There's a Whole Lot of Taiwanese AI Winners Not Called TSMC
[Opinion] A deep supply chain highlights just how much the global AI expansion depends on this one island nation.
Good Evening from Taipei,
Everyone knows the name Taiwan Semiconductor Manufacturing Co. It’s the world’s most important company. TSMC’s position atop the global supply chain comes from its ability to do what no other organization in the world can do.
Revenue figures for Taiwanese companies point to another stark fact: the rest of the country is just as important to the global economy as TSMC, and equally irreplaceable.
Sales across the Taiwan Stock Exchange’s almost 1,000 listed stocks climbed 27% in US dollar terms for the first two months of this year. That’s not just a handful of blockbuster companies, but the average across the entire bourse including makers of chips, clothes, chemicals, rice cookers and cars. By comparison, sales for S&P 500 member companies is expected to climb 8.5%.
Taiwan is an export-driven economy. Its biggest companies by revenue are all tier-one exporters, meaning their business comes from selling stuff overseas. If these companies are enjoying massive growth then it’s because their foreign clients are desperately in need of the products they make. A decade ago, the world needed smartphones, a decade prior it was laptop computers. Supply chains for both were dominated by Taiwanese companies — with an assist from Chinese factories.
Right now, the world craves more computing systems to drive the irreversible rollout of artificial intelligence. Once again, it’s Taiwanese businesses that command the end-to-end supply of chips, circuit boards, server racks, cooling systems, and final assembly. Note also that very few Chinese companies are big players in the AI hardware space.
It’s easy to think this is Nvidia’s show, with TSMC and Foxconn merely playing supporting roles.
There’s another way to look at it, though. Nvidia is without a doubt the name to watch in AI compute, but it’s not doing it alone. In reality Nvidia makes nothing. Its chips and servers are the product of an entire ecosystem of companies that own the factories, manage the components, and handle the logistics. They’re mostly Taiwanese.
Heading the list by revenue growth is Wistron and Hon Hai (aka Foxconn) which sold $10.3 billion and $8.4 billion more in product respectively than they did in the first two months of last year. Remember, that’s additional revenue and it comes almost entirely from AI servers and the parts which go into them.
Wistron brought in $16 billion of business over the past two months largely because of fresh AI orders. The Taipei-based company, a spinoff from PC brand Acer, last year opened a new factory 10km from TSMC’s HQ just to make AI servers. Capacity was sold out almost immediately and almost entirely to make systems for Nvidia, the Economic Daily reported. It’s no less astounding that Foxconn, whose revenue topped $260 billion last year, can still manage 25% growth.
“It’s looking like a very good year,” Foxconn Chairman Young Liu told reporters last week. “Consumer electronics is looking very good because of smartphones, and AI is also looking very good.”
Top-line numbers for assemblers, also called contract manufacturers, is a little misunderstood. A large chunk of the $42 billion in revenue Foxconn posted for January and February was the resale of components that clients like Apple, Nvidia, Cisco and Sony ordered.
Although specifics vary from company to company, and product by product, the business model generally entails the client selecting parts and then having the contractor buy those components on its own dime. They’re then assembled into the devices the manufacturer was contracted to make. Sometimes the contractor can mark up the cost, other times it buys at the contracted rate but charges an agreed upon fee per unit of final product. Sometimes it’s a combination of both.
That’s why watching these companies is so useful. When contract manufacturers like Quanta, Hon Hai and Wistron are adding $2.5 billion to $8 billion in revenue per month, you can be sure their clients and the rest of the supply chain are also going strong. The caveat here is that escalating inventories can be a bad sign. These stockpiles are built based on expectations of future demand. They can usually be digested when the market slowly turns, but a one-off shock like the 2009 financial crisis can cause suppliers and assemblers a world of pain.
What makes the Taiwan supply chain run deep is the plethora of companies that provide esoteric and specialized products. Take Favite, a maker of automated optical inspection equipment, which posted $8.9 million in revenue over the past two months. That doesn’t sound like much, but it’s a whole heap more than the $133,000 it posted a year earlier. That sudden boost comes from its privileged position as a supplier of test equipment for TSMC’s future packaging technologies including next-gen Chip-on-Wafer-on-Substrate (CoWoS) and Chip-on-Panel-on-Substrate (CoPoS).
Another AI winner is United Integrated Services, which posted a 64% jump in sales. On paper, Taipei-based UIS is just a construction company, and not really an exporter. But its biggest client is TSMC, and it recently won contracts to manage the development of upcoming N2 wafer and CoWoS packaging fabs in Taiwan, as well as the chipmaker’s Arizona expansion.
Then there’s Winbond and Nanya Tech, among Taiwan’s few surviving memory-chip makers. Although not major suppliers of the high-bandwidth memory that’s fused with Nvidia and AMD GPUs, both companies are winning from higher DRAM prices and increased demand for faster DDR that goes into both AI servers and edge devices.
There are too many fascinating AI-connected suppliers for me to list them all, but if you want some extra reading then I suggest looking up Accton, Gigabyte and HiWin.
Next week I’ll show you how to use Taiwan monthly sales data to track trends in various industries. Make sure you’re subscribed.
Thanks for reading, and please take the time to fill out the survey.
One interesting element you could add to your analysis is the test & metrology ecosystem around AI semis.