The Underrated "New Narrative" of Nvidia: "Self-Developed Open Source Models" and "Supply Chain Lockdown"

NVIDIA is quietly building a moat along two fronts that are rarely fully priced in by the market. According to the latest research report released by HSBC on August 20, HSBC analyst Frank Lee believes that beyond the ongoing "better-than-expected performance," new catalysts driving NVIDIA's stock price revaluation will come from two previously overlooked narrative lines. First, there is a strong bet on open-source small language models (SLMs), which have the potential to expand the customer base from a handful of mega-scale cloud vendors to millions of developers and sovereign nations. Second, by securing supply chain capacities through a series of multi-year procurement agreements, this advantage will become increasingly pronounced as competitors find it more challenging to secure key manufacturing resources. The market significance of these two new narrative lines is that they collectively point to a core proposition: NVIDIA's growth engine is evolving from a single structure dependent on a few mega-scale customers to a broader and more resilient customer ecosystem. If this transformation is recognized by the market, it will be a key variable driving the revaluation of NVIDIAs valuation. Open-source model offensive: from "selling shovels" to "making shovels" The HSBC report noted that after the decline of the autonomy AI and emerging cloud vendor (neocloud) craze, NVIDIA has yet to establish a new narrative strong enough to significantly drive stock price revaluation, which is one of the key reasons for its underperformance compared to the Philadelphia Semiconductor Index since the beginning of the year. The strategic focus on open-source AI is filling this gap. NVIDIA is currently actively positioning itself in the open-source model arena and branding itself as the world's largest contributor to open-source AI. According to NVIDIAs disclosures, the cumulative scale of open-source models has become the second most popular category in terms of token generation volume. HSBC believes that the strategic significance of this positioning lies in the fact that open-source small language models (SLMs) are becoming the preferred inference engine for intelligent agents and on-device applications. The preference for SLMs is backed by threefold logic: 1. Latency and throughput advantagesautonomous intelligent agents need to operate at high frequency in cycles of interpreting intent, calling APIs, and evaluating outcomes. Each micro-step that invokes large frontier language models (LLMs) will produce severe latency bottlenecks, whereas SLMs have significant advantages in inference cost and throughput. 2. Task-specific intelligenceSLMs excel at constrained, deterministic tasks, and enterprises are embedding specific domain SLMs into software platforms to address complex business problems. 3. Edge deployment capabilityModels with fewer than 10 billion parameters can easily fit into local GPU memory or edge devices, supporting localized operation. NVIDIA's open-source product matrix is already quite comprehensive, covering multiple core scenarios: - The Nemotron series focuses on inference and language tasks; - Cosmos is aimed at "physical AI" in collaboration with Siasun Robot & Automation and the vision field; - GR00T N1 is positioned as the worlds first open universal foundational model for humanoid Siasun Robot & Automation; - Alpamayo focuses on autonomous driving scenarios; - NVIDIA Agent Toolkit and NeMo are used respectively for building and customizing enterprise-grade AI agents and generative AI applications. HSBC pointed out that this free, highly optimized open-source ecosystem is essentially NVIDIAs strategic leverage to guide developers to prioritize running AI applications on its hardware. Supply chain positioning: securing capacity in advance to build competitive barriers HSBCs report noted that the demand for AI computing power continues to exceed supply under the constraints of capacity, and NVIDIA has systematically locked in key supply chain capacities in advanced packaging, memory, optical components, and energy infrastructure through a series of multi-year agreements by 2026. HSBC predicts that supply chain constraints in several sectors may further intensify by 2027, and NVIDIA's advanced procurement strategy will yield competitive value far exceeding that of its peers, potentially attracting a higher market premium. According to The Information, in advanced packaging and memory, NVIDIA signed a multi-year agreement worth $1.5 billion with Amkor Technology in July 2026 to support its expansion of advanced semiconductor packaging and testing capacity in Arizona, USA. In the same month, it reached a comprehensive cooperation agreement potentially worth up to $500 billion with SK Group, encompassing joint development with SK Hynix on next-generation AI memory (including HBM) and planning to build a 2-gigawatt Vera Rubin AI factory in Korea. In terms of foundry capacity, NVIDIA has booked 63% (for 2026) and 52% (for 2027) of TSMCs CoWoS-L advanced packaging capacity, forcing GPU and ASIC competitors to turn to other suppliers for alternatives, which come with potential yield risks. Regarding optical interconnects, as AI network infrastructure transitions from copper cabling to optical interconnects, NVIDIA has signed multi-year strategic agreements with Lumentum and Coherent, each investing $2 billion to support R&D and build domestic manufacturing capacity in the United States while gaining future access to advanced laser component capacity. Simultaneously, it signed multi-year commercial and technical cooperation agreements with Corning to expand the U.S. domestic manufacturing scale of advanced optical connection solutions. In terms of energy and land, NVIDIA is securing critical assets by directly investing in infrastructure developers. According to Bloomberg, NVIDIA has successively invested in Cloverleaf Infrastructure, Lancium, and SB Energy to bind power resources to ensure future availability for its chips, thereby embedding NVIDIA's complete hardware and software stack in the early design phases of relevant facilities. Among these efforts, NVIDIA announced that through its collaboration with SB Energy and OpenAI, it has locked land, power, and construction capacity at the PORTS-Pike Technology Park in Ohio, with the initial design supporting 4.25 IT-GW AI factory capacity, for which NVIDIA's cumulative payment obligation is capped at $105 billion, in addition to a $1.5 billion investment in SB Energy. Furthermore, NVIDIA plans to invest $1 billion in NAVER to expand the "GAK Sejong" AI factory from 55 megawatts to 200 megawatts by 2028, with a long-term plan to reach 1 gigawatt of sovereign AI infrastructure. From the competitive landscape perspective, such investments are essentially a bundling strategy. Cloud computing giants and AI labs typically mix and match multiple suppliers when purchasing chips, network equipment, and custom cables. By holding stakes in infrastructure developers, NVIDIA has secured a significant leverage to ensure future facilities are designed around its complete technology stack. This article is reproduced from the "Wall Street Journal" APP, authored by Bao Yilong; GMTEight editor: Song Zhiying.