Nvidia AI Chip Suppliers: Who Makes Blackwell and Rubin
Last updated: October 15, 2026. Labels: Confirmed by company (Nvidia or the supplier, in a filing or press release), Reported (reputable press or a research firm), Unconfirmed (single secondary source or a product mapping no source states directly).
Nvidia designs its chips and outsources manufacturing. Its annual report for fiscal 2026 (Form 10-K, filed 2026-02-25) states: "We utilize foundries, such as … TSMC, and Samsung … to produce our semiconductor wafers. We purchase memory from SK Hynix Inc., Micron Technology, Inc., and Samsung. We utilize CoWoS technology for semiconductor packaging. We engage with … contract manufacturers such as Hon Hai Precision Industry Co., Ltd., Wistron Corporation, and Fabrinet to perform assembly, testing and packaging of our final products." [1] The 10-K doesn't say which supplier serves which product, so per-product mappings below are labelled. For Nvidia's full supplier map, see Nvidia suppliers.
Quick answer
| Layer | Supplier(s) | Product scope (as stated) | Label | Source |
|---|---|---|---|---|
| Wafers (foundry) | TSMC, Samsung | Company-level | Confirmed by company | [1] |
| Blackwell process | TSMC custom 4NP | Blackwell GPUs | Confirmed by company | [2] |
| US wafer production | TSMC Arizona (Phoenix) | Blackwell, started 2025 | Confirmed by company | [3] |
| Advanced packaging | CoWoS (TSMC's packaging technology) | Company-level | Confirmed by company | [1] |
| US packaging & test | Amkor, SPIL (Arizona) | Partnership announced 2025-04-14 | Confirmed by company | [3] |
| HBM / memory | SK hynix, Samsung, Micron | Company-level (10-K); HBM4 for Vera Rubin (CEO keynote, reported) | Confirmed by company / Reported | [1][4] |
| HBM4 share for Rubin | SK hynix ~60–70%, Samsung ~25–30%, Micron smaller | Industry estimates | Reported (estimate) | [4] |
| Package substrates | Ibiden ("dominant supplier") | Nvidia AI chips, as of Dec 2024 | Reported | [5] |
| Assembly / test (contract mfg.) | Hon Hai (Foxconn), Wistron, Fabrinet | Company-level | Confirmed by company | [1] |
| US AI supercomputer plants | Foxconn (Houston, announced 2025-04-14); Wistron (Fort Worth, opened 2026-07-21) | GB300 now; Vera Rubin cell at Wistron | Confirmed by company | [3][8] |
| Vera Rubin system partners | Dell, HPE, Lenovo, Supermicro, Foxconn, Quanta Cloud Technology, Wistron, Wiwynn, Inventec, Pegatron, Compal and others | Named by Nvidia, 2026-05-31 | Confirmed by company | [6] |
Who makes Nvidia's chips? (foundry)
Nvidia's 10-K names TSMC and Samsung as wafer foundries [1]. Nvidia says Blackwell GPUs are "manufactured using a custom-built TSMC 4NP process" (Confirmed by company) [2]. On 2025-04-14, Nvidia said "Blackwell chips have started production at TSMC's chip plants in Phoenix, Arizona" [3]. The process node for Vera Rubin isn't stated in our sources (Unconfirmed). Which products, if any, Samsung manufactures isn't specified (Unconfirmed).
Packaging: CoWoS
Nvidia's 10-K says it uses CoWoS for semiconductor packaging [1]. CoWoS is TSMC's advanced packaging platform. In the US, Nvidia named Amkor and SPIL for packaging and testing in Arizona (Confirmed by company) [3].
HBM memory: SK hynix, Samsung, Micron
Nvidia buys memory from SK hynix, Micron and Samsung (Confirmed by company) [1]. The Korea Herald reported that at GTC Taipei on 2026-06-01, Jensen Huang confirmed that HBM from all three will go into Vera Rubin (Reported) [4]. The same article cites industry estimates of roughly 60–70% of Rubin HBM4 volume for SK hynix and 25–30% for Samsung, with Micron taking a smaller share (Reported, estimate) [4]. On 2026-02-12 Samsung said it had begun mass production of HBM4 and shipped commercial products to unnamed customers; Reuters reported the same day that Samsung is trying to narrow the gap in supplying parts for Nvidia's AI accelerators (Confirmed by supplier for shipments; Nvidia not named) [7][9].
Substrates
Bloomberg reported on 2024-12-29 that Ibiden is "the dominant supplier of chip package substrates used in Nvidia Corp.'s cutting-edge semiconductors". It added that "all of Nvidia's AI semiconductors now use Ibiden's substrates", while noting Unimicron is eyeing the field (Reported, as of December 2024; may have changed since) [5]. Nvidia's 10-K says it may directly procure "memory, substrates, and a variety of components" [1].
Server and rack assembly
- Company-level: Hon Hai (Foxconn), Wistron and Fabrinet perform assembly, testing and packaging (Confirmed by company) [1].
- US: On 2025-04-14 Nvidia said Foxconn (Houston) and Wistron (named as Dallas) were building plants, with mass production expected to ramp "in the next 12-15 months" [3]. On 2026-07-21 Nvidia said Wistron opened its Fort Worth, Texas D1 plant, which produces the GB300 Grace Blackwell Ultra Superchip and has a cell that will produce the Vera Rubin Superchip (Confirmed by company) [8]. We haven't confirmed the current status of Foxconn's Houston plant from a primary source (Unconfirmed).
- Vera Rubin: on 2026-05-31 Nvidia said Rubin is "ramping into full production" and named system partners including Dell, HPE, Lenovo and Supermicro, plus Foxconn, Quanta Cloud Technology (QCT), Wistron, Wiwynn, Inventec, Pegatron and Compal. Production shipments are "set to begin starting this fall" (Confirmed by company) [6].
- Market shares by ODM (e.g. Foxconn's share of GB200 racks) circulate in secondary sources that we haven't verified (Unconfirmed, not published).
What we don't know yet
Rubin process node; Samsung foundry's product role; current substrate share; ODM shares by platform; optics/CPO suppliers for Spectrum-X Photonics.
FAQ
Who makes Nvidia chips? Nvidia designs them. Its 10-K names TSMC and Samsung as wafer foundries, and Nvidia says Blackwell uses a custom TSMC 4NP process. (Confirmed by company) [1][2]
Who supplies HBM to Nvidia? Nvidia's 10-K names SK hynix, Micron and Samsung as memory suppliers. The Korea Herald reported that Jensen Huang confirmed all three for Vera Rubin's HBM4. (Confirmed by company / Reported) [1][4]
What is CoWoS and does Nvidia use it? CoWoS is TSMC's advanced packaging technology. Nvidia's 10-K says it uses CoWoS for semiconductor packaging. (Confirmed by company) [1]
Who builds Nvidia AI servers? Contract manufacturers named by Nvidia include Hon Hai (Foxconn), Wistron and Fabrinet. For Vera Rubin, Nvidia names partners including Dell, HPE, Lenovo, Supermicro, Foxconn, QCT, Wistron and Wiwynn. (Confirmed by company) [1][6]
Are Nvidia chips made in the US? Nvidia said Blackwell production began at TSMC's Arizona plants in 2025, with Amkor and SPIL packaging in Arizona. Wistron's Fort Worth, Texas plant opened in July 2026. (Confirmed by company) [3][8]
Sources
- NVIDIA, Form 10-K for fiscal year ended 2026-01-25, filed 2026-02-25. sec.gov
- NVIDIA, Blackwell Architecture page (accessed 2026-10-15). nvidia.com
- NVIDIA Blog/Newsroom, "NVIDIA to Manufacture American-Made AI Supercomputers in US for First Time", 2025-04-14. blogs.nvidia.com
- The Korea Herald, "Huang confirms Vera Rubin in full production with Korean memory", 2026-06-01. koreaherald.com
- Bloomberg via Yahoo Finance, "Nvidia Supplier Ibiden Weighs Faster Expansion for AI Demand", 2024-12-29. finance.yahoo.com
- NVIDIA Newsroom, "NVIDIA Vera Rubin Ramps Into Full Production…", 2026-05-31. nvidianews.nvidia.com
- Reuters, "Samsung ships latest HBM4 chips to catch-up in AI race", 2026-02-12. reuters.com
- NVIDIA Blog, "Built in Fort Worth: Wistron Opens Advanced Manufacturing Plant to Produce NVIDIA AI Systems", 2026-07-21. blogs.nvidia.com
- Samsung Global Newsroom, "Samsung Ships Industry-First Commercial HBM4", 2026-02-12. news.samsung.com