Chip Expert Explains Why Samsung Lags Behind in HBM Market Competition
Why is Samsung, the leader of the memory chip industry, lagging behind in the HBM competition? Is it possible to stage a comeback and counterattack TSMC in foundry?
During a recent podcast interview with CommonWealth Magazine, Chih-yuan Lu, President of Macronix and an academician at Academia Sinica, shared his insights on the current challenges and dynamics in the semiconductor industry, particularly addressing Samsung’s struggle to keep up in the high-bandwidth memory (HBM) market.
He discussed how, despite Samsung’s strong technical foundation, the company has faced significant delays in gaining Nvidia’s certification for its HBM products.
At the 2024 Computex event, Nvidia CEO Jensen Huang stole the spotlight, but one moment stood out—when a South Korean journalist asked him whether Samsung could be Nvidia’s partner.
Huang, usually enthusiastic, responded coolly, stating that while Samsung and SK Hynix are excellent memory partners, he was waiting for Samsung's HBM to pass testing, noting that Samsung still had engineering hurdles to overcome.
This left many in the industry surprised, as Samsung, once a dominant player, seemed humbled, waiting for Nvidia’s approval, while SK Hynix and Micron had already started shipping their HBM products.
Lu emphasized that the memory industry has always been Samsung’s stronghold, yet in the HBM sector, it is now trailing behind SK Hynix and Micron, ranking third in what has become one of the hottest areas of the market.
This gap has raised concerns about Samsung's market value, which is now only about half of Taiwan Semiconductor Manufacturing Company’s (TSMC) and has impacted South Korea's economic growth. Lu pointed out that despite Samsung’s superior technology, its focus on short-term profits may have caused it to miss out on long-term opportunities, particularly in AI-driven growth sectors like HBM.
Lu went on to explain the technicalities of HBM, describing it as multiple memory modules stacked together to increase bandwidth. This stacking allows for faster data transmission, which is crucial in AI applications where large amounts of data need to be processed quickly.
While the concept of stacking memory has been known for decades, Lu noted that it wasn’t a priority in the past because there wasn’t enough demand. However, with the rise of AI, the need for high-bandwidth memory has surged, and companies like SK Hynix have capitalized on this opportunity, while Samsung has been slow to adapt.
He also delved into the broader industry dynamics, touching on TSMC’s new "Foundry 2.0" strategy, which aims to redefine its business scope by incorporating advanced packaging and mask-making, while excluding memory production. This move, according to Lu, could be a strategic maneuver to avoid potential antitrust scrutiny from regulators in the U.S. and Europe. By expanding its definition of the market, TSMC can claim a smaller market share, reducing the likelihood of being seen as a monopoly.
As the conversation shifted to global politics, Lu addressed concerns about the potential impact of the U.S. presidential election, particularly former President Donald Trump’s remarks accusing Taiwan of "stealing" America’s semiconductor industry.
While Lu refrained from commenting on the political aspects, he acknowledged that political instability could severely impact the industry, noting that Taiwan’s success in the semiconductor sector has been supported by decades of political stability. He expressed concern that uncertainty surrounding the U.S. election could affect investor confidence, especially in sectors like AI and semiconductors.
On the topic of a potential AI bubble, Lu is noncommittal. He compared the AI boom to the dot-com bubble of the early 2000s, where investors poured money into companies that were not yet profitable.
While AI companies like OpenAI and divisions of tech giants like Google and Meta are currently operating at a loss, the fear is that the bubble could burst if these investments don’t yield returns, and the major institutional investors get cold feet. Lu suggested that the AI sector’s long-term success would depend on how widely and easily its technology becomes integrated into everyday life, just as the internet did post-dot-com bubble.
In closing, Lu’s remarks underscored the importance of long-term strategic thinking in the tech industry, whether in memory production, semiconductor manufacturing, or AI development. He pointed out that while companies like Samsung have the technical capabilities, their business strategies must align with emerging trends to maintain their competitive edge.
As of October 23, Samsung's fifth-generation high-bandwidth memory chip (HBM3E) remains uncertified by Nvidia, despite being delayed for over a year.
Edited by Kwangyin Liu
In this episode, I’ll take you through some of the most important developments in the tech world, particularly focusing on the situation at Samsung and the broader implications for the semiconductor industry. From the rise of high-bandwidth memory (HBM) to the challenges of artificial intelligence (AI), there’s a lot happening right now, and I’ll provide my insights on what it all means.
Let’s start with Samsung. Not too long ago, Samsung was an unstoppable force in the tech world, known for its dominance in memory chips and its ability to compete toe-to-toe with Apple. But recently, Samsung’s fortunes have changed, particularly in the area of HBM. I was at Computex in Taipei this June, and all eyes were on NVIDIA CEO Jensen Huang, the so-called "AI king." His every move was carefully watched, but something else grabbed my attention at the international press conference.
A Korean journalist raised his hand and asked Huang, “How do you see South Korean companies, especially Samsung, as potential partners for NVIDIA?” Huang, usually known for his enthusiasm, gave a surprisingly cold response. He said Samsung and SK Hynix are excellent memory partners, but that’s it. He then added that Samsung’s HBM is still undergoing testing and that they have some engineering problems to overcome. This was a moment of reckoning. HBM is crucial for NVIDIA’s GPUs, a high-end, expensive memory type that’s incredibly hard to manufacture. And yet, Samsung—once a leader in tech innovation—was now seen as lagging behind.
What made this situation more striking was that both SK Hynix and Micron had already begun shipping their HBM products, while Samsung was still waiting for certification. In an industry where speed and timing are everything, being late to market can have devastating consequences. As a result, Samsung now finds itself ranked third in a field it used to dominate. No wonder their market valuation is only about half of TSMC’s, and their struggles are even impacting South Korea’s economic growth.
So, what went wrong for Samsung? Why is their HBM falling so far behind?
From my perspective, the core issue isn’t that Samsung lacks the technical prowess. In fact, Samsung is still the best when it comes to single-chip memory production. The problem lies in the stacking of multiple chips—what we call HBM. Stacking memory is incredibly complex. You’re not just dealing with one chip; you’re dealing with 8 or 16 stacked on top of each other, which dramatically increases the technical challenges. It’s no surprise that HBM is so expensive to produce.
Interestingly, Samsung was one of the first companies to experiment with stacking memory chips. But back then, HBM wasn’t a significant business opportunity. The demand wasn’t there, and the market was small, so Samsung didn’t put much emphasis on it. It seems they underestimated how quickly AI would drive the need for more sophisticated memory solutions like HBM. Meanwhile, SK Hynix, traditionally seen as Samsung’s "little brother" in the memory industry, took this opportunity seriously. While Samsung brushed off HBM as a niche business, SK Hynix saw it as a crucial market and invested heavily. Today, SK Hynix is a leader in HBM production, while Samsung has fallen behind.
This doesn’t mean that Samsung has suddenly lost its technical edge. The reality is, Samsung is still the most capable memory manufacturer overall. Their current financial struggles are partly because the entire memory industry is suffering right now. When you’re the biggest player, like Samsung, a downturn in the market hits you the hardest. While smaller companies like my own, Macronix, may lose a few million dollars, Samsung’s losses can easily reach hundreds of billions. That’s just the nature of being a giant in this industry.
But there’s something more concerning about Samsung’s current situation. I believe that Samsung has become too focused on short-term profits. Their leadership has become too "business-oriented," almost like they’re thinking like a CFO instead of a CTO.
In a fast-moving industry like semiconductors, you need to have a long-term vision. You need to invest in technologies that may not pay off immediately but will be critical in the future. That’s where Samsung has stumbled.
They’ve been so focused on immediate financial returns that they didn’t adequately prepare for the rise of AI and the demand for HBM.
Shifting gears, let’s talk about another major player in the semiconductor industry: TSMC. Recently, TSMC introduced the concept of "Foundry 2.0," which has generated a lot of buzz. What they’re doing is expanding their business beyond just producing logic chips. They’re now including things like advanced packaging and mask-making in their business scope—essentially, they’re redefining what it means to be a foundry. However, they’ve made it clear that they’re not getting into memory manufacturing. By broadening their business definition, TSMC is positioning itself to handle more parts of the semiconductor production process, while still staying out of areas like memory.
Why is TSMC doing this?
One reason could be to avoid antitrust investigations from the U.S. or the European Union.
By expanding the market definition, they can argue that their market share isn’t as dominant as it seems. It’s a clever strategy, and it reminds me of what Amazon did when it faced antitrust scrutiny. Amazon argued that they weren’t just competing in e-commerce, but also against physical stores like Walmart, which reduced their market share on paper.
Finally, let’s touch on the topic of AI and whether we’re heading toward an AI bubble. Right now, everyone is talking about AI, and companies are pouring massive amounts of money into AI development. TSMC is claiming that its AI-related products, like its CoWoS packaging for AI servers, are in such high demand that they’ll be sold out until 2026. But does this mean that AI’s growth is sustainable, or are we in the middle of a bubble?
I think the signs of an AI bubble are already there. AI companies, including giants like Google and Meta, aren’t making money from their AI divisions. They’re spending billions on data centers, AI servers, and software development, but none of these ventures are profitable yet. It reminds me of the dot-com bubble in the late ’90s. Everyone was convinced that the internet would revolutionize business, and while it eventually did, many companies went bust along the way.
The same thing could happen with AI. Right now, investors are throwing money at AI companies because they believe in the long-term potential. But if these companies can’t start turning a profit, investors will eventually pull back, and that’s when the bubble will burst. When that happens, the entire industry will suffer. The question is not if the bubble will burst, but when.
At the end of the day, for AI to be truly successful, it has to become as accessible and easy to use as the internet is today. Right now, AI still feels complicated and expensive, but that will change. Eventually, AI will become just another tool that everyone can use without thinking twice about it, just like setting up a website is today.
In summary, Samsung’s HBM struggles, TSMC’s Foundry 2.0 strategy, and the potential AI bubble are all interconnected pieces of a larger story about how the semiconductor and tech industries are evolving. These are complex times, and it’s clear that the decisions companies make today will have far-reaching implications for years to come.
Have you read?