Direct:
Two Taiwanese server ODMs used their May shareholder meetings to tell investors the AI server cycle still has years to run. Pegatron (4938.TW) on May 28 reiterated its tenfold AI server revenue target for 2026, with co-CEOs Gary Cheng and Johnson Teng outlining four AI server business lines they want to grow into. Wistron (3231.TW) on May 29 went broader, with chairman Simon Lin telling shareholders AI is still in an early ramp and could expand more than 10x from here, eventually accounting for over 10% of global GDP.
Pegatron's tenfold target was first floated on the March 11 earnings call, against a 2025 AI server base that is a small fraction of the company's roughly NT$1.12 trillion 2025 revenue. Texas production launched in Q1 2026 with Nvidia (NVDA) GB300 racks for Together AI and 5C, GTC 2026 brought Vera Rubin NVL72 and HGX Rubin NVL8 reference designs, and Q1 2026 earnings printed a 60%+ profit drop on consumer and PC weakness that AI server is supposed to absorb in the back half. The explicit ASIC and inference tracks that Pegatron has talked up since January matter, because those are the segments where Foxconn (2317.TW) and Quanta (2382.TW) have a head start and where custom silicon margins beat Nvidia reference pass-through.
On Wistron, the order book is where the numbers firm up. Server sales were about 70% of revenue in Q3 2025, mostly AI, and Economic Daily reported that Nvidia secured the available capacity at Wistron's Zhubei server plant through 2026, which neither company has formally confirmed. Lin paired the macro thesis with the previously announced one million GPU hours per year donation to Taiwanese and overseas startups and academic labs through Epoch Foundation's Garage+, which functions as goodwill and as talent pipeline development.
For anyone watching consumer SSD and memory pricing, ODM-side AI capacity is still ramping. That is the demand signal that has kept HBM and enterprise SSD allocation tight all year, and neither shareholder meeting suggests it loosens this calendar year.
Drafted with AI assistance against parallel reporting.
Sources:
- DigiTimes, “Pegatron expects tenfold growth in AI servers, defines four key business models” (May 28, 2026, paywalled stub: headline and lead only)
- DigiTimes, “Wistron says AI market could expand more than tenfold, pledges GPU hours to academia and startups” (May 29, 2026, paywalled stub: headline and lead only)
- Taipei Times, “Pegatron eyes 10-fold AI server growth” (Mar 12, 2026): tenfold target, GB300 platform, co-CEOs Cheng and Teng
- Pegatron Corporation, 2025 Annual Report (May 2026): NT$1,117.2 billion consolidated revenue
- DigiTimes, “Pegatron eyes triple-digit AI server growth in 2026, targets ASIC opportunities” (Jan 26, 2026)
- DigiTimes, “Pegatron eyes 1Q26 production launch for Texas plant with Nvidia GB300 servers” (Oct 29, 2025)
- PR Newswire, “Pegatron strengthens AI infrastructure collaboration with Together AI and 5C for Nvidia GB300 NVL72 and HGX B200 liquid-cooled rack deployment in US data centers” (2025)
- StorageNewsletter, “Nvidia GTC 2026: Pegatron unveils next generation AI platforms powered by Nvidia Vera Rubin NVL72 and HGX Rubin NVL8” (Mar 18, 2026)
- DigiTimes, “Pegatron 1Q26 earnings fall more than 60% amid off-season, eyes stronger 2Q26 on AI PC demand” (May 14, 2026)
- Taipei Times, “Wistron reports 109.94% revenue surge” (Oct 9, 2025): Q3 2025 server sales ~70% of revenue, mostly AI
- Tom's Hardware / Economic Daily, “Nvidia reportedly secures entire Wistron AI server plant capacity (Zhubei) through 2026” (not formally confirmed)
- DigiTimes, “Wistron launches AI computing power donation program, donating 1 million GPU hours annually” (Nov 3, 2025): Epoch Foundation's Garage+ partnership, 5,000-35,000 GPU hours per project
- DigiTimes, “Wistron chair sees AI growth entering 1.5 wave, believing AI bubble concerns still premature” (Feb 8, 2026)
Reddit: https://ift.tt/MFmsLjK