Nvidia's latest financial report highlights the most challenging point in the AI computing power industry chain. In Nvidia's financial report yesterday, the median revenue guidance for Q1 was 4% higher than the most optimistic expectation of the buying institution, indicating that the guidance is stronger than the market expected, which means that the market expectation needs to be further raised. Goldman Sachs clearly stated in its latest report yesterday that, unlike some previous quarters, there are three major factors that will lead to a clearer path in the future: 1. There is still room for upward growth in the 2026 capital expenditure forecast of ultra large scale cloud vendors, and early signs of capital expenditure growth in 2027 have begun to emerge, indicating that the demand support for Nvidia's core downstream customer base will continue to extend to the far end. 2. Non traditional clients represented by OpenAI and Anthropic will significantly increase the visibility of their procurement plans until 2027 with the implementation of their respective financing rounds. Nvidia has disclosed that it is still actively negotiating with OpenAI on investment and cooperation matters, and is expected to complete the signing in the near future; At the same time, a $10 billion investment in Anthropic has been completed, and the agreement includes Anthropic training its large language model based on the Blackwell and Rubin architectures. 3. As the new generation of AI models trained on the Blackwell architecture are introduced to the market, Nvidia will once again demonstrate its technological leadership over its AI chip competitors in the coming months, providing more intuitive evidence of differentiated competition for the market. Each of these points points to the same word: capital expenditures, buyers are becoming more diverse and 'capital intensive', and big model companies themselves are becoming infrastructure buyers. During the earnings conference call, Nvidia admitted that the "cost pressure" came from HBM. This means that AI not only requires more computing chips, but also more, more expensive, and more difficult to scale storage. The Asia Pacific Technology Report released by Bank of America two days ago clearly stated that it is bullish on storage and Korean technology for three reasons: 1) Higher DRAM unit price/shipment amount (compared to their January forecast); 2) Stronger capital expenditures (primarily directed towards HBM); 3) The more visible AI demand theme - directly stated in the material: Amazon's FY26 target expenditure is $200bn. Institutions expect DRAM wafer production to only increase by 7% year-on-year by the end of 2026 (due to output growth from Samsung and Hynix), and supply cannot keep up with demand, pointing out that HBM/storage is the easiest link to form a "supply premium". Previous lengthy article on the war of capital expenditures: https://(x.com)/qinba frank/status/202436193082819302? As discussed in s=46&t=k6rimWs Ebo2D2TXolYcM-A, key links in the computing power industry chain can still capture the maximum value: chips, packaging and testing, storage, optical modules, etc. Those whose production capacity is not easy to expand rapidly and those with extremely high moats will enjoy the dividends of huge capital expenditures. Although the market has been debating whether capital can be sustained and whether capital expenditures have peaked? But in the foreseeable 26 years, the capital expenditures of big tech companies still need to be spent, which puts us on the side of 'not yet at the top', but requires close tracking of 'budget/order/guidance' evidence. We will continue to monitor whether the "cloud vendor capital expenditure revision" continues to occur as Goldman Sachs has stated, and whether the "2027 growth signal" has been announced earlier. If the capital expenditure on AI infrastructure (CapEx) continues to rise, and the resulting "supply bottleneck premium" falls on high bandwidth storage (HBM) and advanced packaging. Previously, I saw that the UBS report also included HBM's expected share in the customer dimension: 1) The supply structure provided by UBS for Nvidia servers is: Hynix 60%+, Micron 24%, Samsung 15%; 2) For Amazon's ASIC servers, we offer Hynix 84% and Samsung 16%; Different customer combinations vary, but the commonality is that HBM has a high degree of structural concentration. In fact, it mainly falls on SK Hynix and Samsung. This is also a clear signal that the weight of the Korean semiconductor industry is significantly increasing. When looking at the storage sector in December last year, I chose Micron because the US stock market did not have Hynix and Samsung. In further analysis and learning, I came across the Korean index ETF, which is actually a good target: if AI memory becomes the core variable in the next decade, it will not only benefit a single company, but the entire Korean technology sector. For investors who wish to participate in this round of structural opportunities but are unwilling to bear the risk of individual stock volatility, obtaining exposure through Korean index ETFs may be a more balanced approach. BlackRock's EWY and Franklin's FLKR both provide overall exposure to SK Hynix and Samsung. These targets can be traded on StableStock using stablecoins. https://app.stablestock.finance/trade/stock/ewy https://app.stablestock.finance/trade/stock/FLKR Of course, it's not about chasing the high. Last night, Nvidia's stock price sold the news on the first day after its financial report. Today, the semiconductor sector has experienced a pullback. From a personal perspective, this is not a peak, but a pullback after the sentiment is full. Waiting for the pullback to be in place is a good buying point.