Will Google’s TPU Break NVIDIA’s Monopoly? Why Samsung Electronics Could Be the Biggest Beneficiary
Editor
This morning, as I read the news that Samsung Electronics has recovered to 100,000 KRW in the pre-market, I couldn’t help but think that a truly interesting market shift is underway. KB Securities announced that they are maintaining a target price of 160,000 KRW for Samsung Electronics, citing it as the biggest beneficiary of Google’s TPU ecosystem expansion. But is this merely wishful thinking from analysts, or is it a realistic scenario?
Until now, the AI semiconductor market has been almost entirely dominated by NVIDIA. As of 2024, NVIDIA held a staggering 80% market share in the AI training GPU market. However, Google’s recent move to sell its self-developed TPU (Tensor Processing Unit) to external companies is beginning to alter this monopolistic structure. Notably, the news that Meta is considering equipping its data centers with Google’s TPU by 2027 is particularly noteworthy.
Personally, I believe this change is happening faster than expected. Google’s decision to start selling TPU, which was previously used only for its own cloud servers, indicates a significant level of confidence. Google’s TPU is known for its superior power efficiency compared to NVIDIA’s GPU, especially in inference tasks.
But why is Samsung Electronics seen as the biggest beneficiary here? According to KB Securities analyst Kim Dong-won, Samsung Electronics has a high market share in memory supply to North American big tech companies. In fact, Samsung holds about 42% of the global DRAM market and over 20% in NAND flash.
Changes in the Memory Market Due to AI Ecosystem Diversification
So far, big tech companies’ reliance on NVIDIA GPUs has concentrated memory demand on specific standards. For instance, NVIDIA’s H100 GPU uses HBM3 memory, which has led to a rapid growth in the HBM market. SK Hynix, which holds more than a 50% market share in the HBM market, has greatly benefited from this.
However, as Google’s TPU ecosystem expands, memory demand patterns may change. TPU uses a different memory architecture than NVIDIA’s GPU. Google’s latest TPU v5e uses both general DDR memory and high-bandwidth memory, which is advantageous for comprehensive memory suppliers like Samsung Electronics. Samsung is one of the few companies capable of producing DRAM, NAND, and HBM.
More importantly, there’s a strategy to diversify the supply chain. As analyst Kim mentioned, companies like Google, Broadcom, Amazon, and Meta are looking to diversify their memory supply chains. Until now, there has been an over-reliance on SK Hynix, but now there’s a move to distribute orders to Samsung Electronics. In fact, Samsung’s HBM supply began in earnest from the fourth quarter of 2024 and is expected to expand further in 2025.
Interestingly, while NVIDIA’s stock fell by 2.59% on the New York Stock Exchange yesterday, the M7 (Magnificent 7) companies generally showed strength. NVIDIA closed at $177.82, and AMD also fell by 4.15%. This can be interpreted as a signal that the market views the diversification of the AI semiconductor ecosystem positively.
NVIDIA quickly responded, emphasizing on its official social media that it is “a generation ahead of the industry and the only platform that powers all AI models and computing everywhere.” However, such a strong reaction could be seen as an acknowledgment of the threat posed by Google’s TPU.
Is the 160,000 KRW Target Price Realistic?
Let’s examine whether the 160,000 KRW target price for Samsung Electronics set by KB Securities is realistic. With the current stock price of Samsung Electronics around 100,000 KRW, they are seeing about a 60% potential increase. Is this reasonable?
Looking at the positive factors first, the memory market itself is expected to continue growing in 2025. According to market research firm Gartner, the global memory market size is expected to grow by about 17%, from $120 billion in 2024 to $140 billion in 2025. AI-related memory demand is particularly driving this growth.
Samsung Electronics’ memory business performance is also improving. The operating profit of the memory business in the third quarter of 2024 was 7.1 trillion KRW, marking a turnaround from the previous year’s loss. HBM sales also began to contribute significantly from the fourth quarter of 2024 and are expected to expand further in 2025.
However, there are concerns as well. Chinese memory companies are catching up quickly. Companies like YMTC and ChangXin Memory are narrowing the technology gap, which could lead to intense price competition. Additionally, concerns about an AI bubble have not completely disappeared. It is also questionable how long the excessive capital expenditure by big tech companies can continue.
Personally, I don’t think the 160,000 KRW target price is entirely unrealistic. However, this would depend on the expansion of Google’s TPU ecosystem and the actual diversification of the memory supply chain. The market seems to have already started reflecting some of these changes. I believe the recovery of Samsung Electronics to 100,000 KRW in the pre-market is a result of such expectations.
Ultimately, the key lies in how aggressively Google will sell TPU externally and whether big tech companies like Meta will actually reduce their dependence on NVIDIA. If these changes materialize, Samsung Electronics could indeed see significant benefits. However, investors should consider these uncertainties thoroughly when making investment decisions.
The shift in the AI semiconductor market landscape is something to watch with great interest. Whether NVIDIA’s monopoly will be broken, or if it will maintain its strong competitiveness, and how much Samsung Electronics will benefit in the process, remains to be seen.
This article was written after reading a news article and adding personal opinions and analysis.
Disclaimer: This blog is not a news medium, and the content is the author’s personal opinion. The responsibility for investment decisions lies with the investor, and no responsibility is taken for investment losses based on the content of this article.
