mtn drew

Why did Samsung SELLOFF despite record earnings? Samsung just reported its third quarter earnings and I think the biggest takeaway here is that 2027 could be an even bigger year for the memory industry than 2026. The company generated approximately $146 billion in revenue and $80 billion in operating profit in just three months. Revenue increased 14% quarter over quarter, while operating profit jumped 20%, bringing operating margins to an incredible 55%. To put that into perspective, Samsung generated just 12.2 trillion won in operating profit during the same quarter last year, compared with 107.4 trillion won today. And almost all of that came from semiconductors. Samsung's semiconductor division generated approximately 147.1 trillion won in revenue and 106.5 trillion won in operating profit, largely driven by higher memory prices and demand for AI infrastructure. However, despite these record numbers, Samsung actually missed expectations. Revenue came in 3% below consensus and 7% below UBS estimates, while operating profit missed UBS estimates by approximately 9%. Net income also came in 14% below consensus, partly because of 9.5 trillion won in net finance expenses. But I think investors had VERY high expectations going into this and are focusing too much on the earnings miss and not enough on what Samsung said about 2027. One of the biggest takeaways was just how much memory prices are increasing. Samsung disclosed that DRAM average selling prices increased approximately 20% quarter over quarter, while NAND prices increased by a high teens percentage. Meanwhile, shipment volumes increased by only single digit percentages. So most of the growth wasn't coming from Samsung selling significantly more memory but rather coming from charging significantly higher prices. And this is happening across several different memory categories simultaneously. We've been talking a lot about HBM lately, especially following SK Hynix's earnings yesterday, but conventional DRAM and NAND are becoming increasingly important parts of the AI infrastructure buildout. AI data centers need HBM for accelerators, conventional DRAM for CPUs and servers, and NAND for storage. Samsung is benefiting from all three. Now, here's where things get REALLY interesting. Samsung reported that HBM4 sales increased more than threefold quarter over quarter and HBM4 now accounts for approximately 60% of its total HBM sales. But the most important comment came from Jaejune Kim, Samsung's memory business executive, who confirmed that the company's entire planned HBM supply for 2027 is already SOLD OUT, with pricing negotiations concluded at its target levels. And despite that, customers are still requesting additional supply. Also, earlier Korean media reports suggested Samsung was quoting HBM4 prices in the mid to high $4 range per gigabit, compared with approximately $1.50 per gigabit for HBM3E. That's potentially more than THREE TIMES the pricing of the previous generation. Now, Samsung hasn't confirmed the exact final contract prices, but management saying negotiations concluded at its target levels suggests the company has secured favorable pricing for next year. And remember, there were earlier rumors that Nvidia was pushing back against Samsung's proposed HBM4 pricing. So there's speculation that Nvidia may have ultimately accepted Samsung's asking price, although that hasn't been officially confirmed. If those reported prices are accurate, Samsung could generate substantially more revenue from its HBM4 business in 2027, even without a proportional increase in production volumes. There's also speculation that Nvidia could offer separate Rubin configurations using Samsung's HBM4 at a premium compared with configurations using SK Hynix or Micron memory. That's not confirmed, but it's something worth following. Samsung is also preparing to begin mass production of HBM4E in 2027. And I think this is particularly important because Samsung doesn't necessarily need to take significant market share away from SK Hynix to benefit. If demand continues growing faster than supply, both companies can increase sales and maintain stronger pricing. Another part of the earnings call that I found interesting was Samsung's discussion of AI agents and how they're creating additional demand for NAND storage. We've already seen how AI training created enormous demand for HBM but AI inference and agentic workloads are creating entirely different memory requirements. Samsung specifically mentioned KV cache, which is essentially information an AI model stores from previous tokens so it doesn't have to repeatedly process the same information. As AI agents become more advanced, handle longer conversations, and perform tasks over extended periods, the amount of information that needs to be stored and retrieved increases significantly. Samsung believes this is creating additional demand for high performance TLC SSDs. But the company also mentioned that cloud based personal AI agent services are increasingly moving toward allocating dedicated virtual machines to individual users. Essentially, individual AI agents could have their own dedicated computing environments and storage rather than sharing the same resources. And that creates demand for high density QLC SSDs, which provide larger amounts of storage at a lower cost per bit. Samsung is already preparing a 256TB enterprise SSD for commercial launch in early 2027. And all of this is happening while supply remains extremely constrained. Samsung expects the memory shortage to continue through 2028, even as manufacturers significantly increase capital spending. Samsung has also signed multiyear memory supply agreements with 10 major customers and is negotiating similar agreements with another 10. These agreements aren't limited to hyperscalers either. Samsung is seeing demand from automotive, mobile, and networking customers. The company ultimately wants approximately 60% to 70% of its production capacity covered by multiyear agreements by 2030. That would give Samsung much better visibility into future demand and make it easier to plan production and capital expenditures. Speaking of capital expenditures, Samsung expects to spend approximately 84.4 trillion won in 2026, an increase of 31.7 trillion won from last year. Of that amount, approximately 76.8 trillion won will go toward semiconductors, representing roughly a 62% increase from 2025. And management expects semiconductor capital spending to increase again in 2027. This is also important for companies like Lam Research, KLA , ASML, which supply the equipment needed to expand semiconductor manufacturing. Samsung also provided some updates on its foundry business. The company expects its 2nm project wins to more than double in 2026 compared with last year, with some of those projects coming from major cloud providers and AI customers. Its Taylor, Texas facility is also expected to begin production in 2027. But I wouldn't focus too much on foundry just yet. The business remains loss making, and competing with TSMC isn't going to be easy. Memory is still the main reason Samsung's earnings are growing so quickly. Now, another important announcement was Samsung's shareholder return program. The company expects to return approximately 90 trillion to 110 trillion won to shareholders for 2026 through dividends and share repurchases. Samsung announced a 30 trillion won cash dividend for the third quarter, equivalent to 4,601 won per common and preferred share. And with approximately 35 trillion won already planned or completed, the company expects another 55 trillion to 75 trillion won to be returned to shareholders. More importantly, management confirmed that a significant portion of those remaining funds will go toward share repurchases and cancellations. The cancellation part is important because Samsung will permanently retire those shares, reducing the total number of shares outstanding. Samsung also acknowledged the valuation gap between common and preferred shares and said it would consider adjusting the proportion of preferred shares included in its buyback program. However, the company won't finalize the exact amounts and execution plans until its board meeting in late January 2027. And that brings us to why Samsung's stock actually sold off following the earnings announcement. Part of the reaction was the earnings miss, but investors were also disappointed that Samsung didn't immediately announce more specific buyback detail. I actually viewed the shareholder return guidance positively because the additional 55 trillion to 75 trillion won was above investor expectations, which ranged from no additional returns to approximately 50 trillion won in the fourth quarter. So the problem wasn't necessarily the size of the shareholder returns. It was the lack of clarity around when and how the remaining funds would be distributed. The other issue was Samsung's outlook compared with SK Hynix. SK Hynix said yesterday that its DRAM and HBM supply was already sold out through 2029 and that memory shortages could potentially continue into 2030. Samsung, meanwhile, expects supply to remain constrained through 2028. That's still extremely bullish, but investors had just heard SK Hynix provide an even more aggressive outlook. And with expectations already so high, even slightly more conservative commentary can disappoint.  There's also an interesting theory going around that Samsung might be holding back some positive announcements because of its controlling family. Long story short, Samsung's chairman is set to receive approximately $1.9 billion worth of shares from his mother, and the gift tax bill is calculated using the average share price over a four month period ending in January 2027. The lower the stock price during that period, the lower the potential tax bill. And what's interesting is that Samsung has delayed announcing the full details of its massive share buyback program until late January, right after that valuation period ends. This makes a lot of sense to me. Now, there are still some risks worth monitoring. Samsung's semiconductor business is benefiting from higher memory prices, but those same prices are putting pressure on its smartphone and consumer electronics businesses. The company's Device eXperience division reported an operating loss of 0.8 trillion won during the quarter, partly reflecting higher component costs. And eventually, higher memory prices could begin affecting demand for smartphones, PCs, and other electronics. But when I look at Samsung's earnings alongside SK Hynix's results yesterday, I think the broader takeaway is pretty clear, this is a 2027 story, not a 2026 story. Samsung is already generating extraordinary profits, but its HBM4 ramp is accelerating, its planned 2027 HBM supply is already sold out at negotiated target prices, and the company expects memory demand to exceed available supply by an even greater margin next year. And that's before considering additional demand from AI agents, inference, and enterprise storage. I am happy as a shareholder, if you are a long term investor, ignore the noise and you will very happy over the next 2 years.