Foxconn made its name assembling Apple's devices. Now a meaningful share of its sales comes from building the servers that run artificial intelligence. On Thursday, Nov. 12, it said that business is still strong. It also named a few things that could slow it down.
The Journal's Sherry Qin reported the numbers. Foxconn's profit rose 34% to US$2.36 billion in the three months through September. Revenue rose 46%. Cloud and networking gear now brings in more than half of its sales. Foxconn also earns a meaningful share of its revenue building AI servers for leaders such as Nvidia (NVDA). Foxconn expects shipments of AI server racks to grow "in the high double digits" next year. It was careful too. It cautioned about uncertainty after the U.S. midterm elections. Chairman Young Liu said tight parts supply and shaky geopolitics could make next year less certain. The Journal added that worries about a data-center slowdown have rattled markets.
A second report points the same way. Semiconductor Manufacturing International, or SMIC, is China's largest contract chip maker. The leading chip makers are busy with AI chips, so orders for simpler chips are landing at SMIC. Its revenue rose 34% to a record $3.19 billion. Profit was $428.4 million, up from $191.8 million a year earlier. Taiwan Semiconductor (TSM), known as TSMC, still leads. TrendForce data give it more than 70% of the market for making chips that other companies design. Even so, SMIC's Hong Kong shares lost 32% in the third quarter. They've lost another 20% so far this quarter.
Where the limits are
The first limit is cost. Meta Platforms (META) gives away an AI agent called Muse. Deutsche Bank puts the capital cost at $25 to $50 for each user. Vals AI, a testing startup, says a heavy user could cost more than $50 a month. The second limit is room to build. Americans voted for data-center moratoriums last week, the Journal wrote. We covered that vote in our Nov. 5 piece on the data-center backlash. Meta and OpenAI executives have since said building will get harder. Kansas's governor-elect, Cindy Holscher, promised to halt new data centers. The third limit is money. Rising interest rates are lifting the cost of borrowing for the build-out.
Our read
This is an Investments/Risk (IN04) story. A company's results and its share price can go different ways. SMIC had record revenue and a falling stock in the same season. The AI supply chain is the line of companies that make the chips, the servers and the power. We'd own that chain on purpose, across several links. We wouldn't chase the one name in the news. In our model books TSMC is held at weight in 30 books and Vertiv (VRT) in 20. On the power side, GE Vernova (GEV) is in 30 books, NextEra (NEE) in 21, Vistra (VST) in 17 and Quanta (PWR) in 14. The North Star books added about half a point to GE Vernova, NextEra and Vistra on Friday, Nov. 6, after the open. That was the desk's decision for its model books. Nothing was added in this edition.
Here's where it touches your money. Open your 401(k) or brokerage statement this week. Write down the AI names you own, funds included. Then count the links. Chip makers, server builders, power companies and software firms are different links. If every name does the same job, one limit can hit them all at once. A limit on new data centers can still reach every link, so also add up how much of the account rides on AI in total. These are general planning principles, not advice. If the list looks lopsided, bring it to a review. It's easier to check the spare tire in the driveway than on the highway.
