Business
AI servers now drive most of Foxconn’s revenue
AI servers generated 51% of Foxconn’s Q2 2026 revenue, but falling margins, supplier diversification and chip packaging limits cloud its growth.
Foxconn’s cloud and networking division generated 51% of company revenue in the second quarter of 2026, marking the first time its AI-server business has accounted for more than half of sales. As The Next Web reports, the shift has pushed the Taiwanese contract manufacturer further from its traditional dependence on consumer electronics, including the iPhone.
Foxconn’s smart consumer electronics division contributed 29% of revenue during the three months ending June 2026. Consolidated sales reached NT$2.53 trillion, up 41% year over year and 19% quarter over quarter, while net profit rose 35% to NT$59.97 billion (about $1.86 billion), a record for a second quarter. Earnings per share increased to NT$4.27, from NT$3.19 a year earlier.
The cloud and networking segment represented 48% of revenue in the first quarter of 2026, according to figures cited from Taiwan’s Central News Agency. First-half profit reached NT$109.89 billion, up 27%. Foxconn’s July 2026 revenue then climbed another 54.2% year over year to a record NT$946.5 billion, Bloomberg reported.
Foxconn bets on Nvidia’s next AI-server platform
Rotating CEO Michael Chiang told analysts that cloud investment “will become the most critical growth driver for Hon Hai over the next few years.” He described demand for AI infrastructure as steadier than the seasonal consumer-electronics business and said adoption by governments and corporations remains in its infancy, with cloud providers and AI labs currently the main buyers.
Foxconn is preparing for Nvidia’s Vera Rubin platform, whose racks are scheduled to enter mass-production preparation in the third quarter of 2026 and ship in the fourth quarter. Chiang expects Vera Rubin to become Foxconn’s major product in 2027.
The company is also expanding its manufacturing footprint to hundreds of sites across 24 countries, nearly twice its previous footprint. Planned or expanded capacity includes facilities in California, Texas, Wisconsin and Ohio—part of the broader push that has made Foxconn’s AI infrastructure business central to its strategy, alongside [its move to replace VMware with Arcfra for AI workloads]( /foxconn-arcfra-vmware-broadcom-samsung/).
That expansion does not guarantee rising market share. Morgan Stanley expects Foxconn’s high-end AI-rack share to fall to 39% in 2026, from 51% in 2025, according to the Wall Street Journal. The decline reflects customers seeking additional suppliers rather than weaker output. Microsoft, for example, is ordering hundreds of thousands of its own chips from TSMC, while other hyperscalers are pursuing similar strategies.
Analysts also worry that increasingly standardized AI-server designs could turn assembly into a commodity business. Chiang rejected that argument:
“Standardisation highlights the advantages of suppliers with scaled delivery and complete vertical integration.”
— Michael Chiang, rotating CEO, Foxconn
AI-server revenue is growing faster than margins
Foxconn’s revenue mix has changed more dramatically than its profitability. Gross margin fell to 6.12%, down 0.21 percentage points from a year earlier. Operating margin improved to 3.75%, up 0.60 points, while net margin slipped slightly to 2.37%.
The figures underline the economics of contract manufacturing: AI-server assembly remains a high-volume, single-digit-margin business. Foxconn’s next constraint may not be customer demand but CoWoS advanced packaging capacity, which is supplied by TSMC and is essential for the AI chips used in these systems. Market expectations call for CoWoS capacity to grow by more than 50% in 2027, but Chiang said the number of racks Foxconn can ship will “depend on chip supply.”
Investors have been less enthusiastic than the revenue figures suggest. Foxconn shares were up 17% in 2026, compared with a 57% gain for Taiwan’s broader index. The company has demonstrated that AI-server demand is real and that it can scale production; the unresolved question is whether customers will keep funding the buildout at its current pace.