At a glance. Cloud and networking crossed half the company for the first time, at 51%. Vera Rubin finally got a ship date: production-ready in Q3 2026, shipments in Q4, with a 50% share target. ASIC racks were given the same 40%+ share ambition. The call still did not say 800 V.
What changed this quarter
The mix flipped. Cloud and networking 51%, consumer electronics 29%. Two years earlier those weights were reversed.
Vera Rubin was dated. Management held to Q3 2026 for production readiness and Q4 for the first shipments, with 2027 as the mainstay year. Final volume depends on chip supply. The GB platform was said to last at least through 2027.
ASIC was given a share target, not just a growth rate. GPU share is already high; ASIC share is now aimed at 40% or more as well. ASIC was about 10% of AI-server revenue in 2025 and a little higher in the first half of 2026.
The numbers
| Q2 2026 | Value | Change |
|---|---|---|
| Revenue | NT$2.53tn (~US$80bn) | +41% YoY, +19% QoQ |
| Gross profit | NT$154.5bn (~US$4.9bn) | +36% YoY |
| Operating profit | NT$94.8bn (~US$3.0bn) | +68% YoY, a quarterly record |
| Net income | NT$60.0bn (~US$1.90bn) | +35% YoY |
| EPS | NT$4.27 (~US$0.14) | +34% YoY |
| Gross margin | 6.12% | slightly down on mix |
| Operating margin | 3.75% | +0.6pp YoY |
First half 2026: revenue NT$4.65tn (~US$148bn), +35%; operating profit NT$170.5bn (~US$5.4bn), +65%; operating margin 3.67%; net income NT$109.9bn (~US$3.5bn), +27%; EPS NT$7.84, +26%; ROE 6.21%, +0.73pp.
Cash and deposits NT$1.52tn (~US$48bn); net cash NT$219.8bn (~US$7.0bn). First-half operating cash outflow NT$69.1bn and free-cash-flow outflow NT$150bn (~US$4.8bn) as receivables and inventory rose with the AI ramp. First-half capex NT$80.9bn (~US$2.6bn), +5%, with the full year still guided more than 30% higher than 2025.
Segment mix: cloud and networking 51%, consumer electronics 29%, computing 15%, components 5%.
NVIDIA 800V read-through
This is the first Foxconn call with a dated NVIDIA next-gen rack, and it is still not an 800 V call. Vera Rubin is the 2026 cabinet. NVIDIA's 800 VDC architecture is written against Kyber in 2027. Treating Q4 2026 Rubin shipments as 800V evidence would be a misread.
What the call actually adds to the 800V file:
- Self-made content above 50%, chips excepted, from Level 1 components through Level 11 system assembly. That is the same 50–60% target from 2025, now stated as already there, and named as the barrier to entry as racks standardise.
- Capex is going into servers, racks, liquid cooling and test, in Taiwan, the United States, Mexico and Vietnam — the physical plant an 800V cabinet will use, even if the voltage has not been discussed.
- CPO switches have started to ship. LPU cabinets and Vera CPU racks enter production in Q3, ramp in Q4, and scale in 2027; lower ASP, so a limited P&L effect.
- Analysts again asked what Foxconn supplies into TPU. Management would not provide a TPU bill of materials. It answered with the ASIC share target.
Foxconn remains a customer and a competitor to the Taiwan component names in this collection. The 50%+ self-made figure is why.
Segments that moved
Cloud and networking at 51% is now the core of the P&L. Q3 AI-rack shipments are guided to a high-double-digit rise versus Q2; the full year is still a multiple of 2025. ASIC projects, on the back of the GPU position, are expected to grow substantially in 2026. Switches at 800G and above are guided to double; CPO is in production.
Consumer electronics at 29% is the seasonal second-half engine, guided to significant sequential and annual growth in Q3.
Computing at 15% is guided slightly down sequentially in Q3 as the new-product peak passes and memory stays tight, but still significantly up year on year on component prices.
Components at 5% — camera modules, connectors — is guided to significant annual growth in Q3. The reported share is small because so much CMM is internal.
Guidance
- Q3 2026: significant sequential growth, strong annual growth.
- Full year 2026: strong growth held. 2027 is described as still growing.
- Operating margin: first-half 3.67%; the full year is expected to beat 2025's 3.2%, with 3% still the medium-term floor.
- Capex: more than 30% up, no cash equity raise, funded from internal cash flow and debt. R&D tied to the 3+3+3 agenda is about NT$20bn (~US$0.64bn), +40%.
- Order book: 2027 orders already seen moving higher. New cloud operators and AI-model companies are expected to contribute by the end of 2026.
Management Q&A
Q: How will the next-generation AI rack affect second-half shipments, and how does that change the competitive landscape?
A: Mass production of the next-generation AI rack starts in Q3 2026, and GB-platform demand is expected to last at least through 2027, so the sequential rack-shipment climb is unchanged. Management argued that standardisation actually favours suppliers that can deliver at scale with full vertical integration. Foxconn's offering is one-stop from Level 1 components through Level 11 system assembly, with self-made content already above 50%, so barriers only rise and share on the new platform is expected to increase further.
Q: Capex keeps rising — what is the return on that spend, and will the company raise equity?
A: Earlier capex is already showing up in scale and profits: first-half operating profit was up NT$67.4bn and EBITDA up NT$77.6bn. Full-year capex is still expected to rise more than 30%. Funding follows match short-to-short and long-to-long, mainly from internal cash flow and competitive debt; there is no cash capital increase planned at this stage, and net cash stands at NT$219.8bn.
Q: How should second-half operating margin be read, especially as high-ASP AI racks mix in — can 3% still hold?
A: Mix will move the percentage, but Foxconn is using design-for-automation, standardised modules, and a flexible buy-and-sell versus consignment mix to protect profitability. Highly customised ASIC programmes are mostly consignment, which helps asset turns. First-half operating margin was 3.67%; the full year is expected to beat 2025's 3.2%, and the medium-term 3%+ floor is unchanged.
Q: What does Foxconn supply into TPU, and how should general servers be viewed?
A: It would not comment on a single customer or product, but said 2026 ASIC programmes have added new customers and projects, so ASIC share should rise substantially alongside an already-high GPU position; Foxconn offers both GPU and ASIC solutions. AI is also lifting general-server demand: the second half is guided to double-digit growth versus the first half, visibility is better than three months earlier, and industry CPU shipments are cited at about +16% in 2026 and possibly +23% next year — Foxconn thinks it can grow faster than that average.
Q: What is the scale and margin of networking (switches, CPO), and has CSP capex revision improved order visibility or brought in more new-cloud customers?
A: First-half switch revenue grew hard on both volume and ASP; CPO switches have started to ship, mainly for scale-out between racks, with scale-up later, and optical communications is described as the next AI-infrastructure growth wave, with a market that could exceed US$100bn. CSP investment has lengthened visibility, with 2027 orders still moving higher; besides large CSPs, more new-cloud operators and AI-model companies are adding demand, and that contribution is expected before the end of 2026.
Q: Will GPU servers move to consignment, and what share of first-half AI-server revenue came from ASIC?
A: Consignment is mainly used on ASIC programmes today and may be used on GPU programmes later, depending on the customer; the mix is assessed deal by deal, and the consignment share is expected to rise. ASIC was about 10% of 2025 AI-server revenue and a little higher in the first half of 2026, though consignment means reported sales understate the real book; the ASIC share target is 40% or more, matching the GPU band.
Q: What is the Vera Rubin volume ramp and the share target?
A: The plan is unchanged: the next-generation AI rack enters production readiness in Q3 2026 and starts shipping in Q4, becoming the mainstay in 2027. Foxconn is still working toward a 50% share target. Next year's CoWoS capacity is expected to grow more than 50%, but how many racks that becomes still depends on chip supply.
Q: When do standalone LPU or Vera CPU racks go to volume, how do they compare on gross margin, and how does the AI-platform strategy show up in the P&L?
A: LPU and Vera-related products are scheduled to enter production in Q3 2026, ramp in Q4, and reach scale in 2027; unit prices are lower than GPU AI racks, so the effect on group revenue and gross margin is relatively limited even as volume grows. The AI-platform shift is meant to lift internal efficiency and profits through smart manufacturing (including the Genesis platform) on one side, and to create external revenue from platforms such as smart EVs and smart cities on the other.
Q: The Sharp investment is ten years old — how are the next ten years positioned?
A: After a decade of restructuring together, the next ten years are aimed at new growth with Sharp in AI, robotics, EVs and smart living, with the goal of making Sharp a brand Japan is proud of and the world watches again. Sharp's first-generation AI server, launched the previous week, was cited as a joint product and a template for more of the same.
Disclaimer
English notes on Hon Hai Precision Industry's investor conference of 12 August 2026, covering Q2 2026. Figures are as presented by management and have not been independently verified. USD equivalents are approximate at about NT$31.5 to the dollar. For reference only; not investment advice.