides to delay their orders as punishment. Former AMD vice president Scott Herkelman, who previously worked for Team Green, said "They [Nvidia] are the GPU cartel, and they control all supply."

Speaking about the price of Nvidia's AI accelerators and whether competitors' alternatives would offer better value for money, Huang said it's only those who buy and sell chips that think about the prices, while those who operate data centers think about the cost of operations.

The CEO added that Nvidia's chips exhibit an excellent Total Cost of Ownership (TCO) – a measure of both direct and indirect costs – thanks to factors such as time to deployment, performance, utilization, and flexibility. The TCO is so impressive, Huang claims, that even if competitors were giving their chips away for free, it wouldn't be cheap enough.

Although plenty will argue with Huang's claims of Nvidia's benevolence, it's easy to see where his hubris comes from. Nvidia is now the third-most valuable company in the world by market cap ($2.19 trillion) and could soon replace Apple in second position, all thanks to its dominance of the advanced AI hardware market. It's also propelled Huang to the 20th position on Bloomberg's Billionaires Index – his fortune now stands at $77.2 billion.

Permalink to story.

 
If Moore's law is logarithmic then ai makes it synergistic. It would be difficult for anyone to catch up imo even AMD who was on the fence on ai until Nvidia's stock became infamous.
 
With this level of arrogance, this house of cards will soon come crashing down. What happened to the air of wisdom of a while ago?

"Nvidia CEO Jensen Huang says his AI powerhouse is ‘always in peril’ despite a $1.1 trillion market cap: ‘We don’t have to pretend…we feel it’"https://finance.yahoo.com/news/jensen-huang-says-nvidia-always-014309828.html?

Has nothing to do with arrogance, its just reality really and it was funny too.

Its not "only" about building the chips which on paper can compete, but also the entire software ecosystem behind which Nvidia spent years and billions on, THIS is why all the big AI companies want Nvidia right now

Even if, lets say AMD could actually beat Nvidia on raw AI compute power, it would not be enough without the software ecosystem behind to support it - I doubt AMD can afford this development and even if they can, it will take years and take away from their CPU progress (which will lack R&D funds)

Nvidia has AI money coming in fast RIGHT NOW which all can be used for R&D, hard to see anyone catch up anytime soon

AMD might be a smaller player in AI market in some years, Nvidia will remain king

Who should be able to compete with Nvidia? If AMD truly wants to go this route, it would eat all their R&D funds and they will loose CPU marketshare
 
Last edited:
The Leatherman-man is now a Rock Star.

It's so funny but the gamers' only hope is Intel.

After so many years (decades?) I have given up hope that AMD will ever make a GPU that will be even remotely competitive with Huang's GPU's.
 
Has nothing to do with arrogance, its just reality really and it was funny too.

Its not "only" about building the chips which on paper can compete, but also the entire software ecosystem behind which Nvidia spent years and billions on, THIS is why all the big AI companies want Nvidia right now

Even if, lets say AMD could actually beat Nvidia on raw AI compute power, it would not be enough without the software ecosystem behind to support it - I doubt AMD can afford this development and even if they can, it will take years and take away from their CPU progress (which will lack R&D funds)

Nvidia has AI money coming in fast RIGHT NOW which all can be used for R&D, hard to see anyone catch up anytime soon

AMD might be a smaller player in AI market in some years, Nvidia will remain king

Who should be able to compete with Nvidia? If AMD truly wants to go this route, it would eat all their R&D funds and they will loose CPU marketshare
In terms of computing power, AMD's solution is already much faster than Nvidia, and considering the price difference, it becomes up to 4x faster, because you can buy two mi300x instead of one H100.

It's not in the interest of corporations that currently want to capitalize on AI to stay in Nvidia's hands and pay whatever the leather jacket wants. Meta, Google, Microsoft, Oracle are software companies, if you think they're not capable of creating an ecosystem, I'm sorry you have such a limited vision.
 
In terms of computing power, AMD's solution is already much faster than Nvidia, and considering the price difference, it becomes up to 4x faster, because you can buy two mi300x instead of one H100.

It's not in the interest of corporations that currently want to capitalize on AI to stay in Nvidia's hands and pay whatever the leather jacket wants. Meta, Google, Microsoft, Oracle are software companies, if you think they're not capable of creating an ecosystem, I'm sorry you have such a limited vision.
As I said, the ecosystem are the true magic of Nvidias AI chips. AMD lacks that and this means no serious AI companies are interrested in MI300 or MI300X.'

With optimized workloads and proper software in place, AMD is nowhere near Nvidia in AI power. And every AI company makes sure of that.

AMD is much cheaper for a reason. They deliver a half solution.

Building the actual ecosystem and software is the hard and expensive part.
 
As I said, the ecosystem are the true magic of Nvidias AI chips. AMD lacks that and this means no serious AI companies are interrested in MI300 or MI300X.'

With optimized workloads and proper software in place, AMD is nowhere near Nvidia in AI power. And every AI company makes sure of that.

AMD is much cheaper for a reason. They deliver a half solution.

Building the actual ecosystem and software is the hard and expensive part.
No, you just keep insisting on generic baseless allegories, without any critical thinking. Sit back and watch the castle collapse on your "king's" head, corporations are after profit margins, they're not Nvidia's little sheep.

It won't happen in a flash, but it will happen. It's inevitable.
 
No, you just keep insisting on generic baseless allegories, without any critical thinking. Sit back and watch the castle collapse on your "king's" head, corporations are after profit margins, they're not Nvidia's little sheep.

It won't happen in a flash, but it will happen. It's inevitable.
Thats business and meanwhile Nvidia will keep receiving billions upon billions which can be used in other segments going forward, they are not slowing down and AI demand will continue to be skyhigh in the next many years

AMD might have a complete solution ready when AI race is over :joy:

The fact that you thought that AMD, a CPU focused company, was going to beat Nvidia in the GPU sector is funny tho


AMD is the niche player in all GPU markets; Gaming, AI and Enterprise. Nvidia dominates all these markets with absolute ease.

Currently Nvidia is not even trying in the Gaming space and has the - by far - fastest gaming GPU while sitting on 80-85% marketshare, all while dominating Enterprise and AI market completely, which has their full focus. Money drips down into gaming.

Nvidia is prepping 5090 and 5080 for a Q4 launch. Meanwhile AMD readies Radeon 8000 series with no high-end SKU - Sigh. AMD lost their chance, once again.
 
Last edited:
In terms of computing power, AMD's solution is already much faster than Nvidia, and considering the price difference, it becomes up to 4x faster, because you can buy two mi300x instead of one H100.
AMD's MI300x was released well over a year after NVidia's H100; this is near to comparing two separate generations. And while AMD claims "up to" double performance, independent testing shows a much smaller delta.

Furthermore, you miss the salient point. In this market, it's all about the software stack. For instance, the transformer engine for H100 can easily provide 3X or even 5X performance on AI model training over what raw processing power would suggest. And the prevalence of Cuda-based software is something that AMD will require a decade to overcome, if ever.
 
AMD's MI300x was released well over a year after NVidia's H100; this is near to comparing two separate generations. And while AMD claims "up to" double performance, independent testing shows a much smaller delta.

Furthermore, you miss the salient point. In this market, it's all about the software stack. For instance, the transformer engine for H100 can easily provide 3X or even 5X performance on AI model training over what raw processing power would suggest. And the prevalence of Cuda-based software is something that AMD will require a decade to overcome, if ever.
True.

There's zero hope for other GPU brands for now

 
AMD's MI300x was released well over a year after NVidia's H100; this is near to comparing two separate generations. And while AMD claims "up to" double performance, independent testing shows a much smaller delta.

Furthermore, you miss the salient point. In this market, it's all about the software stack. For instance, the transformer engine for H100 can easily provide 3X or even 5X performance on AI model training over what raw processing power would suggest. And the prevalence of Cuda-based software is something that AMD will require a decade to overcome, if ever.
Does Zluda have any chance?

 
Does Zluda have any chance?

Nope

CUDA is Nvidia only

 
Does Zluda have any chance?
In counterpoint to Disiw's post, I would safely venture to state that the AI market will soon be far too large to admit only one single proprietary standard for GPU instruction sets. If Intel's peradventures with the X86 instruction set are any guide, NVidia will eventually be forced -- legally, if not technically -- to admit competition to CUDA.
 
I know all the AMD fans will not agree, but Nvidia is so much better than AMD in every single facet, that you couldn't even pay me to use an AMD card, or CPU for that matter tbh.
 
AMD's MI300x was released well over a year after NVidia's H100; this is near to comparing two separate generations. And while AMD claims "up to" double performance, independent testing shows a much smaller delta.

Furthermore, you miss the salient point. In this market, it's all about the software stack. For instance, the transformer engine for H100 can easily provide 3X or even 5X performance on AI model training over what raw processing power would suggest. And the prevalence of Cuda-based software is something that AMD will require a decade to overcome, if ever.
It was released about 7-8 months after, not a year.

Not to mention AI software companies are working within CUDA converters to other ecosystem. AMD made the right move by going open source.

In the long run, Nvidia will need to open up their ecosystem or they will be dead in the water when the industry move away from these handcuffs.

It will be Adaptive synch story all over again.
 
The software that can be emulated from an ecosystem to another...

Cmon, this is pure denial. If you think Nvidia will hold a monopoly on AI because of CUDA, then you don`t understand the power of open source.
Nvidia invented CUDA, AI software runs best on CUDA

AMD has no chance without CUDA support - ZLUDA is dead in the water
 
It was released about 7-8 months after, not a year.

Not to mention AI software companies are working within CUDA converters to other ecosystem. AMD made the right move by going open source.

In the long run, Nvidia will need to open up their ecosystem or they will be dead in the water when the industry move away from these handcuffs.

It will be Adaptive synch story all over again.

H100 was released one year ago. March 2023. Blackwell AI GPUs are coming later this year on prime 3nm TSMC tech, which AMD can't afford to use yet.

Nvidia won't have to do anything. They are years ahead of AMD and will push AMD even harder with all those AI money coming in. AMD has little to no chance.

Nvidias software stack is what matters. AMD has nothing. Open source won't save them. All the big AI companies want Nvidia for a reason.

Nothing will change anytime soon and Blackwell will obliterate MI300X in optimized AI tasks, just like H100 already easily beats MI300X with 3-5x in optimized tasks. You can cherrypick benchmarks all you want but AI companies don't use that software. They are using the Nvidia stack and this is the true magic.

AMD won't be able to keep up at all going forward. Nvidia has full focus on AI, AMD don't and AMD don't even have the R&D funds to compete. It's a lost cause really.

Nvidia invented CUDA, AMD did not. That is just reality.
 
Last edited:
Back