Skip to the yard
Waxsealai

NVIDIA’s Supply Chain Strategy: Scaling AI Systems Through Supplier Partnerships

Shipped
Shipper
NVIDIA Writers
Readers
781

NVIDIA uses close supplier relationships, chip level tracking and factory coordination to expand production and shorten system turnaround from five weeks to two.

NVIDIA’s Supply Chain Strategy: Scaling AI Systems Through Supplier Partnerships

NVIDIA’s ability to scale AI systems depends on more than chip design. Deb Shoquist, who leads the company’s operations, describes a supply chain built around long standing manufacturing relationships, detailed production visibility and rapid collaboration when factory processes fall short. The approach has helped NVIDIA move from selling chips to building data center systems, which requires more partners and new production capabilities.

Shoquist’s role is to turn designs into a reliable, high performance supply chain. That means matching each product with manufacturers that have the right technical capabilities, can invest in production ramps and are willing to take on work they haven’t done before. If a partner has two of those strengths but lacks the third, NVIDIA may support its development so it can meet the need.

The network includes TSMC, Foxconn, Wistron, SPIL, King Yuan Electronics (KYEC), Unimicron, Lumentum, Coherent and Amkor. TSMC has worked with NVIDIA since 1998; SPIL and KYEC were assembling and testing its chips more than two decades ago. Shoquist meets key suppliers each quarter and spends six weeks a year visiting partners in Taiwan, China, Korea, Vietnam and the United States, including time on factory floors.

Foxconn

Relationships built on factory work

For Shoquist, regular contact with production teams matters as much as executive relationships. Factory workers and production leaders see problems that can be difficult to surface from a meeting room: a need for more technical support, quality engineers, improved designs, automation or additional material buffers. Trust makes those conversations possible, and gives both sides a clearer view of what needs to change.

That familiarity also shapes requests to increase output. Before asking a supplier to push a production line harder, NVIDIA needs to understand its processes and people, Shoquist says. A specific request is easier to assess and act on when it reflects real knowledge of the factory. When the ramp succeeds, both companies benefit, reinforcing the relationship for the next challenge.

One example is Amphenol’s cable cartridge, which Shoquist described as the company’s most complex interconnect assembly to date. The product has 1,500 small cable legs, each requiring precise assembly to reach a quality level below 1 defect per million parts. Amphenol built more than 100 automated lines at its Vietnam factory to produce the cartridge at the required speed and scale. For Shoquist, the effort shows that complex production depends on technical skill and equipment, as well as a supplier’s commitment to solving problems alongside the customer.

Building capacity for systems and optics

NVIDIA’s supply chain needs changed as the company expanded from chips into data center systems. Semiconductor manufacturing already had an established industrial base, so the task was largely to find the best technology match for a product. Systems work has required a broader set of partners. In some newer areas, the supplier base is still developing and lacks the capacity or capital to meet NVIDIA’s target volumes.

Optics is one such area. Shoquist says optical suppliers have relevant technology and the willingness to grow, but don’t yet have enough scale or capital to build factories for the volumes NVIDIA needs. The company has supported partners including Coherent and Lumentum. For co-packaged optics, NVIDIA is also assembling supply chains that involve companies new to the optics sector. Coordinating expertise across those partners will be part of bringing the technology to high volume.

Wistron

Allocating chips by factory performance

When a critical component is in short supply, NVIDIA needs to decide where each unit will have the greatest effect. Shoquist identifies TSMC wafers as essential to production. As chips complete assembly each week, the company allocates them among factories based on performance. A factory with low yield or a longer than desired time of ownership may receive fewer chips the following week.

NVIDIA tracks each chip from its die identification number through to the customer invoice, and suppliers must support that traceability process. The tracking gives operations teams a detailed view of where units are and how quickly they move through production. When supply and demand call for it, NVIDIA has reduced the time from a chip leaving TSMC to a finished system leaving Foxconn or Wistron from five weeks to two.

This method connects chip allocation to conditions on the factory floor. It gives the company a way to direct production toward lines that can meet output and quality needs, while monitoring the movement of individual chips through assembly and system production.

Coherent

Recovering from a difficult production ramp

A ramp for the NVIDIA GB200 compute board at Foxconn and Wistron shows how that coordination works when a new process misses its target. The final stage used an automated handler for testing, a new approach intended to support high volume production of complex products. During the initial ramp, however, throughput was far below plan.

The operations director for Taiwan manufacturing decided to disable the automated process and return to manual test loading to protect the product and shipments. Manufacturing partners supplied hundreds of operators, keeping production moving while the teams worked on a longer term fix. The throughput problem had several contributing factors, including test tray assembly and design, automation hardware and control software, and product yield.

NVIDIA then brought together specialists from design engineering, operations industrial, product engineering and test engineering, alongside Foxconn’s automation engineers. The teams worked on the line to redesign and reactivate it before the scale ramp. Shoquist presents this kind of response as part of operations work: when production hits a difficult problem, teams across company boundaries need to concentrate on getting the process back on track.

Amphenol

Shoquist recently paused her supplier visits to accept the 2026 Dr. Lisa Su Woman of Innovation Award. She says the appeal of operations lies in continually changing the supply chain to meet demand for new products, with teams given resources to tackle difficult production goals. The practical test for NVIDIA’s approach will be whether its network can keep adding capacity, especially in newer areas such as optics, while preserving the production speed and quality its systems require.

Source: NVIDIA Writers · blogs.nvidia.com

Same vesselRelated cargo