News Ex-Intel CEO Pat Gelsinger loads up on Nvidia stock, says the market's reaction to DeepSeek is wrong

Reminds me of that scene in Wall Street Money Never Sleeps:

Main character to banker CEO: “what’s your target number, how much is enough?”

CEO: “More”

There’s more than enough for everyone’s need.

There’s no where near close enough for everyones greed.
 
You know things are rough at silicon valley when intel CEO buys Nvidia stock after a single open source AI model gets released
 
Last edited by a moderator:
"AI is not a bubble!" :rollseyes:

Regards.
I'm just gonna paste what I texted to a friend yesterday when they asked me how this works

"I think it's because of Nvidia, OpenAI, Microsoft etc fap wheel
They can move millions together doing that, hence AI *bubble*

Nvidia says they are going to make millions selling hundreds of thousands of GPUs because they're fundamental to train AI, so investors go and put a lot of money on it, then OpenAI buys the GPUs but say they're gonna make even more millions licensing ChatGPT and so investors go and put millions on that, and so on and so forth.

And so they build this huge and flimsy scaffolding to make profit out of it, that's holding a product that's basically all sellable promises but with mostly fake usefulness.

And then something like this happens and it all goes to heck🤷"
 
Last edited by a moderator:
Nvidia says they are going to make millions selling hundreds of thousands of GPUs because they're fundamental to train AI, so investors go and put a lot of money on it, then OpenAI buys the GPUs but say they're gonna make even more millions licensing ChatGPT and so investors go and put millions on that, and so on and so forth.

and then Microsoft create co-pilot and investors laugh, point to Bing, Nokia, Vista, W8 and put millions elsewhere
 
  • Like
Reactions: bit_user
Lol. I wouldn't bet on it. Now that we know that AI Models can be made without the high end Nvidia chips, even the Chinese can create these AI Models using lower end Chinese made chips within the next year or 2.
 
ai is a fad. Its hot as its new and advancing.

that wont last & unlike the last fad (crypto) ai has no real profit outcome....your normal person has no real use for it and will never pay for access to it to ever recoup the cost sunk into it.
 
Yeah Microsoft is just a bunch of nobodies. No influence on the pc market at all. People….
"Microsoft is going bankrupt" is second only to "I'm switching to Linux" as the most overused and ridiculous statements on the Internet.

If one listens to the Internet on these things, Microsoft was bankrupt by the turn of the century and 18 billion people are using Linux on their desktops. 😜

AI isn't going anywhere. I expect more of a shift in it than anything else. Frankly the data to train a true, Artificial General Intelligence does not exist. The only alternative is the Internet. Let's face it, there is no intelligent life there. I believe that way too much knowledge is stored only in the human brain to make AGI ever work properly.

I think what we'll see is more specialized AI. Think of a CAD program for an architect that not only knows structural analysis, but is also an expert on state and local building codes. THAT is a limited subset of knowledge that an AI could learn easily and be quite useful. But, it probably can't make an aesthetically beautiful building.
 
Yeah Microsoft is just a bunch of nobodies. No influence on the pc market at all. People….

Well most people don't understand where "AI all the things" makes sense. All the fapping has been around generative AI like chatbots or making professional looking meme pictures, but that's cause it's stuff regular people can interact with cheaply day to day. The real use cases are mining your own data to assist with answering questions or looking up corporate information. In the case of CoPilot, you can have a private tenant build that trains just on your own corporate information then lets you query it to tell you what's inside these 100+ page reports that you filed last year.

Our executive team is salivating on being able to use CoPilot to assist them in decision making by querying our own mountain of financial information. Like most corporations there is a bunch of sensitive private information that can't be put on the internet, so we're moving slowly and ensuring that it's properly walled off and the right security permissions are applied. For a user this would be the equivalent of having a private tenant that would have access to your purchasing history and "my documents" that you could then ask questions like "how much money did I spend on snacks last year" or "what did I buy eight months ago".
 
  • Like
Reactions: mitch074
ai is a fad. Its hot as its new and advancing.

that wont last & unlike the last fad (crypto) ai has no real profit outcome....your normal person has no real use for it and will never pay for access to it to ever recoup the cost sunk into it.

I would disagree. The problem is that western companies like Google, Meta and Microsoft spent so much money on AI but didn't really made any profitable products and applications out from it. China however, has already integrated AI into healthcare, manufacturing, energy and already demonistrated industry-specific applications for the use of AI. With Deepseek, more companies in China will integrate AI because the price of entry has been lowered.
 
  • Like
Reactions: George³
I don’t think Pat is the go to for future thought, cause he was in the position to shape the industry and didn’t get it done, but on this he’s right … the metaverse is wild and it’s turned the stock market in tombstone but then again stock market itself is imaginary. Stocks are always emotionally driven with a pinch of reality … optimized or not we are infants in this arena and if people think LLM’s are the end game you need to pick up an Asimov book … and with endless possibilities and the funds that inevitably must be spent to get there … the train is just warming up.
 
I think a big question (if the numbers claimed by DeepSeek hold true and broadly-applicable) is the effect this has on AI investment in general. Granted that a huge percentage has been towards the hardware and power requirements, but a lot is also spent on foundational R&D. Would relative R&D funding increase if hardware costs are greatly reduced?

I have a feeling, though, that costs for GenAI use are going to resist coming down anytime soon, partly for legit investment recoup reasons, partly for much greater profit now that the cost expectation has been set. A race to the bottom would look very interesting.
 
I would disagree.
"ai" itself is good, but the amount of $ corpo push it will never be made back up. Yes, "ai" can help automate some stuff easier, but that isn't making profit for those corpos.

investors don't care if "ai" becomes an every day thing or not they only care for profit in future & that has an end and many ppl are going to lose $ when its hit.

ai itself is fine but there is no long term profit like crypto had.
 
Well most people don't understand where "AI all the things" makes sense. All the fapping has been around generative AI like chatbots or making professional looking meme pictures, but that's cause it's stuff regular people can interact with cheaply day to day. The real use cases are mining your own data to assist with answering questions or looking up corporate information. In the case of CoPilot, you can have a private tenant build that trains just on your own corporate information then lets you query it to tell you what's inside these 100+ page reports that you filed last year.

Our executive team is salivating on being able to use CoPilot to assist them in decision making by querying our own mountain of financial information. Like most corporations there is a bunch of sensitive private information that can't be put on the internet, so we're moving slowly and ensuring that it's properly walled off and the right security permissions are applied. For a user this would be the equivalent of having a private tenant that would have access to your purchasing history and "my documents" that you could then ask questions like "how much money did I spend on snacks last year" or "what did I buy eight months ago".
Yeah I’m aware AI has tons of potential at data mining and optimization. I’m an engineer that uses parametric optimization software a lot and we’re trying to get AI to steer that process away from wasting testing results on clearly bad configuration.
 
  • Like
Reactions: mitch074
Mr. Gelsinger, respectfully, your Gaudi AI plan for Intel fell short of the hype and expectation, please sit down and stop inflating the AI bubble.
 
I mean, anyone with a lick of sense would buy that dip. It's like a freebie gift.

Nvidia still has a good several good years ahead of it, the need for its products is not going anywhere for now.
 
I mean, anyone with a lick of sense would buy that dip. It's like a freebie gift.

Nvidia still has a good several good years ahead of it, the need for its products is not going anywhere for now.
Short term yes, long term no, Nvidia was already overvalued at $3 Trillion dollars and DeepSeek's distillation and optimization techniques (8 bit floating point instead of 32 bit saves RAM, compression key indices saves RAM, multi-token prediction instead of single-token predictions saves RAM, mixture-of-experts saves RAM, all of which reduces compute requirements to the point you only need 2,043 H800s train DeepSeek. This optimization technique is extremely efficient, published on GitHub, and openly reproducible. The demand for Nvidia 's high-end chips will reduce in long-term.
 
Short term yes, long term no, Nvidia was already overvalued at $3 Trillion dollars and DeepSeek's distillation and optimization techniques (8 bit floating point instead of 32 bit saves RAM, compression key indices saves RAM, multi-token prediction instead of single-token predictions saves RAM, mixture-of-experts saves RAM, all of which reduces compute requirements to the point you only need 2,043 H800s train DeepSeek. This optimization technique is extremely efficient, published on GitHub, and openly reproducible. The demand for Nvidia 's high-end chips will reduce in long-term.
That's not how this works, because the demand indeed is not static and just like there are optimizations that will happen plenty, so will be new capabilities that will require more and more capacity.

Chat tools is not exactly the limit of imagination of what would be possible to do with AI. Nvidia will eventually truly dip, but only after there will be actual real mass-produced alternatives to their products. It will happen, but it will take a few more years still.

It's not a forever stock, at some point it will be right to dump, but definitely not now.
 
In fact, "AI" is currently another "dotcom" bubble, like "autopilot" before, the collapse of which was predictable.
Even specialized "AI" with sufficient benefit in some sector of human activity requires powerful resources and energy costs for something really useful, especially in everyday life
The 1st option using big data and neural networks in large data centers remotely - a priori will be paid and inaccessible to everyone, because the higher the fee - the greater the value.
The 2nd option, household (for example, on a laptop or smartphone) is now a priori impossible from a practical point of view, if you look at it from the perspective of real benefit.

Can anyone (standard test) - name at least some useful in practice in everyday life, for an ordinary person, trained "neural network" that is capable of working at least on 32GB of RAM and modern silicon from the point of view of accelerators in the mass segment?

Such neural networks simply do not exist in reality. When I ask such a test question, no one ever answers on the merits on any technical forum.

Serious local* (in your pocket, private) neural networks require resources that are several orders of magnitude greater than the capabilities of "pocket" silicon in 2025.

Large data centers do not have such restrictions - but the cost of a real valuable request to their large neural networks is simply inaccessible to the average person and will not be available in the next 20 years.

Yes, the super-rich will only increase their lead in the capabilities of multifaceted analysis of large volumes of information - so they are able to pay for requests to such neural networks and even get a personal neural network of such a scale. Social stratification in societies will only increase, not decrease. And the social stratification between societies that are advanced in this regard and less developed ones will increase even more.

But to achieve full “AI” (especially one that significantly surpasses the capabilities of even well-coordinated large teams of scientists and engineers), humanity will need centuries more.
 
That's not how this works, because the demand indeed is not static and just like there are optimizations that will happen plenty, so will be new capabilities that will require more and more capacity.
What are these "new capabilities"? o3 reasoning model? DeepSeek can distill o3's output for 3% of training cost and release it publicly as open source. OpenAI's business model is screwed, and guess who is one of Nvidia's biggest customers....
Chat tools is not exactly the limit of imagination of what would be possible to do with AI.
As far as I can tell, the $500 Billion StarGate AI is still about LLMs, and same for $65 Billion 1.2 million GPU mega cluster by Meta is still about LLMs. You don't need these multi million mega cluster farms when DeepSeek will just distill your output and release it public open source anyways.