Why doesn't TSMC need to envy SK Hynix? HBM is not the 'Storage Version of TSMC' | The True Moat for AI Investment | TSMC | SK Hynix | Intel | Google

$Taiwan Semiconductor(TSM.US) $NVIDIA(NVDA.US) $Intel(INTC.US) $AMD(AMD.US)

Live Stream Highlights | The True Moat in the Second Half of AI: Not HBM, but TSMC?
We've compiled the key takeaways from today's discussion with StockPro Xiaofeng on Caijing. Please like and follow before watching.
As this AI wave reaches today, the market is beginning to show a clear divergence.
On one side, TSMC's financial reports continue to hit new highs, but its stock price has started to fluctuate; on the other side, HBM concept stocks like SK Hynix and Samsung are facing market skepticism about whether the "super cycle" is ending after a round of crazy rallies.
The question worth pondering isn't "whether AI will end," but rather, who in the AI supply chain possesses a true competitive advantage that can transcend cycles?

I. TSMC and HBM: Essentially Two Completely Different Business Models
Many investors like to call HBM "the memory version of TSMC," but this analogy is actually inaccurate.
Xiaofeng points out that the biggest difference lies in "customization" vs. "standardization."
TSMC serves the world's most advanced chip design companies. Each chip requires joint development one to two years in advance, with deep binding in the production process, resulting in extremely high customer stickiness and forming a very deep technological moat.
In contrast, memory products like DRAM and DDR are essentially standardized commodities. As long as they meet industry specifications, products from different manufacturers can be interchangeable, meaning prices are ultimately determined by supply and demand, making cyclical fluctuations inevitable.
Simply put:
TSMC sells "irreplaceable capability," while memory manufacturers sell "standardized products."
This is why the market is willing to assign completely different valuations to the two.

II. HBM Is Indeed Stronger Than Traditional DRAM, But It Still Can't Escape Cycles
Over the past two years, HBM has become the hottest track in AI, leading the market to hope it can break the cyclical fate of the memory industry.
Xiaofeng believes that HBM indeed has higher technical barriers than DDR4 and DDR5.
This is because, besides the memory chips themselves, HBM involves non-standardized parts such as Base Dies, packaging, and co-design with GPUs, giving it more pricing power than traditional DRAM.
However, the problem is:
These non-standardized parts do not account for a high enough proportion of HBM's overall value to completely change the nature of the industry.
Therefore:

• HBM can extend the prosperity cycle;
• It can reduce price competition;
• It can improve gross margins;
  But it is hard to truly eliminate the cycle.
  Once the three major original equipment manufacturers fully expand production, the market may still return to a situation where supply and demand determine prices.

III. Why Did Xiaofeng Choose to Reduce Memory Holdings and Increase TSMC?

• Reduce some memory holdings
• Reduce Google holdings
• Reallocate funds to TSMC
  The reason is simple.
  He hasn't changed his long-term view on AI, but believes:
  If AI continues to grow, then TSMC is one of the most certain beneficiaries in the entire AI supply chain.
  Compared to storage prices being subject to supply and demand fluctuations, TSMC's business model is more stable and easier to navigate through economic cycles.

IV. Will US Support for Intel Threaten TSMC?
This is also a concern for many investors.
Xiaofeng believes:
The short-term answer is no.
Currently, in the global most advanced process market, TSMC still holds a de facto monopoly status, with Samsung and Intel playing more of a second-supplier role.
However, in the long run, the US push for manufacturing reshoring and support for Intel may indeed gradually erode some of TSMC's advantages.
Especially in:

• US domestic capacity
• Advanced packaging
• Government subsidies
• AI supply chain security
  In these areas, Intel may still gain more market share in the future.
  Therefore, his strategy is not to deny TSMC, but to continuously track market changes, even allocating a small amount of Intel as a long-term risk hedge.

V. Has AI Demand Really Started to Cool Down?
Xiaofeng believes that what is cooling down now is not AI demand, but market sentiment.
AI computing power demand continues to grow. The real uncertainty lies in whether the future speed of supply expansion will exceed demand growth.
He estimates:

• In 2026-2027, HBM supply will remain tight.
• Starting in 2028, the effects of expanded production will gradually emerge.
  What truly affects stock prices is not just increased supply, but:
  Whether AI demand can continue to grow at a high speed simultaneously.
  If AI deployment remains rapid, new demand can absorb new supply, delaying the impact of cycles; conversely, if demand slows down, the market may return to a cycle of oversupply.

VI. The Biggest Risk of Leveraged ETFs Is Not Direction, But Time
Regarding many investors holding two-times leveraged ETFs on SK Hynix, Xiaofeng admits he has also suffered significant drawdowns.
He reminds us that the biggest problem with leveraged products is not getting the direction wrong, but that the daily rebalancing mechanism causes capital erosion.
Even if you are bullish on the industry in the long term, if volatility is too high, you may permanently lose part of your capital due to the leverage mechanism.
Therefore, he believes that controlling position size is currently more important than continuing to increase leverage.

Conclusion: In the Second Half of AI, It's No Longer Just About GPUs, But Who Has the Deepest Moat
In this AI race, the market has begun to shift from "who can produce more GPUs" to "who can control the hardest-to-replace infrastructure."
HBM remains one of the most important components of AI, but it is still affected by industry cycles.
In comparison, TSMC's true advantage lies not only in advanced processes but also in long-term customer binding, deep customization capabilities, and an ecosystem that is difficult to replicate.
Therefore, the true competition in AI is not just a computing power race, but who can build a moat that transcends cycles. This is the core proposition worth 持续关注 (continuously following) in future AI investments.