Nvidia Rubin Moves to HBM4: What Changes for SK Hynix and Micron?
July 21, 2026
Altsets
Research by Altsets Research
Nvidia's Rubin architecture moves to HBM4 and publicly identifies Micron, SK Hynix, and Samsung in its memory ecosystem, creating a product-specific reason to revisit company-wide supplier exposure.
Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.
Key findings
- Nvidia Rubin uses HBM4, and Nvidia's 2026 GTC materials identify Micron, SK Hynix, and Samsung as HBM4 suppliers for Vera Rubin.
- Micron publicly confirms HBM4 designed for Vera Rubin while SK Hynix has a named multi-year Nvidia memory partnership, but the Altsets relationship percentages remain company-wide rather than Rubin-specific.
Nvidia's move from Blackwell toward Vera Rubin changes the product context around its mapped memory suppliers. In the displayed Altsets data, the SK Hynix relationship carries two large directional weights: 27.88% of SK Hynix revenue is associated with Nvidia, while 27.33% of Nvidia's cost base is associated with SK Hynix. Those company-wide percentages establish economic importance before the architecture question is introduced.
Rubin uses HBM4 memory, and Nvidia's own 2026 GTC materials identify HBM4 from Micron, SK Hynix, and Samsung in the Vera Rubin system. That does not mean the existing relationship percentages can be reassigned to Rubin. It does mean the architecture transition gives investors a product-specific reason to revisit the supplier network.
Rubin materially changes the memory subsystem
Nvidia says Rubin uses HBM4 with up to 22 TB/s of memory bandwidth and up to 288 GB of HBM4 capacity per GPU. The company describes HBM4 as critical to long-context and high-concurrency AI inference. This creates a more specific supplier question than "Who sells memory to Nvidia?" The question becomes: Which mapped memory suppliers are actually shipping products designed for the Rubin architecture?
Micron publicly confirms Rubin-specific HBM4
Micron announced in March 2026 that its HBM4 36GB 12-high product had entered high-volume production and was designed for Nvidia Vera Rubin. Micron said the product delivers more than 2.8 TB/s of bandwidth per stack. Micron also announced SOCAMM2 products and PCIe Gen6 storage designed around the wider Vera Rubin platform.
This is direct product-specific evidence. Altsets separately maps Micron to Nvidia at 17.62% of Micron revenue and 14.00% of Nvidia's cost base. The public Rubin announcement and proprietary relationship metric therefore answer different parts of the same question.
SK Hynix has a named multi-year Nvidia memory partnership
SK Hynix and Nvidia announced a multi-year technology partnership in June 2026. The partnership covers next-generation memory for Nvidia AI infrastructure, including Vera Rubin. That direct customer-specific evidence strengthens the interpretation of the Altsets SK Hynix to Nvidia relationship.
Again, the 27.88% supplier-revenue percentage remains company-wide. It is not a Rubin revenue percentage.
Architecture transitions can change supplier relevance before financial statements show it
A new platform can change: memory generation, capacity per system, bandwidth requirements, qualification requirements, packaging, storage, networking, cooling, and system architecture. Those changes can alter which suppliers matter most even before the supplier's reported revenue mix visibly changes. Product-transition research is therefore inherently forward-looking.
The relationship graph identifies the existing commercial base. The architecture roadmap shows what may change next.
Do not assume the transition creates a winner-take-all outcome
Nvidia's GTC materials identify Micron, SK Hynix, and Samsung as HBM4 suppliers for Vera Rubin. That matters because a product transition can broaden the qualified supplier set rather than concentrate it. An investor should therefore avoid turning one named partnership into an assumption of exclusive supply. The relevant questions are qualification status, shipment timing, capacity, yields, performance, power efficiency, customer allocation, and contract structure.
Relationship concentration and architecture share are different metrics
SK Hynix at 27.88% of supplier revenue associated with Nvidia tells us Nvidia is an important customer for SK Hynix. Micron at 17.62% tells us the same for Micron at a lower displayed concentration. Neither number tells us each company's share of Rubin HBM4 sockets.
That is a different denominator. This distinction matters because investors often take a company-wide customer percentage and turn it into a product-market-share estimate. The data does not support that shortcut.
What the transition does support
The strongest defensible conclusions are:
- Nvidia is economically significant to both SK Hynix and Micron in the displayed relationship data.
- Rubin requires HBM4.
- Nvidia publicly identifies Micron, SK Hynix, and Samsung as HBM4 suppliers for Vera Rubin.
- Micron and SK Hynix each provide direct Rubin-related product evidence.
- The architecture transition justifies renewed supplier analysis.
- The relationship percentages should remain company-wide unless architecture-specific commercial data is disclosed.
A repeatable architecture-transition workflow
- Identify the old and new product architecture.
- List the components that materially change.
- Map current suppliers.
- Find product-specific qualification or partnership evidence.
- Separate company-wide relationship metrics from architecture-specific supply.
- Monitor shipment and production milestones.
- Track new suppliers entering or leaving the qualified set.
- Revisit customer concentration after the architecture ramps.
The company-wide versus segment exposure guide explains why denominator discipline matters when narrowing a company relationship to one product. The Nvidia supply-commitment analysis adds a separate forward signal from procurement commitments. For relationship definitions and limitations, read the Altsets supply-chain data methodology. Browse the Company Dependency Research collection for other company-specific supply-chain analysis.