Chips & Compute

NVIDIA Locks In 3.5x More Substrate Capacity From Kinsus By 2028

NVIDIA signed new deals with Kinsus, Samsung Electro-Mechanics and SPIL, including a 3.5x substrate capacity expansion by 2028 and multibillion-dollar packaging funding through 2029.

NVIDIA Secures More Substrate and Packaging Capacity for AI Platforms
NVIDIA Secures More Substrate and Packaging Capacity for AI PlatformsQuanlecntt2004 / pixabay
By Rebecca Stone5 min read

Why it matters

  • NVIDIA announced the Kinsus, Samsung Electro-Mechanics and SPIL agreements on November 17, 2026.
  • Kinsus will expand NVIDIA-dedicated substrate capacity more than 3.5x by 2028, with production starting in H2 2027 at its Youth campus in Taoyuan.
  • SPIL's multibillion-dollar deal adds dedicated advanced-packaging capacity, including bumping, interposer fabrication, assembly and testing, by mid-2029.
  • Vera Rubin NVL72 and Vera CPUs began volume shipments in Q3 on TSMC 5nm and 3nm processes with CoWoS packaging.
  • Amkor's comparable NVIDIA agreements total $3 billion across Arizona ($1.5B) and South Korea ($1.3B).

NVIDIA announced on November 17, 2026 new and amended supply agreements with Kinsus, Samsung Electro-Mechanics and Siliconware Precision Industries, anchored by a Kinsus plan to expand NVIDIA-dedicated substrate capacity by more than 3.5 times current levels by 2028. The deals extend NVIDIA's campaign to secure the constrained back-end manufacturing layers — substrates and advanced packaging — that determine how fast AI accelerator platforms can ship.

The announcements arrive as NVIDIA's Vera Rubin NVL72 racks and Vera CPUs have already begun volume shipments in the third quarter, built on TSMC's 5nm and 3nm process technologies and CoWoS advanced packaging. The new agreements target capacity that comes online after that ramp, positioning suppliers for later platform generations.

What did Kinsus commit to?

Under a new multiyear agreement, Kinsus will expand advanced ABF package-substrate manufacturing at its Youth campus in Taoyuan, Taiwan. Production is expected to begin in the second half of 2027, with the additional capacity targeted for the third quarter of 2028.

Kinsus CEO Scott Chen said the agreement builds on a supplier relationship with NVIDIA that has lasted more than a decade. The 3.5-fold expansion figure applies specifically to substrate capacity dedicated to NVIDIA, not to Kinsus's total output.

ABF substrates carry the dense interconnects between a GPU die and the package that mounts it to a board. They have been a recurring bottleneck in AI hardware supply chains, which is why NVIDIA is signing multiyear, capacity-specific contracts rather than buying on the spot market.

What changes at Samsung Electro-Mechanics?

The Samsung Electro-Mechanics amendment focuses on existing factories rather than new construction. It aims to improve operational capabilities at current plants to support increased production of NVIDIA's accelerated-computing platforms.

The amendment builds on the companies' multiyear substrate-manufacturing collaboration in Sejong and Busan, South Korea. That collaboration supplies substrates for NVIDIA's compute and high-speed networking technologies.

Unlike the Kinsus deal, the Samsung Electro-Mechanics terms carry no headline capacity multiplier or target date. The change is factory-specific: upgrades inside plants already feeding NVIDIA's supply chain.

What does the SPIL packaging deal cover?

NVIDIA and Siliconware Precision Industries (SPIL) are expanding advanced-packaging operations in Taiwan under a separate multiyear agreement supported by multibillion-dollar NVIDIA funding. The companies aim to increase contracted advanced-packaging capacity dedicated to NVIDIA by mid-2029.

Planned production lines will provide turnkey support spanning:

  • bumping
  • interposer wafer fabrication
  • assembly
  • testing

The turnkey scope matters for output. Housing bumping, interposer fabrication, assembly and test under dedicated lines removes handoffs between vendors — the kind of interface friction that slows advanced-packaging throughput when capacity is split across suppliers.

The SPIL packaging project is distinct from the optical-engine expansion in NVIDIA's earlier October 21, 2026 supply-chain announcement. That package targeted a 15-fold increase in SPIL's optical-engine capacity for co-packaged optics by mid-2027, plus an aggregate 2.5-fold substrate-capacity increase over three years. The new November terms add the dedicated Kinsus expansion and the factory-specific Samsung Electro-Mechanics changes on top of that broader buildout.

How does the SPIL funding compare with Amkor?

The SPIL funding structure follows a pattern NVIDIA has already established with Amkor Technology. Amkor's July 23, 2026 announcement covered a $1.5 billion multiyear agreement with an NVIDIA prepayment supporting expanded U.S. advanced-packaging capacity in Arizona, plus development work on high-density interconnects and heterogeneous integration.

An October announcement added a $1.3 billion South Korea agreement. Amkor stated that the Arizona and South Korea multiyear agreements total $3 billion.

NVIDIA did not disclose a precise figure for the SPIL funding, describing it only as multibillion-dollar. The Amkor comparison shows the scale: NVIDIA is now committing capital in billion-dollar increments to lock in packaging and substrate lines years before the platforms that will use them reach volume production.

How do Vera Rubin shipments fit in?

NVIDIA said Vera Rubin NVL72 and Vera CPUs began volume shipments in the third quarter, using TSMC's 5nm and 3nm process technologies and CoWoS advanced packaging.

TSMC senior vice president and deputy co-COO Kevin Zhang said the collaboration allows the companies to plan future technology requirements and prepare manufacturing capabilities as NVIDIA introduces new platforms. That planning function is the operational core of these agreements: capacity commitments made in 2026 map onto platform ramps in 2027 through 2029.

Foxconn chairman Young Liu said the manufacturer is deepening its collaboration with NVIDIA across next-generation GPU and networking platforms. Foxconn draws on a production network across Asia, North America and Europe to support the volume ramp of next-generation AI infrastructure.

Why does the staggered timing matter?

The practical effect of the November 17 announcements is a staggered buildout across NVIDIA's supply chain:

  • TSMC is supporting the current Vera Rubin shipment milestone now.
  • Kinsus brings new substrate capacity online in the second half of 2027, reaching full additional volume in the third quarter of 2028.
  • SPIL delivers dedicated advanced-packaging capacity by mid-2029.

That sequencing implies NVIDIA expects platform demand to keep scaling past the current Vera Rubin generation — the Kinsus and SPIL capacity would arrive too late for this ramp and is sized for what follows. It also signals where NVIDIA sees structural risk: front-end wafer capacity at TSMC is secured, so the company is spending its capital securing the back end, where substrates, interposers and packaging test lines have historically constrained output.

By signing named-supplier, named-factory agreements with committed dates — Youth campus in Taoyuan, plants in Sejong and Busan, packaging lines in Taiwan — NVIDIA converts supply-chain commitments into scheduled capacity rather than forecasts, and its prepayment-funded structure shifts expansion risk from suppliers to itself.

Source: unite.ai

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

728 articles

Related articles

  1. TSMC Is Weighing 3D DRAM Production at NVIDIA's Request
  2. ASUS Ramps Vera Rubin NVL72 Shipments as Rack Race Shifts to Execution
  3. Nvidia Case Puts AI Chip Supply Chains Under New Scrutiny
  4. Nvidia launches open-weight Nemotron models for enterprise agents and AI PCs
  5. Samsung Electro-Mechanics expands Nvidia deal to substrates and MLCCs

« Previous articleNext article »