Foxconn expects double-digit sales growth in 2027 and sees no bubble in artificial intelligence, its chairman said, pushing back at the doubt that has pulled down technology shares and complicated the financing of the data centers its customers are building.
Young Liu, who chairs the company formally known as Hon Hai Precision Industry, told reporters at its annual technology event in Taipei on Oct. 30 that demand from the customers it serves was clear and strong. He said he had met OpenAI chief executive Sam Altman that morning and that an announcement on a collaboration would come once details were settled.
The event, now in its seventh year, is where the company announces partnerships and shows the machinery it has built for customers. Foxconn has no consumer brand of its own, so the audience it is addressing is the companies that buy its capacity.
The remarks landed at an awkward moment. AI-related shares have fallen as investors question whether spending on chips, servers and power can keep rising while margins narrow, cash flow slows and borrowing costs climb. Several companies tied to the buildout have reported results that disappointed, and the selloff has spread from the pure semiconductor names to the contractors that assemble their hardware.
Foxconn sits close to the center of that chain. It is Nvidia’s largest maker of AI servers, and its cloud and networking products, the segment that includes those servers, accounted for 51% of second-quarter revenue, the first time the unit has passed half the total. The company’s growth now depends on a business that did not exist at scale four years ago.
Liu said Foxconn has six factory buildings in Houston, where it builds AI servers for Nvidia, and that it will need more. Expansion in Texas has become a political asset as well as a capacity decision, since the servers assembled there can be sold to American cloud operators under rules that favour domestic production.
The company’s history explains both the confidence and the caution. Foxconn assembled iPhones for Apple and grew into the world’s largest contract electronics manufacturer by taking the least glamorous part of the business, the assembly, and doing it at a scale nobody else could match. That model produced enormous revenue and a share price that tracked its largest customer’s product cycle. AI servers are the first product in a decade that has given the company a growth story of its own.
Competition for those orders is crowded. Quanta Computer, Wistron and Inventec all build AI servers, and the cloud operators that buy them have an interest in keeping at least two suppliers qualified for every rack design. Foxconn’s advantage is its ability to build at volume in multiple countries, including the United States, at a moment when customers want their hardware made outside China.
The constraint the company describes is supply, not demand. Memory chips, printed circuit boards, power equipment, utility connections and lasers are all short, and the delays and cost increases pass down the chain to the assembler. For a business built on volume and thin margins, missing components are a more immediate problem than missing orders, because a server that cannot be completed cannot be invoiced.
Foxconn’s own results show how thin the cushion is. Contract manufacturing earns a few cents on each dollar of revenue, so a rise in component prices compresses profit unless it can be passed through, and cloud customers with multiyear contracts are hard to pass costs to. Its gross margin has historically sat in single digits, a fact of life for a company that competes on volume.
Liu also used the event to describe a bet on robots. Foxconn Industrial Internet, the Shanghai-listed unit, has put more than ten humanoid robots to work inside its Texas facilities, according to Brand Cheng, its chairman, and plans to extend the deployment beyond Texas. The company showed two wheeled models in October, one that lifts loads vertically and one that bends at the waist for assembly and loading work, and it is developing the control software on Nvidia’s GR00T foundation model.
The logic is the same as the one that built Foxconn’s electronics business: if labour is the constraint in American manufacturing, and the customer wants production in America, then the answer is machines that do not need visas. Whether the economics work at scale is unresolved, and the company has not said what the robots cost or how many it intends to deploy.
Analysts said the chairman’s optimism should be read alongside the guidance he did not give. Double-digit growth for 2027 is a wide range, and Foxconn has historically declined to break out AI server revenue in absolute terms, which makes it difficult to check the claim against the order book. The company’s monthly sales reports are the closest thing to a real-time gauge, and they have been volatile as the mix shifts toward rack-scale systems that ship less often but at far higher value.
There is a structural argument on Liu’s side. The companies building AI infrastructure are not buying on hope; they are buying against contracts with cloud customers, and the power plants and substations being ordered now will take years to connect. A boom financed by debt and depreciating assets can still end badly, but it does not end because sentiment turns.
There is also a structural argument against him. Every generation of hardware buildout has overshot, and the assembler at the end of the chain absorbs the write-downs last and hardest. Foxconn’s customers can cancel orders; Foxconn’s factories and robots cannot be unbuilt.
Liu’s answer, in effect, is that the shortage of components is the proof. A bubble does not produce a queue for memory chips.
