HSBC: NVIDIA's Open-Source Models and Supply Chain Lock-In Create New Catalysts for Valuation Re-Rating
Nashnova编辑部
HSBC analyst Frank Lee argued on August 20 that Nvidia's next revaluation catalysts lie in two under-appreciated narratives: a major push into open-source small language models and multi-year deals locking down supply-chain capacity — both pointing to a customer base evolving from a handful of hyperscalers toward a broader, more resilient ecosystem.
Why does Nvidia need a "new story"?
After the sovereign-AI and emerging-cloud hype faded, Nvidia failed to build a narrative strong enough to drive a meaningful re-rating.
This means → that gap is a key reason it has underperformed the Philadelphia Semiconductor Index year-to-date.
HSBC sees the growth engine shifting from reliance on a few hyperscalers toward a wider, more resilient customer ecosystem. If the market buys this shift, it becomes the key variable for revaluation.
What is the open-source small-model bet really about?
Nvidia now positions itself as the world's largest open-source AI contributor. By its own disclosure, open-source models are the second-most-popular category by token generation volume.
HSBC identifies three reasons open-source SLMs — small language models, far lighter than GPT-class systems — are becoming the preferred inference engine: cost and throughput advantages suit high-frequency agentic loops; they excel at constrained, deterministic tasks and can be embedded in enterprise software; models under 10 billion parameters fit into local GPU memory or edge devices, enabling on-premise deployment.
In plain terms = large models are good at "chatting about anything"; small models are good at "doing one thing fast and right" — and when enterprises actually deploy, the latter is often more practical.
What scenarios do these open-source products cover?
Nvidia's open-source matrix spans multiple core use cases: Nemotron for reasoning and language; Cosmos for robotics and vision; GR00T N1 — billed as the first open foundation model for humanoid robots; Alpamayo for autonomous driving.
On the enterprise-tool side, NVIDIA Agent Toolkit and NeMo are used to build and customize AI agents and generative-AI applications respectively.
This means → a free, highly optimized open-source ecosystem that steers developers to run AI applications on Nvidia hardware first — not just selling the shovel, but designing it so everyone digs with yours.
What exactly has been locked down in the supply chain?
Advanced packaging and memory: in July 2026 Nvidia signed a $1.5 billion multi-year deal with Amkor to expand packaging and testing capacity in Arizona; that same month it struck a $500 billion comprehensive agreement with SK Group covering next-gen AI memory co-development with SK Hynix — including HBM, high-bandwidth memory designed as stacked VRAM for AI chips.
Foundry capacity: Nvidia has reserved 63% (2026) and 52% (2027) of TSMC's CoWoS-L — an advanced process that packages multiple dies together — forcing GPU and ASIC rivals onto alternative suppliers that carry yield risk.
Optical interconnects: $2 billion multi-year strategic agreements with each of Lumentum and Coherent; a separate deal with Corning to scale U.S.-based advanced optical-connectivity manufacturing.
Why take equity stakes in power and land developers?
Nvidia has invested in Cloverleaf Infrastructure, Lancium, and SB Energy, binding power resources to ensure future deployment sites for its chips and embedding its full hardware-and-software stack into facility designs from day one.
Through its partnership with SB Energy and OpenAI, Nvidia locked in land, power, and construction capacity at the PORTS-Pike tech campus in Ohio — initial design supports 4.25 IT-GW of AI-factory capacity, with a cumulative payment cap of $105 billion plus a $1.5 billion investment in SB Energy.
In plain terms = hyperscalers typically mix and match vendors when they build; by taking equity, Nvidia ensures future data centers are architected around its stack from the ground up — trading capital for lock-in.
When does this thesis face its test?
HSBC expects supply-chain constraints across multiple segments to tighten further in 2027, at which point Nvidia's pre-procurement strategy should deliver competitive value well beyond its peers.
This means → whether the supply-chain lock-in advantage actually converts into a market premium when constraints bite is the key inflection point for this narrative.
Nvidia also plans to invest $1 billion in NAVER to expand the "GAK Sejong" AI factory in South Korea from 55 MW to 200 MW by 2028, with a long-term roadmap toward 1 GW of sovereign AI infrastructure — this reflects a lock-in strategy already extending into Asian markets.
Content is for reference only, not financial advice.