Astra's emergence ignites the AGI frenzy, while RSI reshapes the AI training paradigm. NVIDIA Corporation (NVDA.US) continues its growth myth.

NVIDIA Corporation (NVDA.US) stock is experiencing a new wave of strong upward momentum, fueled by robust support from earnings growth, breakthroughs in model capabilities, next-generation training clusters, control over the developer ecosystem, and long-term computing power orders. Recently released earnings show that NVIDIA Corporation's revenue for the second quarter of fiscal year 2027 reached $96.2 billion, a year-on-year increase of 106%; data center revenue soared to $89 billion, up 117% year-on-year; adjusted earnings per share stood at $2.22, a year-on-year increase of 120%. The revenue guidance for the third quarter is set at $108 billion, with a variance of 2%, which suggests around a 12.3% growth from the strong base in the second quarter. Additionally, management has explicitly defined the revenue growth outlook for fiscal year 2028 as constrained by supply. The recently launched Astra large model by OpenAI continues to actively expand the professional tasks that AI can undertake. NVIDIA Corporation CEO Jensen Huang made a significant statement on social media, suggesting that the emergence of GPT-6 Astra means "AGI has arrived." The strong range of revenue confirmed by NVIDIA Corporation, along with subsequent strong shipment guidance, combined with the ongoing development of AI large models entering a new phase of "recursive self-improvement (RSI)," opens a new curve of surging demand for AI computing power. This indicates that Astra is expected to expand AI computing power needs in commercial applications, and the trajectory of AI beginning to "create AI" could lead to increased investment in cutting-edge operator experiments, evaluations, and long-term sustained training, collectively extending the cycle of computing power investment. All of this highlights an almost limitless demand for AI computing power that is transforming into enormous hardware procurement orders centered around NVIDIA Corporation's AI GPU computing clusters. The assessment of AI economics by enterprises is also shifting from price per million tokens to the total cost of successfully completed tasks. Although the Gemini 3.8 Flash maintains a price of $0.75 per million input tokens and $3.75 per output token, an approximate 30% increase in output tokens and higher agent invocation rounds have actually raised the cost of each benchmark task by about 40% compared to the previous generation. Alphabet Inc. Class C parent company Alphabet has also clarified that the new models will improve task completion quality through more reasoning steps and repeated tool calls, which may consume more tokens. The inference can be drawn that a decrease in unit price and an increase in the number of applications could occur simultaneously, driving increases in overall AI inference demand and total inference value, which is known as the "Jevons Paradox." A forecast report from the well-known market research organization TrendForce shows that the shipment volume of NVL72 racks is expected to grow by over 50% year-on-year by 2027, with the total output value of NVL rack systems surrounding NVIDIA Corporation's AI GPUsincluding GB200/GB300 and next-generation systems based on the Vera Rubin architectureexpected to exceed $710 billion, an increase of 214% year-on-year, factoring in product upgrades and price hikes. Wall Street strategists have recently significantly raised their target prices for NVIDIA Corporation, with the fundamental 12-month target price being increased from $265 to above $300, a rise of about 13.2%. The core of this increase lies in raising normalized earnings per share assumptions from the previous $9.80 to a consensus expectation of $12 or higher while lowering the price-to-earnings ratio from 27 times to 25 times, reflecting that upward revisions in earnings forecasts are driving the increase in target prices. The acquisition of Hugging Face also bolsters NVIDIA Corporation's developer ecosystem as a long-term catalyst. Financial giants on Wall Street, such as Citigroup, Goldman Sachs Group, Inc., and Morgan Stanley, are optimistic about the overall demand for NVIDIA Corporation's next-generation computing architecture Vera Rubin, the growth visibility of GPU clusters in 2027 and beyond, and the expansion of AI infrastructure. Their consensus bullish judgment is based on the sustained strength of AI computing power demand, the ramp-up of Vera Rubin, and the ongoing strengthening of NVIDIA Corporation's integrated software and hardware platform advantages. Citigroup raised the target price from $300 to $315; Morgan Stanley increased its price from $288 to $300 while maintaining an "overweight" rating; Goldman Sachs Group, Inc. lifted its target from $285 to $300; and Bank of America Corp maintains a "buy" rating, preferred global stock status, and a target price of $350. New catalysts for a price surge are emerging! Earnings expectations are rising, and the developer ecosystem is opening new spaces. NVIDIA Corporation's stock price currently corresponds to a price-to-earnings ratio of less than 15 times on long-term earnings forecasts. With the halfway mark of fiscal year 2027 reached, CEO Jensen Huang and his team raised the performance forecast for fiscal year 2028 after announcing second-quarter results at the end of last month. Previously, the market had worried about a potential growth slowdown, but the company now expects to achieve growth rates of up to 70%. Despite increasing political noise, AI infrastructure construction seems to be progressing at full speed. Looking back over the past five years, NVIDIA Corporation has also performed outstandingly among global stocks. Since the U.S. Labor Day in 2021, the stock's cumulative return has reached 908%, surpassing the S&P 500 index's total return rate of 82% and far exceeding the second-best performer among the "Magnificent Seven" tech giants, Alphabet Inc. Class C parent company Alphabet (GOOGL.US), which has seen a 135% increase. On the trading day following the earnings report, the stock jumped 8.7%, marking the most robust response to earnings since May 2024. Just before the earnings call began, the stock actually dipped slightly; however, when the company revealed longer-term performance guidance, NVIDIA Corporation's stock quickly soared. In contrast, the volatility priced into the options market for the day of earnings disclosure was 5.6%, and in the previous eight earnings releases, NVIDIA Corporation's actual stock price movements had not exceeded the ranges implied by straddle options, with the last four times seeing significant declines post-earnings disclosure. In August, NVIDIA Corporation reported strong quarterly results. Data center revenue spiked, and the outlook for fiscal year 2028 is extremely optimistic, with the recent acquisition of Hugging Face attracting new attention. Acquiring Hugging Face undoubtedly enhances the long-term catalyst of its developer ecosystem: on September 3, NVIDIA Corporation announced that it had reached an acquisition agreement worth approximately $13 billion, aimed at expanding the coverage of its model development, distribution, and deployment platforms while committing to continue support for various cloud platforms and computing hardware. The investment's value lies in enabling more enterprises, research institutions, and developers to deploy AI, thereby expanding the potential market for NVIDIA Corporation's AI GPU computing power. Wall Street analysts are focusing on NVIDIA Corporation's gross margin metrics, and the 75.0% final result did not trigger market anxiety. Furthermore, with the support of $21.3 billion in free cash flow, NVIDIA Corporation returned a record $26 billion to shareholders through stock buybacks and dividends during the second quarter, while increasing U.S. GAAP operating expenses to $8.4 billion to support strategic expansion and R&D. Once the transaction is finalized, it could become a key "Trojan Horse" strategic move, as Hugging Face hosts over 3 million open-source weights or fully open-source AI models, acting as a model distribution layer and driving wider computing infrastructure demand beyond major hyperscale cloud providers. Similar to Microsoft Corporation (MSFT.US) completing its acquisition of GitHub in 2018, NVIDIA Corporation's acquisition of Hugging Face is expected to effectively integrate the company's architecture into developer workflows and a broader open-source ecosystem. Of course, we must be aware of potential regulatory hurdles, but as of now, this acquisition seems to be progressing smoothly. Another risk is that Hugging Face will maintain openness and interoperability, allowing developers to freely choose models, frameworks, cloud services, inference service providers, and computing platforms. Regarding earnings prospects, Wall Street analysts unanimously expect NVIDIA Corporation's earnings per share for fiscal year 2027 to nearly double, with only a mild slowdown in growth the following year. Analysts predict that by fiscal year 2029, the company's adjusted earnings per share based on operating profit could reach $20 to $21 or even higher. Unsurprisingly, in the past 90 days, Wall Street's sell-side analysts have raised earnings per share forecasts 40 times while only reducing them once. The earnings performance has indeed been impressive, with free cash flow per share reaching as high as $4.70 over the past 12 months. In terms of valuation, Wall Street analysts have updated evaluations for the second quarter of 2026. Assuming a normalized operating earnings per share of $12 over the next 12 months and a P/E ratio of 25, the stock price should be around $300. This is a notable increase from the previous $265, and this figure comes even after a two-point reduction in the P/E multiple. Wall Street's most optimistic target price is as high as an astounding $515, suggesting that NVIDIA Corporation's market capitalization could exceed $13 trillion. Looking further ahead, NVIDIA Corporation currently has a P/E ratio of only 11.2 times corresponding to its earnings per share in fiscal year 2029, indicating a very low valuation. Its PEG ratio is an attractive 0.5 times, while its price-to-sales ratio is 32% lower than the stock's long-term average. The primary risks for NVIDIA Corporation include its large size, with a market cap currently approaching $6 trillion rather than $5 trillion. Any setbacks in AI data center construction could lead to a decline in the P/E ratio; given the proximity of the midterm elections, the rising public opposition to AI and local data center construction could make such a scenario plausible. From a macro perspective, AI infrastructure construction is fundamentally a cycle of capital expenditure, and any slowdown in spending by hyperscale cloud providers, emerging cloud providers, enterprises, or sovereign investments could simultaneously suppress revenue growth and the valuation multiples that investors are willing to pay. An even more imminent risk may arise from intensifying competition in the chip development field, with top chip companies like Broadcom Inc. (AVGO.US) and Marvell Technology, Inc. (MRVL.US) making advancements in customizable semiconductors (i.e., AI ASIC/TPU/XPU). China remains an uncertain factor, although some Wall Street analysts predict that even NVIDIA Corporation's aforementioned guidance does not hold overly optimistic expectations for demand from the world's second-largest economy. Internally, NVIDIA Corporation also faces significant customer concentration: according to quarterly filings, one direct customer accounted for 16% of second-quarter revenue; three direct customers contributed 16%, 15%, and 13% of first-half revenue, respectively. Additionally, the company's newer AI cloud business arrangements have added another layer of execution risk, with an outstanding commitment amount of $36 billion as of July 26. Looking ahead, company event data provided by Wall Street Horizon shows that NVIDIA Corporation has confirmed it will report its third-quarter results for fiscal year 2027 after the market closes on Tuesday, November 17. Prior to this, Jensen Huang and his team plan to attend a series of industry and AI conferences, starting with one of the largest corporate events of the yearGoldman Sachs Group, Inc.s 2026 Communacopia Tech Conference held this week. Wall Street analysts will also focus on NVIDIA Corporation's AI Day event in Singapore scheduled for September 22 later this month. From a technical analysis perspective, the stock's long-term 200-day moving average is on an upward trend, indicating that bulls dominate the major trend. In addition, just before the upcoming long weekend, NVIDIA Corporation quietly set a new historical high for weekly closing prices, which typically reflects underlying technical strength. After a period of poor performance over the summer, the short-selling pressures may have been cleared, paving the way for Wall Street's previously set technical target price of $265 to be actively realized before the end of the year. In terms of relative strength index (RSI) momentum indicators, the stock is currently operating within a relatively neutral range of 35 to 65. Investors hope to see improvements in RSI, with the stock price closing above the previously set intraday historical high of $236.54 established in May, which would thoroughly revive bullish sentiment on the technical front. Finally, due to a considerable historical trading volume existing below the current price level, there should still be ample buying support when the stock price experiences a pullback. The launch of the Astra model and the participation of AI in the development of AI large models: cutting-edge large models and RSI pathways significantly extend investment cycles in computing power. OpenAI's recently launched GPT-6 Astra large model, along with the RSI technology path focused on by AI leaders, is expected to be one of the two core drivers of exponential expansion in AI computing power demand, namely, more powerful AI large models and wider use of AI application tools, with the next-generation AI training paths demanding even stronger computing power, enhancing the basis for continued growth in AI infrastructure demand. A research report released by Rich Pivovaritsky, head of Delta One trading at Goldman Sachs Group, Inc., highlights the trading desk's opinion that Astra could shift the entire demand curvewhen AI large models become smarter, businesses can attempt tasks that were previously unreliable, and competitors will also need to continue investing in R&D and training, providing new support for the AI expenditure cycle. The report notes that SoftBank's ADR surged over 10%, and Oracle Corporation rose about 5.5%, reflecting the market's repricing of the potential for Astra-led computing power expansion. In relation to NVIDIA Corporation, Astra and RSI strengthen demand for cutting-edge computing power, Hugging Face expands developer coverage, and Blackwell and Vera Rubin undertake product delivery, while profit growth supports valuation. Decreasing costs per token stimulate increases in invocation volume, reasoning depth, and agent cycles, with RSI opening a second curve of computing power demand for sustained training. Recursive Self-Improvement (RSI) is not equivalent to a model suddenly gaining self-awareness; its currently verifiable form allows AI to gradually participate in the planning of training strategies, algorithm generation, code writing, experiment design, result evaluation, error correction, and next rounds of training, as well as subsequent rounds of AI training. The trend of RSI brings an important direct inference: AI computing power advantages directly research advantages, and research advantages directly model advantages; those AI labs with sufficient GPUs can complete all validations in a day, while cutting-edge AI labs lacking GPUs may only accomplish 5 or fewer task validations per day, creating a significant speed disparity. RSI could create a "second demand curve" for computing power, independent of terminal user demand: training is no longer "trainreleaseend," but rather a continuous closed loop where the model proposes algorithms, runs hundreds to thousands of experiments in parallel, evaluates results, and initiates the next round of training; agents will also increase test-time compute through longer reasoning chains, repeated tool calls, and multi-model collaboration. If top labs like OpenAI and Anthropic redistribute computational resources from high-income inference operations to training, short-term revenue may plummet, but the overall computing power consumption will not decline; rather, it could increase as competition for leading models intensifies. With the significant launch of Astra and the entry of AI large model R&D led by OpenAI, Anthropic, and SSI into the new AI research phase of "recursive self-improvement (RSI)," along with the recent strong performance and outlook from NVIDIA Corporation and a series of contracts for AI computing power capacity at the megawatt and even gigawatt levels driving continuous positive expansion of computing resources demand, the existing fundamentals of the AI computing power industry do not support the pessimistic assumption that "the entire industry has systematically excess computing resources." Instead, they demonstrate that as prices for AI computing power decrease, demand is sky-rocketing. The core change brought by Astra is that advancements in model capabilities begin to expand the range of tasks that can be assigned to AI. Jensen Huang stated on September 6 that Astra training utilized over 100,000 NVIDIA Corporation GPUs based on the Grace Blackwell NVLink72 platform, and predicted that the next batch of 400,000 GPUs will come online. The former describes the training resources, while the latter pertains to the upcoming deployment of AI computing power resources. OpenAI's announcement regarding Astra showcases achievements including 98% on FrontierMath Level 4 and 99.9% on ARC-AGI-3, demonstrating its capabilities leap in specific high-difficulty tests. For the capital markets, this is expected to exponentially increase commercial space in professional software operations, scientific research, code development, and cybersecurity scenarios. From a systems engineering perspective, recursive self-improvement (RSI) will further tighten the connection between reasoning and training. AI agents propose candidate algorithms, write experimental codes, invoke training tasks, analyze results, and feed effective improvements back for the next round of research; each iteration consumes reasoning compute, experimental training compute, and data storage resources. A decrease in reasoning prices allows more experimental schemes to fall within cost-effective ranges, and the capability improvements expand the set of researchable problems, collectively supporting the growth of computing power demand. AI server storage chip components remain the clearest supply bottleneck on the AI computing power industry chain level. TrendForce predicts that in the third quarter of 2026, contract prices for traditional DRAM and NAND Flash will rise by 13%-18% and 10%-15%, respectively, compared to the previous high base; HBM and server RDIMM are expected to account for 51% of the global DRAM bit supply in 2026, with HBM contract prices anticipated to increase by 70%-140% in 2027. TrendForce's forecast data also indicates that contract prices for DRAM on the server side of AI data centers are expected to rise approximately 270% cumulatively by 2026, with enterprise SSD prices expected to rise about 235% cumulatively; HBM contract prices in 2027 may also increase by 70%-140%. The agency anticipates that capital expenditures among major cloud service providers will grow by 98% year-on-year in 2026, followed by another 50% growth in 2027, with the proportion of DRAM and NAND spending rising from 47% in 2026 to 68% in 2027, potentially increasing by a full 21 percentage points within a year. On Wall Street, the most optimistic target price of $515 comes from Simon Leopold, a senior analyst at Raymond James. This analyst raised the target price from $352 to $515 on August 27, maintaining a "strong buy" rating; this adjustment came before the September Astra release. The analyst's core bullish logic is that supply constraints are obscuring NVIDIA Corporation's potential revenue scale: storage, packaging, and manufacturing capacity limits shipments, with expansion expected to unlock unfulfilled demand; the ramp-up of Vera Rubin, along with the expansion of AI labs, enterprises, and professional cloud customers, will further broaden revenue sources. He forecasts that by January 2029, NVIDIA Corporation's annual revenue could approach $1 trillion, which is about one-third higher than the then consensus market estimate of under $750 billion. His $515 valuation assumes a projected earnings per share of approximately $23.41 for the calendar year 2028, with a 22-fold P/E ratio, with the central bet being sustained earnings growth.