DeepSeek V4 will CRUSH U.S. Tech Stocks and Corporate Jobs
Mike Adams warns that DeepSeek V4, a Chinese model with 1 trillion parameters running on domestic Huawei chips, will crush U.S. tech stocks by rendering Google and Microsoft obsolete while replacing middle-manager jobs. He argues failed sanctions cannot stop China's rising chip dominance, predicting NVIDIA profits will surge from selling high-end workstations despite the shift to free, locally run AI. Ultimately, this technological pivot signals a geopolitical realignment where China secures global AI leadership, necessitating personal preparedness for potential domestic disruptions. [Automatically generated summary]
We are probably now just days away from the release of DeepSeek version 4, which is it's almost certain to be the most impressive frontier model that's open source in the world.
It is, of course, released by the Chinese company DeepSeek, which shocked the world a little over a year ago by releasing DeepSeek R1, which was called the Sputnik moment. for reasoning AI.
And R1 was really remarkable.
It was the first AI that did thinking tokens.
It would reason through problems.
And it just showed the innovation of the DeepSeek team led by some really talented and innovative engineers.
And now DeepSeek has been working for the last year on building this new multimodal model that can, apparently, it can do images, it can do, of course, text and reasoning, but also video.
I don't know that I believe it until I see it, but it's supposed to be able to do video and images and other modalities.
In addition, it's supposed to have 1 trillion parameters as a model size, which is very large.
And it's supposed to be a mixture of experts engine that has 32 billion parameters active at any given time, or for any given token, let's say.
So 32 billion parameter active mixture of experts model that is 1 trillion total parameters with a 1 million context window or 1 million tokens as a context window.
Now, if you don't know what that means, it's a game changer.
See, a context window is how much stuff you can put into the model in your prompt.
And right now, most of the world's models are limited to something like, well, maybe 256,000 tokens.
That's actually on the high end.
A lot of them will only support 128K tokens or the older ones, 64K or even 32K.
So to go to 1 million tokens means that you could put in a massive amount of context.
You could put in, you know, an encyclopedia almost.
You could put in, I don't know how many lines of code, but hundreds of thousands of lines of computer code, probably.
You could put in multiple book texts, you know, memory context, all kinds of things.
And it would process all of that efficiently.
Now, DeepSeek is the company that came up with DeepSeek Sparse Attention, which is a special algorithm that routes tokens and problems through its internal structure, its internal neurology in a way that only activates the necessary tokens while avoiding activation of all the tokens, which means faster throughput, lower compute costs, lower inference costs, etc.
Lower electricity usage per unit of cognition, etc.
Anyway, I'm not going to bore you with all the specs because they may or may not be meaningful to you.
My point is that as an AI developer myself, who's built numerous successful online platforms, and as someone who has created AI engines and worked with AI now for two and a half years, I can tell you that if the rumors are true, this is a game changer for human civilization.
This is a game changer that will be so disruptive that the effects of it are not well understood, I don't believe.
I'm going to try to discuss some of those with you here.
One of the effects will be probably a stock market crash in the U.S. among high-tech stocks, like maybe even Google, Microsoft, you know, Meta, et cetera.
Because once DeepSeek version 4 comes out, I think it will be obvious to people that China has by far the world's best technology and that even OpenAI can't compete.
And that this technology is being given away to the world for free.
where you can download it and run it locally on your own hardware.
And thus, you don't need to pay a subscription to OpenAI or Google or whoever in order to get really great coding or great answers or great whatever.
Now, I'm not yet convinced that the image model capabilities of DeepSeek will be world-class.
I'm doubtful.
I don't think they'll compete with models like Flux, for example.
I also don't think that the video capabilities of the engine will compete with, let's say, LTX or something.
I'm doubtful, but I'm open to the possibility.
It's just that I can't imagine how DeepSeek could create a world-class frontier engine with the best image generation and the best video generation and the best text generation all at the same time in one year.
I'm just skeptical of that.
I do believe that DeepSeek version 4 will be the best coding engine, certainly the best coding engine available in any open source model.
I have no doubt that that's true.
Will it beat CloudCode?
Maybe.
It's rumored to beat Cloudcode.
If it beats Cloudcode, then Anthropic is in trouble.
And hopefully Anthropic is scrambling to put out a better model.
I mean, their model is already amazing, Opus 4.6.
It's amazing.
I use it every day.
But they better get ready for Opus 5.0 or something if DeepSeek 4 launches.
And if DeepSeek 4 beats Opus 4.6 on the SWE bench for writing code, then, you know, Anthropic is going to be scrambling.
But here's the thing.
Even though I'm a user of Anthropic and I love the Anthropic Cloud Code features and functionality, it's written most of the code that runs my sites.
I would still switch to DeepSeek if it's better.
And especially if it's free.
I mean, I don't really mind how much I pay to Anthropic because it's worth it.
You know, it's a very valuable code writing engine.
It's incredible.
It's hard to imagine not using Anthropic.
But if DeepSeek is better, I'm going to switch to DeepSeek and I'm going to run it locally.
That's if it's better.
And I'm not the only one.
Everybody in the world who writes code, especially all these software companies and the corporations that have internal code and finance and everything.
Of course, they're all going to switch to the best model.
If that best model happens to be DeepSeek version 4, they're going to run it.
And they're going to stop using Anthropic or stop using Google Gemini or stop using whatever.
Why wouldn't you?
Wouldn't you want the best cognition?
So that right there could put incredible downward pressure on the revenues and profits of companies like Anthropic, which is already suffering under the U.S. government maliciously designating it to be a security risk or something.
That's all absurd.
Now, on top of that, though, you see, DeepSeek version 4 is a reasoning model with apparently a very high level of internal reasoning.
And what that means is that it can think through problems and it can offer incredible detailed problem-solving capabilities that will allow it to think through problems like a business manager.
It will be able to function like a middle manager in a corporate environment.
It will be able to make decisions about supply chains, logistics, manufacturing, scheduling, whatever, in the same way that humans do, or maybe even better.
As a result, this is the first model I believe that is going to begin mass replacement of middle manager jobs, white-collar jobs in the corporate workplace across the world.
DeepSeek version 4 could wipe out more human jobs than any AI model that ever came before.
And yet, at the same time, it could also add benefits to humanity at a scale that's never been witnessed before because it will do these jobs better than humans in many cases.
Perhaps in most cases.
We'll have to see.
But, you know, it's not just the engine itself.
Also, using engines to replace human jobs requires very clever prompting and programming and automation by capable humans who know how to work with AI.
So it's not automatic.
It doesn't mean it's going to happen instantly.
But if the model is capable, humans in the corporations are going to figure out how to deploy it in order to cut costs, which investors have repeatedly said makes your company more valuable.
Every time these companies cut jobs and say that they're being replaced by AI, Wall Street responds by hiking their stock price even more.
So there's going to be a financial incentive for replacing humans with DeepSeek version 4.
Now, it's going to be so dramatic that the federal government will likely at some point, Trump may announce, that DeepSeek version 4 is not allowed on government computer systems.
Okay.
They just don't want government to be efficient, obviously.
They don't want to replace the government workers, you know, because they've got a cushy job going there, benefits and pensions and vacations.
And every once in a while you show up and pretend to do something, you know, government jobs, especially federal government jobs.
But in the private sector, that's not going to fly.
In the private sector, you either replace or die.
That is, as a company, if your competitor is replacing 10% of the workforce with AI effectively and thus cutting costs without cutting quality or performance, then you, you have to do the same thing or you risk being obsolete, right?
So you as a company, you have to consider that.
And any company that's not using AI is going to be obsolete either way.
So the race is going to be on to implement and deploy AI as quickly as possible with the best engines possible, which is probably going to be DeepSeek version 4, which is also free.
Running Trillion Parameter Models on Consumer GPUs00:09:22
Now, I said before, it's a 1 trillion parameter model that is also 32 billion parameters active at any one time.
It's rumored that there's a version of this that's going to be able to run on a consumer grade 5090 GPU.
And I've got a few of the 5090s, but they only have 32 gigs of RAM.
I am very skeptical that you can run a 1 trillion parameter model in 32 gigs of RAM.
I don't think it's possible.
There is the 6000 Pro Blackwell GPU that has 96 gigs of RAM.
And I would suspect that, yeah, maybe you could run it there with an aggressive enough quantization.
But on a 50-90 card, I'm highly doubtful.
I guess we'll see.
Maybe there's a super light version that they're going to release at the same time that does fit on a 50-90 card.
We'll have to see.
But it wouldn't be the flagship version.
It would be something kind of scaled down.
Well, guess what?
NVIDIA just announced that systems integrators like Dell and HP and others, MSI, I think, are now beginning to ship the NVIDIA Spark stations.
These are sort of enterprise-class workstations that have the GB300 chipset, the Blackwell chipset, on an integrated motherboard with, I think, 496 gigs of RAM built in.
I think that's what it is.
And these systems are going to cost.
I don't know if it's $40,000 per system or $50,000 or I don't know.
They're going to be pricey.
Dell is apparently shipping these right now.
I've got my staff calling Dell to find out what they cost.
I don't yet have the answer.
HP is going to be shipping them next month.
MSI and other companies will be shipping these also, you know, maybe April, May, June, whatever.
NVIDIA, see, okay, let me back up.
NVIDIA is going to do well because NVIDIA sells the hardware that will run the DeepSeek version 4 full-size model.
Because if you have 496 gigs of RAM on your motherboard, that is, I'm talking about unified memory, high-bandwidth memory for the GPU, you can run DeepSeek version 4.
And so every one of these corporations, once they realize that DeepSeek 4 can replace even one person, it's going to be worth $50,000 to go out and buy one of these systems from, you know, Dell or whoever that has huge profits for NVIDIA.
I don't know how much NVIDIA profits per one of these workstations, but it's probably like 10 grand or something.
You know, there's a lot of profit.
NVIDIA is a very profitable company.
Because if you're a corporation and you're replacing even one human who costs you $150,000 a year, let's say, is that worth a one-time outlay of $50,000 for the workstation?
Yes, of course it is.
And probably this can replace more than one human.
So the race is going to be on for these corporations to go out and buy these NVIDIA spark station systems, as many as they can, and load DeepSeek onto it, the new DeepSeek.
And then human jobs start dropping like flies everywhere.
So effectively, the Chinese company DeepSeek will both clobber the U.S. tech stock market for the AI companies, including Google, Microsoft, et cetera,
but will boost up NVIDIA's valuation dramatically while also causing widespread unemployment in middle-level middle manager jobs in corporations across America as this cognition becomes more widespread and universally deployed.
And again, any company that doesn't deploy this cognition will be left behind.
I want to deploy this cognition.
I want to run DeepSeek version 4 on a full-size Spark station.
But again, I don't even know how much it costs yet.
I guess we're going to have to find out.
But this isn't, you know, a gamer on a desktop PC that costs a thousand bucks.
These are serious machines, $50,000 machines, I'm guessing, that can run human-level cognition, that can do thinking, that can solve problems, and that can think faster than humans.
They can do math and science and reading and writing and, you know, editing and freaking everything.
One workstation in a corporation will be able to serve an entire department of 25 people or something like that, depending on the usage patterns.
And it's going to make employees that aren't fired more valuable and more efficient, more effective in whatever it is they do.
So this is the trend of where all of this is going.
Is human jobs just keep falling like dominoes month after month as more and more capable engines get rolled out.
NVIDIA keeps making more billions of dollars of profit.
The U.S. AI tech companies become increasingly obsolete.
And China increasingly emerges as the world leader in AI technology.
And I believe all of those things are about to happen when DeepSeek version 4 is released.
Because I've seen people's comments who use the engine and they're blown away.
Incredible people who have used it.
They're blown away.
Is there a chance that DeepSeek would be disappointing?
That it would suck?
Yeah, okay, there's a chance.
I doubt it, though.
I don't think that's what's going to happen.
Because some of the best science papers on AI technology have come out of the DeepSeek engineers.
Like, what is it, manifold-constrained hyper-connections?
There's a training method that constrains the manifold between layers of the model in order to maintain coherence and reduce escalating noise factors.
Yeah, read that science paper.
When you understand that the DeepSeek version 4 engine was trained using that method, oh my God, that means it's going to be 10 times smarter, maybe, than other engines trained with the old techniques.
On top of that, did you know that the DeepSeek engine was trained entirely on Chinese microchips?
I think largely Huawei.
So this wasn't trained on Western chips from NVIDIA and H100s and H200s or whatever.
This was trained almost exclusively, is my understanding, on Chinese chips, Huawei chips.
And what this means is that all the microchip export controls from Trump and the White House that say to China, oh, you're not allowed to buy NVIDIA chips.
You can't have the H100s.
China is like, yeah, who cares?
We make our own microchips.
They may not be as advanced yet, but we can make a lot more than you can.
We could just scale it.
And that's what they're doing.
And by the way, China is also inventing their own UV lithography microchip fabrication technology, which means that within a couple of years, they won't even need chip fabs from anywhere else in the world.
They'll be entirely, domestically, self-sufficient.
They'll be able to make their own microchips and build their own AI models and do their own lithography, you name it.
And when they do that, combined with their abundant domestic energy and lower energy prices and strong skills in industrial output, China's going to become the microchip factory hub of the world.
And before long, it won't be that the U.S. will threaten China with sanctions, like you can't buy our microchips.
It's going to be China telling the world or telling the U.S., you can't buy our microchips because ours are better and we have a lot more of them.
You just do the math on this, connect the dots.
China will have more data centers, more electricity by far, more microchips, more engineers by far, better science, better math, better AI models.
China is going to dominate this.
And I think the pivot point is the release of DeepSeek version 4.
U.S. will suddenly realize, holy crap, we are behind.
We are no longer the world leaders in AI technology.
Yep, that's about to happen.
U.S. got lazy and complacent and stupid, while China got smart and clever and self-reliant.
Sound familiar?
Yeah.
So that's my guess, but we'll have to wait and see for sure.
Stock Up for Nuclear Survival Scenarios00:04:20
Nevertheless, you can follow my continued coverage of this topic at brightvideos.com.
You can read my articles at naturalnews.com.
And you can use my free AI tools, such as our book creation engine at brightlearn.ai, where you can now download 50,000 books free of charge.
And we've started producing audio books, full-length audio books.
You can find those at books.brightlearn.ai and just start downloading audiobooks completely free.
Or you can use our deep research engine that I also built.
That's at brightanswers.ai.
Or follow news analysis and trends at brightnews.ai.
I built all those engines using AI, mostly using Claude.
But maybe the next wave of those engines will be DeepSeek built.
Who knows?
I will keep you posted.
Thank you for listening.
I'm Mike Adams, AI developer, also the HealthRanger.
And again, check out all my videos at brightvideos.com.
And thank you for listening.
Take care.
Yes, the world is getting crazy, but here at the Health Ranger store, we're putting together a survival supply assortment for you.
If you go to healthrangerstore.com/slash survival, you'll see what we put together for you, including iodine and IOSAT.
That's a specific brand name of potassium iodide that's FDA approved.
Or we have the nascent iodine here, which is less expensive in terms of the iodine that you get.
These are available in case things go nuclear.
It's clear that you will not be able to find any of this for sale anywhere.
All the inventories will be wiped out like what happened after Fukushima in 2011.
So if you want to get your hands on some iodine, this is a chance to get it right now.
Healthrangerstore.com slash survival.
In addition, we have many other survival items for you here, including some silver solutions, some spirulina available in bulk and at a discount, and then a large assortment of storable organic food that's laboratory tested, including our Ranger bucket sets.
Here's a 195-day supply.
We've got the mini buckets, and we've also got number 10 cans available of freeze-dried fruits and vegetables and other things like miso soup powder.
Here's some of the buckets.
There's a big variety available.
Here are some of the number 10 cans right here.
Remember, a lot of people are missing fruit.
They don't have enough vitamin C in their storable food.
So, you know, getting bananas and pineapples and strawberries, especially, again, certified organic, freeze-dried.
That is the highest quality with the highest nutrient preservation that you can get in any kind of a storable food format.
All of this is available right now and so much more.
Just go to healthrangerstore.com slash survival.
And because the freeze-dried foods last for so long, you know, even if you don't eat them this year or next year, just keep them on the shelf.
They're going to last a very long time with good preservation, a long shelf life, and they will have value no matter what happens in the world.
Now, of course, I'm praying for peace.
I'm praying for de-escalation.
I don't want to see World War III break out, and I certainly don't want it to go nuclear.
But we're dealing with insane times and insane leaders and insane situations.
Who knows what could happen tomorrow or next week?
Disruptions could happen here in the United States.
There could be, you know, domestic attacks that disrupt supply chains here in the U.S.
So stock up early, stock up now, get your emergency food, emergency medicine, iodine, anything else that you think that you might need.
Get it now.
And by doing so, by shopping with us, you'll be supporting our platforms and our AI engines that we offer for free.
That's funded in part by sales from our store.
So shop with us at healthrangerstore.com slash survival and help yourself get prepared and also help us bring you more free tools and platforms that can keep you informed no matter what happens in the world.