Hon Hai Technology Group, better known as Foxconn, is centering growth on the Foxconn AI server market as chairman Young Liu says a deeper OpenAI collaboration is taking shape. On Friday in Taipei he told reporters he had met OpenAI chief executive Sam Altman earlier that day, and he told CNBC the company’s global AI server share could reach 50% within a couple of years. The remarks come as technology stocks face pressure over whether AI infrastructure spending can continue.
Key Takeaways
- In Taipei, Liu said he met Sam Altman earlier that day; terms are unfinished.
- Foxconn’s AI server share, about 40% now, could hit 50% in a couple of years.
- Capacity is fully booked for at least the next year and a half.
- Cloud and networking reached 51% of second-quarter revenue for the first time.
Foxconn AI Server Share Could Reach 50%
In an interview with CNBC’s Christine Tan, Liu said Foxconn’s share of the global AI server market stands at around 40% and could rise to around 45% next year.
“I think we can grow up to 50 percent,” Liu said.
Formally known as Hon Hai Precision Industry, Foxconn remains a major assembler of Apple’s iPhones, but AI infrastructure is now a key growth driver. Liu expects it to account for 60% to 65% of revenue in the next three to five years, excluding edge AI devices that run algorithms on local hardware.
In the second quarter, its cloud and networking division, which includes AI servers, accounted for 51% of revenue, the first time above 50%, Finimize and The Next Web reported.
Capacity is fully booked for at least the next year and a half, he told CNBC. Demand so far has come largely from AI model developers and cloud service providers, while governments and enterprises have yet to fully embrace the technology. He added that AI demand remains “quite strong.”
| Metric | Cited figure |
|---|---|
| Current global AI server share | Around 40% |
| Potential share next year | Around 45% |
| Share within a couple of years | Up to 50% |
| Cloud and networking, second quarter | 51% of revenue |
| AI infrastructure, next 3 to 5 years | 60% to 65%, excluding edge AI |
| Planned AI spending per year | $3 billion to $5 billion |
New OpenAI Deal Still Awaits Terms
At Foxconn’s annual technology event in Taipei, Liu said a new partnership is in the works and would be announced once terms are finalized, The Next Web reported, citing Reuters. He gave no details. OpenAI has not commented.
The talks would build on a November 2025 collaboration to design several generations of AI data center racks. Planned U.S. components included:
- Cables
- Cooling systems
- Power equipment
That deal included no purchase commitments. OpenAI received early access to test the hardware and an option to buy it.
Foxconn is already Nvidia’s biggest AI-server manufacturer, so the talks matter beyond one purchase order. Analysis: early access and a buy option are not shipments. The rack talks are separate from OpenAI’s move to train frontier models without live internet.
Houston Plants, Humanoids, and the Supply Chain
“What we are going to do is we’re going to move up the supply chain,” Liu told CNBC. A broader U.S. presence, he said, would let Foxconn respond more quickly to rising demand. Higher manufacturing costs mean an initial focus on stronger-margin products, including AI server racks and data center equipment.
Greater scale, he said, could eventually help lower manufacturing costs. The company is looking to expand six factory buildings in Houston, where it makes AI servers for Nvidia, The Next Web reported, and Finimize said it expects to keep expanding.
Foxconn Industrial Internet has deployed more than 10 humanoid robots for internal use in Texas, with robot brains it is developing on top of Nvidia’s GR00T foundation model, Finimize reported. Finimize described those lines as a way to learn which tasks robots could standardize, such as handling parts, loading and unloading, and repetitive assembly. It recently partnered with Schneider Electric on AI infrastructure equipment.
Analysis: if those robots raise the share of units that pass inspection the first time, output can grow without labor and rework costs rising one-for-one. That test matters while financing costs stay high, a question also taken up in how a SpaceX GPU debt deal closed as credit access shut off.
Liu Rejects Bubble Talk, Then Sets a Limit
Liu pushed back on concerns that the AI industry is heading for a bubble, The Next Web reported, citing Reuters. Technology stocks have come under pressure in recent weeks as investors question whether enormous AI infrastructure spending can continue, particularly as borrowing costs rise. Foxconn expects double-digit growth next year.
“Overinvestment will kill the company,” he told CNBC.
He said Foxconn would keep a tight rein on spending. Liu said the company could spend more than its planned $3 billion to $5 billion a year on AI, depending on how far vertical integration expands.
Weaker demand from one customer could be offset by growth elsewhere. An industry-wide slowdown would be a much bigger challenge for Foxconn and the broader AI supply chain.
Analysis: a share near 40%, a path toward 50%, and capacity booked for at least a year and a half show demand already in the factory book, with the risk concentrated in model developers and cloud providers.