Huang Renxun's Speech in Taipei: Vera Rubin Mass Production Initiated, Taiwan's AI Computing Power to Double
Taylor Wilson
Nvidia announced its next-generation Vera Rubin architecture has entered production ramp, with plans to double AI supercomputing capacity in Taiwan by 2026 — placing the island's supply chain at the center of global AI infrastructure buildout.
What is Vera Rubin, and why is it entering production now?
Vera Rubin is Nvidia's latest AI supercomputing architecture. Jensen Huang announced on May 29 in Taipei that it has officially entered production ramp.
A single Vera Rubin system weighs roughly 2 tons, contains 1.3 million components, about 8 kilometers of copper cabling, plus liquid cooling and power busbars. This means → this is not a chip — it is an entire "AI factory."
In plain terms = Nvidia is no longer selling graphics cards. It is shipping building-scale computing systems. Production ramp means this goes from lab to mass delivery.
Doubling Taiwan's compute — who is taking the orders?
Huang said Nvidia will double its AI supercomputing capacity in Taiwan by 2026 to meet surging demand.
The event drew a dense lineup of Taiwanese firms: Hon Hai (Foxconn), TSMC, Delta Electronics, Lite-On, Kenmec, Unimicron, and others. TSMC's VP of advanced packaging Jun He showcased a Rubin wafer on stage.
This reflects a core fact: Nvidia's AI factory cannot be built by one company. 150 ecosystem partners collaborate, and Taiwanese firms hold central roles.
"You're so rich now" — what was Huang really saying?
Huang thanked his supply-chain partners on stage and joked: "You're so rich now."
He noted that multiple partners had achieved 3x, 4x, and even 10x business growth over the past few years.
In plain terms = the AI compute boom is not just an Nvidia profit story — the entire supply chain is capturing value. Huang's quip was really saying: the pie is still growing, and this is just the beginning.
MGX platform and 800V power — why does infrastructure matter?
Nvidia's MGX modular reference architecture gives partners a standardized platform for rapid builds, from single-node servers to full-rack AI factories.
As AI factories scale up, Nvidia is pushing a next-generation 800V DC power architecture to improve energy-conversion efficiency and supply stability.
This means → the bottleneck in AI compute is shifting from "are the chips fast enough" to "is there enough power, and can cooling keep up." Infrastructure itself is becoming the competitive moat.
Mainland Chinese firms are on the list — what is the signal?
The partner roster was not Taiwan-only. Mainland firms Innoscience, Luxshare Precision, Megmeet Electrical, and Yuanda Technology were also listed.
This reflects the cross-regional nature of Nvidia's AI factory supply chain. Even against a backdrop of geopolitical tension, mainland suppliers remain part of the ecosystem in specific segments.
Put simply = this supply chain runs on "who can build it, who is cost-effective, who is reliable" — not strictly along political maps.
Content is for reference only, not financial advice.