Foxconn wants to build half the world's AI servers within a couple of years, giving the Apple supplier a much bigger role in the equipment powering the AI boom.

Chairman Young Liu puts its current share at about 40%, with roughly 45% possible next year. His 50% goal applies specifically to the global AI server market, rather than every part of AI infrastructure.

Speaking to CNBC's Christine Tan in an interview published October 30, Liu said:

I think we can grow up to 50 percent.

He said production capacity is booked for at least the next year and a half. He expects AI infrastructure to bring in 60%-65% of company revenue within three to five years, excluding AI devices that run models locally. Those are management forecasts.

AI demand is already lifting sales

The company's financial reports show why servers have become such a priority. Its October 5 report put third-quarter revenue at NT$3.03 trillion, up 47.12% from a year earlier and 20.44% from the previous quarter. That was a record for the July-September period. The figures are unaudited.

Cloud and networking products led the quarterly growth ranking, ahead of components, computing products and smart consumer electronics. The report credited demand for AI products for the cloud business's growth.

September alone generated NT$1.16 trillion, the first month revenue crossed NT$1 trillion. Consumer electronics benefited from a new product launch and grew strongly from August, but sales were roughly flat compared with September last year. Cloud and networking revenue grew strongly on both comparisons.

For the first nine months, revenue reached NT$7.67 trillion, up 39.53% year over year. These are group sales, covering much more than AI servers. Foxconn has not presented those totals as AI infrastructure revenue.

The same report said AI-related operations were expected to keep growing in the fourth quarter, alongside the traditional second-half peak for information and communications products. It also warned that political and economic uncertainty still needed watching. The company lists November 12 as the date of its third-quarter financial-results investor conference call.

Building AI systems and using them

The manufacturing push extends into the company's own factories. In an October 26 announcement, Foxconn described using Nvidia computing equipment to train models on industrial data and use AI agents across more than 200 facilities.

Its FoxBrain language model adapts existing open foundation models for industrial work. It supports Genesis, a manufacturing platform, and CityGPT, which handles urban services. Nvidia DGX SuperPOD provides the training infrastructure.

The company reported training 120-billion-parameter models 2.48 times faster than on its previous H100-based systems, processing one trillion tokens in 12 days. These are company-reported results from its own setup.

It also reported a 25% reduction in labor hours spent on production scheduling and 50% less engineering work for injection-molding process changeovers. On surface-mount production lines, it said it had addressed more than 20 major causes of defects, with a correction rate above 90%.

Other tools simulate factories and robots before putting them to work, helping the group reproduce production lines in different locations. Through subsidiary Visionbay.ai, it is also offering AI services to enterprises and governments across Asia using Nvidia-certified systems.

More U.S. manufacturing, with a spending warning

Liu also wants to expand U.S. production further up the supply chain. Higher costs mean the initial focus will be products with stronger margins, including AI racks and data center equipment.

Annual AI investment could exceed the planned $3 billion-$5 billion depending on how much of the supply chain it takes on. But he warned:

Overinvestment will kill the company.

He said demand from other customers could offset a pullback by one buyer. A slowdown across the industry would be a bigger problem.