Investing

Is Advanced Micro Devices (AMD) the Biggest Winner From Monday's Bloodbath?

Rich Duprey

Published: January 28, 2025 11:46 am

Advanced Micro Devices (NASDAQ:AMD) fared better than most chipmakers during the tech sector rout on Monday.

Where Nvidia (NASDAQ:NVDA) was crushed by 17%, losing $590 billion in the process, and Broadcom (NASDAQ:AVGO) plunged by a similar percentage, AMD only fell 6.4%. While that puts it in the red for the year, it could make for the right time to jump into the chipmaker’s stock.

Trending Video

▶️

The DeepSeek revolution

The tech sector got crushed after Chinese artificial intelligence lab DeepSeek released its latest large language model R1 that outperformed across various benchmarks all other AI models. What made the feat especially pernicious is it was built on inferior, underpowered chips from Nvidia, yet was faster and more efficient. Worse, because it is open-source, users are able to modify the algorithm to meet their needs, fine-tune it, and build upon it. It also makes it free.

Why would businesses pay a $200 per month subscription fee to use OpenAI‘s ChatGPT Pro when they can use a better product for free? We might be saying R.I.P. to an OpenAI IPO.

A high-risk gamble

Yet it might not be all that bad. There are significant and serious national security concerns to using DeepSeek.

The AI shop collects massive reams of personal, financial, and operational data and stores it all in China. From names, dates of birth, email addresses, and telephone numbers to keystrokes, payment and transaction data, and following users across devices, these are virtually keys to every user’s personal and financial kingdom.

Congress banned TikTok for less and it seems unlikely DeepSeek will have a very long shelf life here. Nvidia should actually come out okay after this, and it actually hailed DeepSeek’s achievements. The chipmaker said, “DeepSeek’s work shows how new models can be created, leveraging widely-available models and compute that is fully export control compliant.” It argues it shows more of its AI chips will be needed.

DeepSeek opens new opportunities for AMD

Yet this is exactly the reason AMD should emerge in a better position. The runner up semiconductor stock has suffered through a series of analyst downgrades in recent days unrelated to DeepSeek. Most see it struggling to match the demand for Nvidia’s H100 chips that are used by data centers and hyperscalers, let alone the soon-to-be-released Blackwell chips. And Nvidia is already working on its next-generation Rubin chips.

Yet I see the DeepSeek revelation as a catalyst for AMD’s stock. While analysts are correct that enterprises were demanding its rival’s chips, what DeepSeek showed was that companies do not need to pay for premium-priced AI chips when cheaper, even underpowered chips, can produce a superior result.

The H800 chips DeepSeek used were developed by Nvidia to get around technology export restrictions imposed by the Biden administration on China. Nvidia significantly reduced chip-to-chip data transfer rates making them supposedly inferior to its H100 accelerators. Yet the Chinese AI lab was able to use work-arounds to overcome their shortcomings.

DeepSeek’s original V3 model only took 2,048 H800 chips to be trained, or the equivalent of 2.8 million GPU hours, versus the 16,384 H100 graphic processing units (GPUs), or 30.8 million GPU hours it took to train Meta Platform‘s (NASDAQ:META) leading Llama 3.1 model.

Further, the cost to train DeepSeek was about $5.5 million versus over $100 million for OpenAI’s ChatGPT.

Taking on the leader

One of the advantages Advanced Micro Devices has over Nvidia is that it’s MI300x GPU is touted as “the most advanced AI accelerator.” It is also cheaper and more powerful than the H100 chip. And sales have been brisk.

In the third quarter, AMD reported data center revenue shot up 43% higher sequentially and 38% year-over-year. It also notched record server CPU and data center GPU revenue, the latter of which exceeded management’s forecast of $400 million as MI300X GPU gained traction.

Although analysts suggest demand for AMD’s newest high-bandwidth memory accelerator MI325 is “lukewarm,” it will be releasing the MI350 later this year with the MI400 to follow in early 2026.

Key takeaway

AMD could regain attention for its powerful and cheaper AI chips in this new DeepSeek era. Knowing that a pricey, full-bore chip is no longer essential to produce a superior AI model, AMD could be the one reaping the rewards of the heightened interest in a low-cost version.

Especially since you don’t have to use a dumbed-down chip, but could get a fully functional chip to do the same work, this seems like the perfect opportunity to buy AMD stock.