HSBC identifies open-source models and supply chain lockups as hidden catalysts driving Nvidia’s re-rating.
HSBC analyst Frank Lee argues in his latest research note that Nvidia is building defensive moats along two underappreciated fronts. First, the company is making an aggressive push into open-source small language models (SLMs), positioning itself as the world’s largest open-source AI contributor. Second, Nvidia has systematically locked in critical supply capacity across advanced packaging, memory, optical components, and energy infrastructure through a series of multi-year procurement agreements extending into 2026. These strategic moves could serve as key catalysts for a valuation re-rating.
Published on August 20, Lee’s note emphasizes that these previously overlooked narratives point to a fundamental evolution in Nvidia’s growth engine: a transition from a single structure dependent on a handful of hyperscale customers toward a broader, more resilient ecosystem. After the sovereign AI and neocloud boom faded, Nvidia struggled to establish a new narrative strong enough to drive a meaningful stock re-rating—a factor that contributed to its year-to-date underperformance against the Philadelphia Semiconductor Index. The strategic bet on open-source AI is now filling that void.
According to Nvidia’s disclosures, open-source models have cumulatively become the second most popular category by token generation volume. HSBC highlights that open-source SLMs are rapidly becoming the inference engine of choice for intelligent agents and edge applications. This market preference is driven by three factors. First, latency and throughput advantages: autonomous intelligent agents operate at high frequency across execution loops—parsing intent, calling APIs, and evaluating results—and invoking large frontier language models at every micro-step creates severe bottlenecks. SLMs offer significant advantages in inference cost and throughput. Second, task-specific intelligence: SLMs excel at constrained, deterministic tasks, prompting enterprises to embed domain-specific models 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 comprehensive, targeting multiple core scenarios. The Nemotron family addresses reasoning and language tasks; Cosmos focuses on “physical AI” for robotics and vision; GR00T N1 is positioned as the world’s first open general-purpose foundation model for humanoid robots; Alpamayo targets autonomous driving; and the NVIDIA Agent Toolkit and NeMo are used respectively for building enterprise-grade AI agents and generative AI applications. HSBC notes that this free, highly optimized ecosystem serves as a strategic lever to steer developers toward running AI applications preferentially on Nvidia hardware. This approach has the potential to expand the addressable customer base from a handful of hyperscale cloud providers to millions of developers and sovereign nations.
Meanwhile, AI compute demand continues to outstrip supply amid persistent capacity constraints. To secure its position, Nvidia has executed a series of multi-year agreements across critical supply chain segments. In advanced packaging and memory, Nvidia signed a $1.5 billion (approximately NT$48 billion) multi-year agreement with Amkor Technology in July 2026 to support expanded semiconductor packaging and test capacity in Arizona. That same month, the company announced a comprehensive partnership with SK Group valued at up to $500 billion (approximately NT$15.9 trillion), covering joint development of next-generation AI memory, including high-bandwidth memory (HBM), with SK Hynix, alongside plans to build a 2-gigawatt Vera Rubin AI factory in South Korea. On the foundry side, Nvidia has pre-booked 63% of Taiwan Semiconductor Manufacturing Company’s (TSMC) CoWoS-L advanced packaging capacity for 2026 and 52% for 2027. This effectively forces GPU and ASIC competitors to seek alternatives, where yield risks remain elevated.
In optical interconnect, as AI network infrastructure shifts from copper cabling to optical solutions, Nvidia has secured separate multi-year strategic agreements with Lumentum and Coherent. Each commitment totals $2 billion (approximately NT$64 billion) to support research and development and U.S. domestic manufacturing capacity buildout, while securing future access to advanced laser components. Additionally, Nvidia signed a multi-year commercial and technology collaboration agreement with Corning to expand U.S. domestic manufacturing scale for advanced optical connectivity solutions. On energy and land, Nvidia is locking in critical assets through direct equity stakes in infrastructure developers, including Cloverleaf Infrastructure, Lancium, and SB Energy. This strategy ties up power resources to ensure its chips have dedicated facilities while embedding Nvidia’s full hardware and software stack into the early design phases of these sites.
Through partnerships with SB Energy and OpenAI, Nvidia has secured land, power, and construction capacity at Ohio’s PORTS-Pike technology park, with initial designs supporting 4.25 IT-GW of AI factory capacity. Cumulative payment obligations for this project are capped at $105 billion (approximately NT$3.3 trillion), supplemented by a $1.5 billion equity investment in SB Energy. Separately, Nvidia 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 ambitions to scale toward 1 gigawatt of sovereign AI infrastructure. From a competitive standpoint, these investments represent a deliberate bundling strategy. While cloud computing giants and AI labs typically mix and match vendors for chips, networking, and cabling, Nvidia’s equity stakes in infrastructure developers grant it significant leverage to ensure future facilities are architected around its complete technology stack.
HSBC expects supply chain constraints across multiple segments to intensify further in 2027. At that juncture, Nvidia’s early procurement strategy will generate competitive value far exceeding that of its peers, potentially commanding a higher market premium. As rivals find it increasingly difficult to secure critical manufacturing resources, Nvidia’s supply chain advantage will grow more pronounced. The combined momentum of these two narratives—the expansion into open-source AI and the systematic locking of supply chain capacity—could mark a pivotal turning point, prompting the market to reassess Nvidia’s valuation and driving a sustained re-rating.