Storage chip breaks through mid-summer consolidation! The AGI frenzy combined with a new training paradigm for RSI has Goldman Sachs sensing the onset of a new storage bull market.

Wall Street financial giant Goldman Sachs Group, Inc. recently released a research report stating that following the AI deleveraging storm in July, which saw aggressive sell-offs leading to a continued global decline in storage themes due to extreme crowding in bearish positions, the storage chip stocks and investment targets related to the frenzy of AI data center construction have begun to significantly break through the recent soft downtrend, moving towards a new round of bullish market trajectory. The core opportunity identified by Goldman Sachs Group, Inc. is the expansion of high-performance AI computing power demand brought about by the launch of OpenAIs Astra and the increasing prospects for storage chip demand as the recursive self-improvement (RSI) training paradigm starts to dominate AI training, intersecting with the low positions of traditional Wall Street asset management firms and hedge funds. Goldman Sachs Group, Inc. has recently compiled data indicating that the net leverage of long-short funds is at the 4th percentile of the past year, meaning bullish exposure is close to annual lows; the implied volatility of global semiconductor themes has dropped from 65 to 36, a decrease of approximately 44.6%, indicating that the previously abnormal risk pricing has eased significantly. If the strong profit trajectory driven by the upgrade iteration of large models and the new paradigm of AI training continues to provide positive catalysts, institutions re-increasing their risk exposure could significantly amplify the rally of storage-related stocks that have broken out of their consolidation. The storage chip components of AI data center server clusters remain the clearest supply bottleneck in the AI computing power supply chain. Market research firm TrendForce predicts that by 2026, the contract price of server DRAM will increase by approximately 270%, and enterprise SSD prices will rise by about 235%; by 2027, HBM contract prices may still rise by 70% to 140%, reflecting the combined impact of AI computing power expansion and rising storage prices. TrendForce's latest estimates show that the combined share of DRAM and NAND in the capital expenditure of major cloud service providers will rise from 47% in 2026 to 68% in 2027, indicating increased purchasing volume and rising prices. The Korean stock market has shown specific signals of expanding capital participation. On September 7, Samsung Electronics rose by 5.68%, SK Hynix increased by 8.26%, and the KOSPI index, known as the AI computing power barometer, surged by 4.61% to 6,995.39 points; foreign and institutional investors net bought approximately 25.5 trillion won and 26.4 trillion won respectively, with buying extending beyond corporate buybacks. Calculating from the July 30 figure of 5,593.56 points, the index has rebounded approximately 25.06%, placing it within a technical bull market range. On September 8, the KOSPI index retreated by 0.58% to 6,954.52 points, still up about 24.33% from that low point, indicating strong storage sentiment, although the overall market remains influenced by interest rate and energy risks. Another Wall Street financial giant, Nomura Securities, recently stated that the demand driven by AI training and inference expansion continues to grow, while supply expansion is constrained, and shortages are expected to persist until 2028. Moreover, Nomura emphasized that long-term supply agreements (LTA), which lock in volume, protect prices, and require prepayments, continue to enhance the predictability of future profits for storage chip giants. Nomura forecasts that approximately 50% to 70% of sales will be covered by long-term contracts, thus the market continues to price according to traditional cyclical stock patterns at about three times the expected price-to-earnings ratio for 2027, underestimating the changes in the business model. This resonates with Goldman Sachs Group, Inc.s emphasis on the opportunity to replenish low positions: Nomura values the long-term profit foundation supporting the replenishment rally. Based on the closing prices on September 8, 2026, which are approximately 269,500 won for Samsung Electronics and 1,793,000 won for SK Hynix, Wall Street financial giant Nomuras target price of 670,000 won for Samsung Electronics implies a potential upside of 150% over the next 12 months, while the target price of 4,700,000 won for SK Hynix indicates an approximate 160% potential upside. Storage theme stocks are emerging from a summer consolidation, and Goldman Sachs Group, Inc. has identified a breakthrough as a positive signal. Lee Koppersmith, Managing Director of Fixed Income, Currency, Commodities and Equities at Goldman Sachs Group, Inc., noted in the report that several critical investment targets in the storage industry chainMicron Technology, Inc. (MU.US), SanDisk (SNDK.US), iShares MSCI South Korea ETF (EWY.US), and Roundhill Storage ETF (DRAM.US)exhibit similar bullish technical patterns. The latest charts compiled by Goldman Sachs Group, Inc. show that these targets are beginning to exit summer consolidation technical indicators, although the related trends remain in the early to mid-stages. As these potential breakthroughs emerge, a broader AI computing power trading theme is also encountering a more favorable market environment: investors' bullish positions are lighter compared to the peak periods of stock prices, implied volatility has decreased, and a series of potential catalysts are set to emerge, including Goldman Sachs Group, Inc.s Communacopia+ technology conference. According to data from Goldman Sachs Group, Inc. in the prime brokerage business, the total leverage indicator for fundamental long-short hedge funds in the U.S. stock market is in only the 27th percentile of the past year, while net leverage is merely at the 4th percentile. This highlights that hedge fund institutional investors have begun to rebuild AI computing power positions, but the level of exposure remains significantly lower than it was about two months ago. The options market has also undergone a substantial adjustment. The implied volatility of semiconductors, as measured by the Cboe Global Markets Inc. semiconductor ETF volatility index (VXSMH), has dropped from about 65 in July to 36, nearly halving and falling close to levels seen in early 2026. Meanwhile, even though there was a significant rebound of nearly 7% last Friday, the basket of technology stock long-short momentum trading combinations compiled by Goldman Sachs Group, Inc. remains approximately 50% lower than the late June highs. Koppersmith stated that this round of adjustment has narrowed the potential performance range reflected in the pricing of AI computing power-related stocks but has not formed a clear bearish market trend. The Korean market may also provide further upward momentum for the computing power sentiment theme. Goldman Sachs Group, Inc. noted that over the past four weeks, the inflow of funds into the Korean stock market has been fully supported by corporate buybacks, while other investors are in a net selling state of funds overall. However, the magnitude of the sell-off has significantly narrowed. This opens up space for more investors to engage deeply and positively in the Korean stock market, particularly in the new round of bullish frenzy for the two largest storage chip stocks globallySK Hynix and Samsung Electronics. Astra ignites the AGI craze + AI begins to participate in AI R&D, leading to a dual demand expansion curve for storage chips. OpenAIs recently launched Astra large model continues to actively expand the professional tasks that AI can undertake. NVIDIA Corporations CEO Jensen Huang stated on social media that the advent of GPT-6 with Astra means AGI has arrived, and NVIDIA Corporation has confirmed strong revenue bands and subsequent strong shipment guidance. With the added layer that AI large model research is entering a new stage with recursive self-improvement (RSI), which opens up yet another curve for the surge in AI computing power demandAstra is expected to broaden AI computing power demand on the commercial application end, while AI beginning to create AI in R&D may increase investment in cutting-edge operator experiments, evaluations, and long-term continuous training, collectively extending the computing power investment cycle. The investment significance of Astra and recursive self-improvement (RSI) lies in the fact that cutting-edge high-performance AI large models, and the R&D of AI itself are becoming new scenarios that continuously consume computing power. OpenAI revealed on September 6 that it has achieved the goal of the automated research intern, capable of completing some tasks that would originally require skilled researchers days under human guidance. As of mid-August, every human working day corresponds to approximately 3.1 agent operational workdays. This measures runtime rather than increasing research outputs by 3.1 times. From this, it can be deduced that research automation will simultaneously increase the reasoning required for code generation, experimental evaluation, and candidate model training demands. However, complete RSI has yet to become an established dominant paradigm, with research directions and resource allocations still determined by humans. Astra represents the mechanism of cutting-edge performance demand expansion: enhanced large model capabilities bring previously difficult tasks into the realm of commercialization. Furthermore, Astra may shift the entire demand curve outwardwhen large AI models become smarter, enterprises can attempt tasks that were previously unreliable; competitors will also need to continue investing in R&D and training, providing new robust support for the AI expenditure cycle. OpenAI's release of the GPT-6 Astra large model and the focus of AI leaders on the RSI technical path are expected to become the two core drivers of exponential expansion in AI computing power demandnamely, stronger performance of AI large models and broader use of AI application tools, along with the next-generation AI training path that possesses even stronger computing power demand, which is strengthening the fundamental basis for sustained growth in AI computing power infrastructure. OpenAI disclosed that Astra achieved a score of 98% on the FrontierMath level 4 test and 99.9% on ARC-AGI-3. Huang expressed a judgment that "AGI has arrived," and mentioned that training models made use of over 100,000 NVIDIA Corporation GPUs, with an additional 400,000 GPUs expected to come online soon. It is noteworthy that the assertion of "AGI has arrived" remains contentious, while the forecast for the deployment of larger-scale NVIDIA Corporation AI GPU clusters directly reinforces the strong expectations for continued investment in training resources for cutting-edge AI large models. From a technological perspective, storage demand depends on parameter scale, context length, concurrency numbers, and experimental density. HBM carries the model weights, training intermediate states, and active key-value caches on the GPU side; server DRAM handles data processing, running environments, and cache offloading; NAND enterprise-level SSDs store datasets, training checkpoints, and reusable caches. A sample provided by NVIDIA Corporation shows that loading weights for Llama 3 70B at FP16 precision requires approximately 140 GB of memory, while the KV cache corresponding to a single users context of 128,000 tokens still requires about 40 GB. As more powerful models handle longer tasks and more agents run concurrently, along with RSI research processes increasing parallel experimentation and checkpoint saving, the demands for capacity, bandwidth, and read-write throughput will increase significantly. KB Securities from South Korea expects that the share of storage in AI infrastructure investment will rise from 14% in 2025 to 40% in 2026, and further to 57% in 2027. KB Securities core bullish logic for SK Hynix and Samsung focuses on the gap between extremely thin inventory buffers and low forward profit valuations. KB Securities states that Samsung Electronics and SK Hynix have storage inventory of less than 10 days and expect that AI infrastructure investment from ultra-large cloud vendors will reach $1.3 trillion by 2027, a year-on-year increase of about 60%. Based on the stock prices and earnings forecasts used in their report, the two companies are approximately 38% down from previous highs, corresponding to a price-to-earnings ratio of only about three times for 2027. Therefore, KB bets on demand expansion and price increases driving upward revisions for profit expectations, along with valuation recovery. For the Korean stock market benchmark index KOSPI, Goldman Sachs Group, Inc. has even set a target of 12,000 points. As of September 8, the KOSPI index fell 0.58% to 6,954.52 points; Goldman Sachs Group, Inc. highlights that the market systematically underestimates the duration of the AI-driven storage chip demand cycle and has significantly raised its forecast for the capital expenditure of large U.S. tech companies to $1.2 trillion for next year, believing that the "storage shortage" driven by data center expansion will further intensify by 2027.