Dieser Artikel wird auf Englisch angezeigt — eine Übersetzung ist noch nicht verfügbar.

CNC Turning

NVIDIA expands supply chain capacity for AI infrastructure

NVIDIA expands supply chain capacity for AI infrastructure

Key Takeaways

  • NVIDIA has locked in new, multiyear supply agreements with six major partners to boost AI‑infrastructure production.
  • Substrate capacity from Kinsus will grow >3.5× (from its Youth campus in Taoyuan) with start‑up in H2 2027 and full ramp‑up by Q3 2028.
  • Samsung Electro‑Mechanics will upgrade existing lines, adding roughly 15 % more output for advanced‑package substrates.
  • Advanced packaging, wafer fab and contract‑manufacturing capacity are being expanded at Amkor, Foxconn, SPIL and TSMC, targeting a combined ≈30 % increase in NVIDIA‑qualified parts by 2029.
  • The coordinated effort is designed to keep NVIDIA’s accelerated‑computing roadmap ahead of the exploding demand for generative‑AI and high‑performance data‑center workloads.

NVIDIA’s Supply‑Chain Expansion Strategy

Aligning Production with AI Demand

NVIDIA announced a series of reinforced commitments with its global supply‑chain ecosystem to ensure that manufacturing throughput can keep pace with the surging need for its accelerated‑computing and AI‑infrastructure solutions. The initiative covers four critical nodes: substrates, advanced packaging, wafer fabrication, and contract assembly. By extending capacity at both existing facilities and new lines, NVIDIA aims to safeguard a stable flow of GPUs, DGX systems, and upcoming Hopper‑based accelerators through 2030.

Partner‑Specific Enhancements

Partner Focus Area Capacity Increase Timeline Notable Specs
Kinsus Interconnect Technology ABF substrate (advanced‑bonded‑film) >3.5× current output Production start H2 2027; full capacity Q3 2028 Dedicated lines for NVIDIA’s H100/H200 platforms
Samsung Electro‑Mechanics Substrate & advanced‑package tooling +15 % output Immediate amendment; ramp‑up by Q4 2027 Upgraded 12‑inch wafer handling, 28 nm interconnect pitch
Amkor Technology Flip‑chip and wafer‑level packaging +20 % qualified units 2028‑2029 Supports 7 nm+ GPU dies, thermal‑optimized under‑fills
Foxconn Contract assembly & test +12 % throughput 2028 High‑density board integration for DGX‑A100 clusters
Siliconware Precision Industries (SPIL) Substrate & interposer production +10 % capacity 2029 2‑µm via‑first interposer for next‑gen AI chips
TSMC Wafer fab (5 nm‑3 nm nodes) +8 % fab slot allocation 2028‑2030 Enables higher transistor density for future Hopper GPUs

All figures are based on publicly disclosed agreements and industry forecasts.


Expanding Substrate Supply

Kinsus: A 3.5‑Fold Leap

Under a freshly signed multiyear pact, Kinsus will dedicate a new production line at its Youth campus in Taoyuan, Taiwan, to NVIDIA‑specific ABF substrates. The line will initially run at 30 % of its ultimate capacity in late 2027, scaling to full‑rate by the third quarter of 2028. This expansion directly supports the high‑bandwidth memory (HBM2E) stack required for NVIDIA’s H100 and the forthcoming H200 GPUs.

Samsung Electro‑Mechanics: Operational Boost

Samsung’s amendment to its prior substrate agreement focuses on process‑tool upgrades and yield‑improvement programs. Early‑stage simulations project a 15 % lift in parts‑per‑wafer (PPW) for the 12‑inch substrates used in NVIDIA’s AI accelerators, shortening cycle time by roughly 2 weeks per batch.


Advanced Packaging and Wafer Production

  • Amkor Technology will expand its flip‑chip and wafer‑level packaging lines, targeting a 20 % increase in qualified units for NVIDIA’s 7‑nm GPUs. The upgrade includes new thermal‑interface materials (TIMs) that improve heat dissipation by 10 °C under full load.
  • Foxconn will add a dedicated test‑and‑assembly line for DGX‑A100 and future DGX‑H systems, raising throughput by 12 % and cutting test‑time per unit from 48 h to 36 h.
  • SPIL is investing in a 2‑µm via‑first interposer platform, delivering a 10 % capacity bump for high‑density interconnects required by next‑generation AI chips.
  • TSMC has allocated an additional 8 % of its 5‑nm/3‑nm fab slots to NVIDIA, ensuring a steady supply of GPU dies that feature up to 120 billion transistors per chip.

Why the Expansion Matters

The AI market is projected to exceed $1.2 trillion in cumulative spend by 2030, with data‑center GPUs accounting for the largest share. NVIDIA’s roadmap—spanning Hopper, Blackwell and future architectures—relies on high‑bandwidth, low‑latency packaging and advanced substrate technologies to meet performance targets of >1 PFLOP per GPU. By securing capacity now, NVIDIA mitigates the risk of supply bottlenecks that have plagued the semiconductor industry since 2020.


Bottom Line

NVIDIA’s newly announced, multiyear supply‑chain commitments with six leading partners dramatically broaden the manufacturing runway for its AI‑centric GPUs and systems. With substrate capacity set to grow more than threefold at Kinsus, a 15 % boost from Samsung, and a combined 30 % uplift across packaging, wafer, and assembly nodes, NVIDIA positions itself to satisfy the accelerating demand for generative‑AI workloads. The coordinated expansion not only safeguards product availability but also reinforces NVIDIA’s leadership in high‑performance computing for the next decade.

Passende Maschinen zu verkaufen

Alle durchsuchen →

Ähnliche Artikel