The company's third-quarter revenue and operating profit exceeded Street and buyside expectations, driven by a high-20s% DRAM ASP lift, indicating strong pricing power.
The fourth-quarter DRAM bit guide, projecting a mid-20s% quarter-over-quarter increase, positions fourth-quarter revenue and operating profit 10-20% above pre-print Street estimates.
Long-term agreements (LTAs) now cover over 50% of 2027/2028 capacity across 14 customers, with DRAM/HBM effectively sold out through 2029, providing significant revenue visibility and stability.
The company's FY26 shareholder return plan of approximately KRW100T (US$75B), representing about 10% of market cap, surpassed buyside expectations, demonstrating a strong commitment to capital return.
The resolution of the HBM4 overhang, with yields beating plans, HBM4 crossing over in 4Q, NVIDIA relaxing the Rubin spec, and 2027 allocation locked, addresses a key bear argument and supports future performance.
The company's strong position as the HBM share leader with the highest AI revenue exposure among major memory players, coupled with industry-leading yields and first-supplier status at a key accelerator customer, makes it a pure play on AI infrastructure.
The long-term agreements (LTAs) are expected to transition the company from a spot-priced commodity to contracted infrastructure revenue, potentially leading to higher multiples over time.
The company's management views customers moving to lower HBM stacks as a symptom of shortage, not a demand negative, indicating sustained strong demand for HBM.
The broadening buyer base for AI infrastructure, including enterprise, sovereign, and on-prem customers, suggests that any pause from one buyer can be backfilled by others, ensuring sustained demand.
The company's commitment to quarterly buybacks and management's view that the shares are undervalued further support a bullish outlook.
The EPS miss was entirely due to non-operating factors, including a significant FX loss and investment valuation losses, which could lead to continued earnings volatility.
Price growth is decelerating, with DRAM ASP increases slowing from ~30% to high-20s% and NAND from mid-50s% to mid-20s%, suggesting a shift in the earnings engine from price to bits.
Customer economization, evidenced by stack-count cuts, a relaxed Rubin spec, and falling PC/smartphone units, indicates price elasticity and potential demand sensitivity.
The claim of being "sold out through 2029" partly relies on volumes still under discussion, and there is no capital return framework beyond FY26, raising questions about long-term visibility.
Despite buybacks, diluted shares actually rose in the third quarter as a July ADR raise outweighed share repurchases, and additional stock was bought for employee compensation outside the stated capital return program.
HBM's share of revenue has retreated as commodity DRAM outpaced total top line, implying that a larger portion of the company's earnings is dependent on commodity DRAM pricing rather than the "HBM moat."
The company's record gross margin, significantly above prior-cycle highs, and the tripling of DRAM pricing this year suggest a peak in the memory cycle, which historically precedes a decline in stock performance.
The increasing visibility of capacity response, including significant capex increases and new fab ramps, suggests that record margins will attract more supply, potentially leading to oversupply.
The "AI is different" argument requires AI capex to continue compounding off an enormous base, with memory becoming a substantial line item in the AI bill of materials, which could invite pushback and scrutiny.
The marginal buyer, particularly neoclouds, relies on increasingly expensive external funding, and any capex pause would first impact non-LTA, spot-priced bits, making the company vulnerable to funding risks.
Long-term agreements (LTAs) are perceived to protect volume more than price, as pricing floats within undisclosed ranges, and customers may push to renegotiate if spot prices fall significantly below contract levels.