Nvidia, beyond consistently delivering better-than-expected results, is quietly building moats along two fronts that the market has yet to fully price in. HSBC analyst Frank Lee wrote in a research note published on August 20 that the new catalysts for Nvidia’s stock re-rating will come from two previously overlooked narratives: an aggressive bet on open-source small language models (SLMs), and the pre-emptive locking of supply chain capacity through a series of multi-year procurement agreements.

The market significance of these two narratives lies in the fact that they both point to the same core proposition: Nvidia’s growth engine is evolving from a single structure dependent on a handful of hyperscale customers toward a broader, more resilient customer ecosystem. HSBC believes that if this shift gains market recognition, it will be a key variable driving a valuation re-rating.

The Open-Source Model Offensive: From “Shovel Seller” to “Shovel Maker”

HSBC’s note points out that after the sovereign AI and neocloud boom faded, Nvidia had struggled to form a new narrative strong enough to drive a meaningful stock re-rating — a key reason it has underperformed the Philadelphia Semiconductor Index year-to-date. The strategic bet on open-source AI is now filling that void.

Nvidia is actively positioning itself in the open-source model space, casting itself as the world’s largest open-source AI contributor. According to Nvidia’s disclosures, open-source models have cumulatively become the second most popular category by token generation volume. HSBC argues the strategic significance of this push is that open-source SLMs are becoming the inference engine of choice for intelligent agents and edge applications.

There are three underlying reasons behind the market’s growing preference for SLMs. First, latency and throughput advantages. Autonomous intelligent agents need to operate at high frequency across execution loops — parsing intent, calling APIs, evaluating results — and invoking large frontier language models (LLMs) at every micro-step creates severe latency bottlenecks. SLMs offer significant advantages in inference cost and throughput. Second, task-specific intelligence. SLMs excel at constrained, deterministic tasks, and enterprises are embedding domain-specific SLMs into software platforms to solve complex business problems. Third, edge deployment capability. Models with fewer than 10 billion parameters can easily fit into local GPU memory or edge devices, enabling on-device execution.

Nvidia’s open-source product portfolio is already quite comprehensive, covering multiple core scenarios: the Nemotron family targets reasoning and language tasks; Cosmos addresses “physical AI” for robotics and vision; GR00T N1 is positioned as the world’s first open general-purpose foundation model for humanoid robots; Alpamayo focuses on autonomous driving; and NVIDIA Agent Toolkit and NeMo are used respectively for building and customizing enterprise-grade AI agents and generative AI applications.

HSBC notes that this free, highly optimized open-source ecosystem is fundamentally a strategic lever for Nvidia to steer developers toward running AI applications preferentially on its hardware, with the potential to expand the addressable customer base from a handful of hyperscale cloud providers to millions of developers and sovereign nations.

Supply Chain Positioning: Locking Capacity Early to Build Competitive Barriers

HSBC’s note highlights that AI compute demand continues to outstrip supply amid capacity constraints, and Nvidia has systematically locked in critical supply capacity across advanced packaging, memory, optical components, and energy infrastructure through a series of multi-year agreements in 2026. HSBC expects supply chain constraints across multiple segments to intensify further in 2027, at which point Nvidia’s early procurement strategy will generate competitive value far exceeding that of its peers, and could command a higher market premium.

In advanced packaging and memory, Nvidia signed a multi-year agreement worth $1.5 billion (approximately NT$48 billion) with Amkor Technology in July 2026 to support the expansion of advanced semiconductor packaging and test capacity in Arizona. That same month, Nvidia reached a comprehensive partnership with SK Group valued at as much as $500 billion (approximately NT$15.9 trillion), covering joint development of next-generation AI memory (including HBM) with South Korea’s SK Hynix, as well as plans to build a 2-gigawatt Vera Rubin AI factory in South Korea.

On foundry capacity, Nvidia has pre-booked 63% of TSMC’s CoWoS-L advanced packaging capacity for 2026 and 52% for 2027, forcing GPU and ASIC competitors to seek alternatives from other suppliers — and turning to alternative suppliers carries potential yield risks.

In optical interconnect, as AI network infrastructure evolves from copper cabling to optical interconnect, Nvidia has signed separate multi-year strategic agreements with Lumentum and Coherent, each committing $2 billion (approximately NT$64 billion) to support R&D and U.S. domestic manufacturing capacity buildout, while securing future capacity access to advanced laser components. Meanwhile, Nvidia signed a multi-year commercial and technology collaboration agreement with Corning to expand U.S. domestic manufacturing scale for advanced optical connectivity solutions.

In energy and land, Nvidia is locking in critical assets through direct equity stakes in infrastructure developers. Nvidia has successively invested in Cloverleaf Infrastructure, Lancium, and SB Energy, tying up power resources to ensure its chips have “somewhere to run” in the future, while embedding Nvidia’s full hardware and software stack into the early design phases of these facilities.

Nvidia announced that through its partnership with SB Energy and OpenAI, it has secured land, power, and construction capacity at the PORTS-Pike technology park in Ohio, with initial design supporting 4.25 IT-GW of AI factory capacity. Nvidia’s cumulative payment obligations for this are capped at $105 billion (approximately NT$3.3 trillion), with an additional $1.5 billion equity investment in SB Energy. Nvidia also plans to invest $1 billion (approximately NT$32 billion) in South Korean internet giant NAVER to expand the “GAK Sejong” AI factory from 55 megawatts to 200 megawatts by 2028, with longer-term plans to scale toward 1 gigawatt of sovereign AI infrastructure.

From a competitive landscape perspective, these investments are fundamentally a bundling strategy. Cloud computing giants and AI labs typically mix and match multiple vendors when procuring chips, networking equipment, and custom cabling. By taking equity stakes in infrastructure developers, Nvidia gains significant leverage to ensure future facilities are designed around its complete technology stack.

HSBC believes that as competitors find it increasingly difficult to secure critical manufacturing resources, Nvidia’s strategy of locking in capacity early through multi-year agreements will make its supply chain advantage increasingly pronounced. The combined momentum of these two narratives could mark an important turning point for the market to reassess Nvidia’s valuation.