NVIDIA Partners with Supply Chain Allies to Expand Production, Substrate Capacity to Reach 2.5x in Three Years
nashnova research
Nvidia announced a joint expansion with multiple supply-chain partners, targeting a 2.5× increase in substrate capacity over three years across substrates, packaging, and testing — its largest capacity lock-in to date, signaling that the AI-chip bottleneck is shifting from design to manufacturing.
Substrates at 2.5× — who is building Nvidia's "foundation"?
Nvidia is partnering with Ibiden, Samsung Electro-Mechanics, and Unimicron to boost substrate capacity to 2.5× current levels within three years.
Ibiden's Ono plant in Japan is the most aggressive — targeting 3× current output, with new capacity ramping from Q1 2028.
This means → substrates — the circuit "baseboards" that carry chips; the bigger and more complex the chip, the harder the board is to make — are becoming a key chokepoint for AI-chip mass production. Nvidia locking in capacity three years out signals exceptionally strong demand expectations.
Why are the packaging numbers the most dramatic?
ASE Group (SPIL) plans to scale co-packaged optics engine capacity to 15× current levels by mid-2027.
Amkor Technology plans to raise advanced packaging capacity by over 300% by end of 2028.
In plain terms = co-packaged optics — embedding optical communication modules right next to the chip so data travels on light instead of electricity, delivering higher speed and lower power — is the core upgrade path for next-gen AI chips. A 15× ramp means this technology is moving from the lab to full-scale production.
Why does testing capacity need to double too?
Nvidia plans to double total semiconductor testing capacity from 2026 to 2027 and will launch large-scale test production in Singapore in February 2027.
This means → the more advanced the chip, the more complex and time-consuming testing becomes. If testing can't keep up, chips produced upstream sit in a queue. Nvidia singling out testing for its own expansion track shows this link is already a real bottleneck.
How did the market vote?
On the day of the announcement, the compute-hardware chain rallied broadly: Super Micro Computer rose over 9%, Dell Technologies gained over 5%, and Nvidia itself added over 2%.
The Nasdaq Composite climbed 1.05% to close at 26,702.86.
This reflects the market reading the expansion as a "hard confirmation" of AI demand — not a slide-deck forecast, but real capital committed to physical capacity.
Can these expansion promises deliver? What to watch?
Whether each supply-chain partner can ramp on schedule and hit yield targets will be the key checkpoints for turning expansion commitments into actual revenue.
In plain terms = announcing an expansion is step one. The real test is whether factories, once built, can produce qualifying products on time. Semiconductor expansions running late or missing yield targets is far from rare historically.
The earliest verification windows are SPIL's optics-engine line in mid-2027 and the Singapore test facility going live in February 2027.
市场有风险，内容仅供研究参考，不构成投资建议。