HBM4 Shortage Worsens, Focus on AI Semiconductors

This was due to the US services and manufacturing PMIs recording 58.7 and 57.0 respectively in September, highlighting concerns about re-accelerating inflation and further interest rate hikes rather than an economic slowdown. Rising international oil prices also acted as a factor intensifying interest rate hike pressure.

As discussions on a phased ceasefire between the US and Iran led to a reduction in the rise of oil prices and interest rates, the inflow of call option purchases for QQQ, AMD, and Micron acted as leverage bets on tech stocks, supporting their lower end.

This week, overseas retail investors net purchased 118.3 billion won of Amazon, indicating expectations for cloud, data center, and AI revenue. Net purchases of Tesla and Nvidia also continued, confirming the AI investment cycle.

Funds flowed into high-volatility themes such as quantum computing (IONQ), and investors pursued a strategy of diversifying their portfolios by securing cash assets (SGOV) and making aggressive ETF investments (SOXL, etc.).

HBM3 requires 3 times and HBM4 requires 5 times the wafers compared to general DRAM.

The complexity of HBM stacking and packaging processes consumes production resources, creating a structural problem where increased HBM supply leads to a shortage of general DRAM.

This acts as a factor causing finished product manufacturers to downgrade specifications due to rising memory prices.

As HBM4 becomes more wafer-intensive, it could structurally prolong memory supply constraints, indicating an entry into an era where yield, rather than factory expansion, determines supply volume.

When HBM is stacked 16 layers or more, the difficulty of the bonding process escalates sharply, and even assuming a 90% yield per layer, stacking 16 layers causes the final yield to plummet to about 20%.

The problem pointed out is that even if high yields are achieved at individual stages, stacking them up to 16 or 32 layers causes the overall yield to drop sharply. Hybrid bonding is presented as a solution, but its process maturity is still insufficient.

With ASP playing a crucial role in defending revenue and profit, unlike in the past where prices sharply dropped due to capacity expansion, a strong trend is expected to continue until 2027 after peaking in 2026.

Rather than a sharp decline in prices after the peak, as in past memory cycles, there is a high possibility that relatively strong price trends will be maintained even in 2027. Maintaining ASP (Average Selling Price) is expected to play a decisive role in defending revenue and profit.

Around $5 billion in passive funds is expected to flow in related to this in September.

This is due to its new inclusion in SOXX with a 3.32% weighting, placing it around 13th.

Despite being a newly listed stock, the fact that its inclusion was accelerated by about a year compared to the usual timing is expected to have a positive impact on supply and demand.

The US-China summit resulted in a bland outcome, postponing conflict for two months rather than a major agreement on improving relations. In fact, it ended up being a two-month temporary truce, with the risk of trade war resumption deferred to year-end/early next year.

The US-China tariff truce was extended until January 10, 2027, but key issues such as export controls on rare earths and advanced semiconductors remained unresolved.

Core issues such as semiconductor export regulations, rare earth problems, and the Taiwan issue remained unresolved, and the uncertainty in the Korean stock market due to ongoing US-China tensions was not alleviated, with the market showing little reaction.

This week, the release of the US August PCE price index on September 30 is a major event expected to influence the upper bound of interest rates.

On the same day, Micron's earnings announcement is expected to be a barometer for the AI server memory industry's condition.

The US September employment report, to be released on October 2, will be a crucial indicator for confirming whether an economic recession is underway, acting as a three-stage event that will determine market direction.

Open interest also increased during the same period, indicating a significant influx of new futures purchases, which is expected to strengthen support when cumulative positions are adjusted.

Experts advise that a strategy of using HBM and front-end process related ETFs is advantageous, considering the volatility of individual stocks.

Investment stability in HBM and front-end/core process semiconductor ETFs was emphasized over individual stock investment.

As large-cap AI semiconductor stocks and leveraged ETFs rebounded, recovering losses or reaching breakeven levels, retail investors showed a tendency to sell.

This reflects retail investors' desire to offload holdings even with small profits, implying a potential resistance level during market rallies.

Foreign investors reduced their call option selling positions last week and switched to an out-of-the-money call option buying strategy. This is interpreted as an aggressive strengthening of positions aimed at profiting from market rallies.

This suggests that foreign investors' bets on the semiconductor sector are becoming more pronounced, coinciding with the AI growth cycle.