FinmoConf

Taiwan Semiconductor Earnings

Quanta (2382.TW) Earnings Calls

English notes on 1 Taiwan earnings call.

Tier
Self-disclosed AI-rack ODM — not on NVIDIA's published 800V list
Role
Notebook ODM that now prints most of its sales as AI and general servers for CSPs and NeoClouds

Who Quanta is

Quanta Computer (2382.TW) is Taiwan's largest notebook contract manufacturer. The live P&L is no longer that company. From Q1 2026, servers have been more than 80% of sales, and AI has been 75–80% of server revenue. It is not on NVIDIA's published 800 VDC component list. It sits in the same slot as Foxconn and Wiwynn: the listed company that has to ship the cabinet.

Business mix

By the first half of 2026 the weights were:

  • Servers — more than 80% of sales, AI plus general compute. AI is guided to 80% of that server line in the second half of 2026.
  • Notebooks — the rest of the book. First-half 2026 shipments 11.5 million units, better than the prior "flat to slightly up" guide, then guided down more than 20% in Q3 and double-digit down for the full year as memory and CPU prices squeeze brands.
  • Automotive / smart driving — still growing double-digit, but a low-single-digit mix because AI servers crowd it out.

Where it sits in the NVIDIA 800 VDC stack

The integrator on NVIDIA's diagram: compute rack, power shelf, 45°C loop. Quanta does not claim an 800 V SKU. The August 2026 call never said 800 V or HVDC.

What it did date is the factory. AI-server capacity is guided to double by end-2026 versus end-2025 (Thailand and California taking more than 80% of first-half capex). The NT$19.7bn AUO Huaya plant in Guishan is meant for 2028 AI-server backend assembly, next to the R&D HQ and the existing server lines. Management's test-power sentence is that the building already has high-voltage supply from its panel days, which it says fits outgoing test of AI servers. That is plant infrastructure, not NVIDIA's single-ended 800 VDC rail.

Customers: large CSPs as the main AI-server book, plus NeoCloud GPU and ASIC racks in small volume. Some high-ASP parts move to consignment from Q3 2026, project by project — GPU and ASIC both. Visibility is talked about through 2028.

Coverage note

One note, Q2 2026 (13 August). Earlier 2026 decks exist in the Chinese IR archive; this collection starts at the call that dated 2028 capacity and the Huaya plant.

Earnings calls