Foxconn's $95.4B quarter moves AI capex to the rack — and the 18x margin gap shows who really owns the trade

Hon Hai Precision Industry — the parent of Foxconn — reported unaudited Q3 2026 revenue of NT$3.03 trillion ($95.4 billion) on October 5, 2026, an increase of 47% year on year that cleared the LSEG SmartEstimate of NT$2.83 trillion by roughly 7%. September alone cleared NT$1.158 trillion, the first time any month's revenue has crossed the NT$1 trillion mark. The Q3 print is unaudited monthly revenue aggregated; the formal earnings release with margin and segment detail is due in mid-November.

Foxconn's Q3 2026 quarterly revenue is now within a rounding error of NVIDIA's most recent reported quarter. NVIDIA's fiscal Q2 FY2027 (period ended Jul 26, 2026) booked $96.22 billion in revenue, per the 10-Q filed Aug 26, 2026. Foxconn's Q3 2026, ending Sept 30, 2026, came in at $95.4 billion. Two different links in the AI supply chain, two different quarters, virtually identical revenue. That is the headline investors are still missing.

Cloud and networking — the segment that contains Foxconn's AI rack, server, and switching products — crossed 50% of group revenue for the first time in Q2 2026 at 51% of sales (NT$2.53T group; ~NT$1.29T segment), per the Aug 12 earnings release. Q3 segment data has not been disclosed yet with the unaudited monthly update, but at a similar or higher mix the segment now runs at roughly $48 billion per quarter, or $192 billion annualized. Consumer electronics (the iPhone business) is now the smaller, slower-growing half.

Foxconn is the rack-scale beneficiary of a hyperscaler spending wave that has reset the math on what 'AI capex' means. The four largest US hyperscalers — Microsoft, Meta Platforms, Amazon, and Alphabet — are guiding combined 2026 capex of roughly $700 billion or more, up 77–89% year on year. Microsoft alone is tracking around $190 billion in calendar 2026 capex, equal to nearly two Foxconn Q3 quarters.

That capex does not flow to chip-level silicon. The bulk is spent on data-center buildout: land, shells, power, cooling, and the rack-scale integration that Foxconn dominates. When Microsoft announces a $190 billion capex year, the value capture runs through rack integrators first, HBM and GPU suppliers second, and software last.

If Foxconn's Q3 is the order signal, the supply chain read-through is layered. Peer OEM Supermicro (US-listed, fiscal year ended Jun 30, 2026) competes head-to-head in NV AI rack assembly and trades on a fraction of Foxconn's scale ($39B TTM revenue vs Foxconn's $382B annualized Q3 run rate). Vertiv (Q2 FY2026, ended Jun 30, 2026) makes the power and thermal management gear that goes inside and around every rack — a smaller business ($11.5B TTM revenue) but with 19% operating margin and direct exposure to every Foxconn rack shipped. Arista Networks sells the 800G/1.6T switches that link those racks together inside hyperscaler fabrics; Foxconn's guidance that 800G+ switch revenue doubles in 2026 is the demand signal for ANET's silicon partners (Broadcom, Marvell).

Foxconn prints the revenue, but it doesn't print the margin. Group operating margin ran 3.67% in 1H26 — meeting but not crushing management's 3% target — and the rack business is fundamentally a contract-assembly model with thin economics. That is why the article does not put Foxconn Industrial Internet (Shanghai-listed subsidiary, 601138.SHH) in the related-stocks list: the value capture is upstream of the assembly line, not inside it.

Over the next one to two quarters, the catalyst is Foxconn's full Q3 earnings release in mid-November 2026 — that's when segment margins, AI rack units, and 800G+ switch revenue will be disclosed in detail, and where management commentary on Vera Rubin ramp and Q4 guidance will set the tone for NVIDIA's November print. Watch three numbers: AI rack unit growth QoQ, cloud/networking segment margin, and 2027 capex commentary.

Over one to three years, the structural question is whether the rack-scale layer can compress margins further upstream by integrating silicon or CPO switching. Foxconn's mid-2026 collaboration with NVIDIA on advanced packaging, and its CPO switch mass-production roadmap, suggest it is trying. If it succeeds, Foxconn's 3.7% margin becomes a 6–8% margin story. If it fails, Foxconn stays the world's most leveraged bet on hyperscaler capex with none of the pricing power — and the right trade stays NVIDIA, SK Hynix, and the US-listed rack-adjacent names.