How a Chinese Firm Restored Capitalism to the AI Space

January 27, 2025, 11:29 am ET, BY Thomas H. Kee Jr- Editor, Stock Traders Daily| Editor: Thomas H. Kee Jr. (Follow on LinkedIn)

How a Chinese Firm Restored Capitalism to the AI Space

DeepSeek Changes the Playing Field

• The AI buildout is going to get substantially cheaper.
• Trumps $500 Billion Rollout will be evaluated.
• Stock Traders Daily has first-hand knowledge.
• This sets the stage for capitalism to prevail.
• DeepSeek’s Model is a game changer.

DeepSeek has developed a model that is likely to change the playing field in the AI space by reducing overall costs.  If this is true, the profiteers from the AI build-out last year are likely to lose momentum.  (This assumes the conclusions are correct and can be validated).

This applies to Nvidia (NASDAQ: NVDA) , Broadcom (NASDAQ: AVGO), Advanced Micro Devices (NASDAQ: ADM), and suppliers.

The demand for top-tier chips is likely to decline considerably.  Here, we will attempt to quantify the potential impact on NVDA, with the expectation that it will reflect the industry.

Before we begin, today’s AI is not perfect.  For example. I used Copilot to help research the data herein, and Copilot was wrong (We assume copilot used the best AI buildout available).  The good news is, after responding with WRONG, Copilot fixed the problem, but it anyone took the first response on face value they would have had the wrong answer.  The current AIs are not perfect, even with the higher processing requirements.

From our research, the difference in cost between DeepSeek’s Model and those being used today is substantial.  The estimates seem to be $6mil to buildout AI using the DeepSeek Model, vs Hundreds of Millions of dollars being spent to do it now (far cheaper), yet the results are largely the same.

DeepSeek’s Model not only uses a lower tier chip, a less expensive one from NVDA (H800), but it also significantly reduces the processing power requirements by simply reducing the number of decimal places being used to compute the data. 

Stock Traders Daily has first hand knowledge of this, we compute data every day for our Predictive AI, and we ran into the same problem.  When we used four decimal places for the stock price quotes it required a vast amount of additional resources, vs the traditional two decimal places, yet the results were the same. 

• Therefore, we believe the probability of today’s news being true is high.
• Our Developers have also been using DeepSeek for a week, and the conclusion is that “DeepSeek is an absolute game changer for AI.  Open AI wants to charge tons for what DeepSeek gives you for pennies.”

From our research, which has been ongoing, the demand for the most powerful NVDA AI chip (GB200) has been robust, and Wall Street has priced in ongoing strong demand, but that may all change beginning immediately.

All projects not yet under contract will reconsider their options, and that includes Trump’s $500 Billion rollout.  Almost for certain, the investors in that project would prefer to invest far less for the same product, and they are doing their homework.

Every business should resolve to seek the best for their investors, so DeepSeek’s Model should be evaluated by all of them.  The question is, how many might prefer to use DeepSeek’s Model, versus the much more expensive one being used today.

Our conclusion is that those applications that require the utmost precision will still require the most powerful chips.  That includes AI applied to technology, healthcare, and cures for diseases, amongst others. 

However, the vast amount of demand will NOT need the most powerful chips, and therefore most of the upcoming evaluations will find value in using DeepSeek’s Model. 

We cannot be sure, but as an executive making a prudent cost analysis, ignoring the potential to save over a hundred million dollars would be a disservice to shareholders.

Therefore, we expect every executive to give this serious consideration now, and we expect many of them will turn to the most cost effective model.

 The United States will not lose its leadership role in the AI space, the industries that need the fastest and most powerful processors will still be in a position to lead, but the cost for everyone else will come down considerably. 

Interestingly, there will probably be a surge in demand in the AI space overall from lower tier companies, as a result.  Businesses that were priced out will now be able to compete, but that spike is not likely to offset the lower overall dollar value of demand as projections adjust down across Wall Street.

Not only will the demand for processing power change dramatically, but so will the projected buildout of new data centers to meet the projected demand.

• Might only one out of ten businesses actually need the best processors and the expensive buildout?  That is probably generous, but we’ll need to wait and see. 
• We believe most businesses do not need top tier processors.
• The ones who do will probably be the biggest spenders. 

Reasonably, we expect almost all executives to recommend the far less expensive buildout option, if it has comparable results, and that seems to be what they are saying. Executives will need to do this not only for sake of their shareholders, but also for their jobs, because if they ignore a massive cost savings they might not be trusted by their Boards.

We expect that 60% of the projected demand for AI processors this year will come from the top tier companies, who need the best processors, while the rest will be able to adjust to the less expensive Model.

Currently, Wall Street estimates that NVDA will sell upwards of 800,000 units of its top tier processor in 2025, with shipments set to ramp up in Q2.  We believe, anything not under contract will be reconsidered, and the demand estimates will likely adjust accordingly.

For example, if none of the chips set to ship were under contract, we would expect the estimated demand for NVDA’s GB200 chip to decline by roughly 320,000 units, but some of these, especially those being shipped in the beginning of the year, are under contract.  That will diminish as the year progresses.

DeepSeek’s Model still uses NVDA’s chips, so NVDA should see a bump in demand for lower priced chips as this adjustment unfolds, but the dollar amount spent will be far less.   The monetary value will be substantially lower, over 90% lower, for those that shift.

We believe similar demand adjustments will occur at competing firms like AMD and AVGO, but those companies may see opportunity if they have chips that compete with the H800.

Competition in the AI space was bound to happen, and here it is.

Recall, back when Intel was a player, chips were considered a commodity, and this DeepSeek revelation helps set the foundation for that to happen all over again.  Before the AI buildout last year, chips were more of a commodity too, the industry sways that way, and we are expecting that to happen again now that there is a way to produce the same result at a far lower price.

The playing field has changed, competition is likely to be restored, capitalism is set to prevail again, and it was all due to a small Chinese company, DeepSeek.

Support and Resistance Plot Chart for NVDA

Real Time Updates for Repeat Institutional Readers:

Factset: Request User/Pass

Bloomberg, Reuters, Refinitiv, Zacks, or IB users: Access Here.