Roundup

The Friday Briefing: Who gets the capacity?

TSMC’s record results arrived in a week of falling technology shares and fresh manufacturing commitments. There is plenty of money going into semiconductors. Where it goes matters rather more than the headline total.

James Creber

16 October 2026

Manufacturing & Assembly

Supply Chain & Trade Policy

Editorial illustration of a semiconductor factory inside an hourglass, with chips falling into the lower chamber against a textured orange background

AI-generated illustration: The Electronics Brief

It has been a strange week to follow the electronics industry.

Read the company announcements and there is money everywhere: new factories, reserved production capacity, expanding materials plants. Look at the financial markets and the mood is considerably less cheerful.

Thursday brought another technology-share sell-off, fuelled by concerns over inflation, interest rates and AI-related debt. TSMC, meanwhile, was reporting record quarterly profit.

You could spend a long time trying to reconcile those two things. Investors are judging what businesses might earn years from now, what they have borrowed and how much optimism is already reflected in their valuations. A chipmaker’s quarterly results tell us something more immediate about what customers have actually bought.

Neither provides a particularly satisfying answer if you are trying to get a product built at an acceptable cost.

For that, the detail matters. Which memory products are suppliers prioritising? Is the manufacturing process you need attracting investment? When does the new capacity become available, and has somebody already secured it?

This week gave us several useful answers. Some were buried well below the billion-dollar headlines.

TSMC has the orders

TSMC’s results on Thursday offered little support for the idea that demand for advanced chips has suddenly evaporated.

Third-quarter revenue reached US$46.72 billion, up 41.2% year on year in dollar terms. Net profit rose 68.7% in New Taiwan dollars to NT$763.06 billion.

Chief financial officer Wendell Huang pointed to “strong demand for our leading-edge process technologies”. Fourth-quarter revenue is expected to reach US$52.3 billion to US$54.3 billion.

The manufacturing mix was revealing, too. Processes of 7nm and below accounted for 79% of wafer revenue, with 2nm already contributing 8%.

That is a formidable quarter. It also needs to be read for what it is: evidence of strong business at the leading edge.

An industrial electronics manufacturer could reasonably welcome those results while recognising very little of its own experience in them. The products, customers and production economics can be quite different. There is room for a booming advanced-chip business and a much more hesitant market elsewhere.

Calling both “the semiconductor industry” is convenient. This week showed why it can also be misleading.

Memory buyers are feeling the competition

Memory is where those apparently separate markets start to overlap.

Samsung’s preliminary guidance, published on 8 October, put third-quarter sales at approximately KRW195 trillion and operating profit at KRW107.4 trillion. Those estimates cover the whole company, so they do not yet provide a detailed breakdown of memory performance.

The more useful purchasing context comes from TrendForce’s fourth-quarter outlook. Published on 30 September, it forecast conventional DRAM contract-price increases of 10% to 15% from the previous quarter and NAND flash increases of 15% to 20%.

Suppliers are prioritising advanced-process capacity for high-performance server products. Packaging and testing constraints are adding to the pressure, even while demand in several consumer categories remains weak.

That is an uncomfortable combination for customers outside the strongest markets. Softer demand for your finished product does not necessarily buy you cheaper memory.

A fresh TrendForce forecast on Wednesday added another wrinkle. It expects combined capital spending by nine leading cloud providers to exceed US$1.5 trillion in 2027. Part of that increase reflects higher server and component prices, along with advance payments to secure supply.

So even the spending figures require some care. More dollars can mean more equipment, but they can also mean paying more for it or paying earlier to obtain it.

For a smaller customer negotiating its next order, that competition is rather less abstract than the industry growth charts suggest.

The factories are coming. Some time is involved.

There was substantial news about future production this week.

On Tuesday, SEMI projected worldwide spending on 300mm front-end fab equipment of US$149 billion in 2026, up 31% year on year. It expects spending to reach US$227 billion in 2029.

Excluding facilities devoted purely to research and development, 20 new 300mm fabs or production lines are expected to begin construction this year, followed by another 18 in 2027.

GlobalFoundries supplied one of the week’s biggest individual announcements.

Under a US–Japan initiative involving up to US$21 billion in funding, it has been selected as technology provider and operator for two proposed factories at Malta, New York. An initial investment of approximately US$10 billion would support the first fab, expected to come online in 2029.

At full operation, that factory would add approximately 900,000 300mm wafers of annual capacity, serving markets including automotive, industrial and connected infrastructure.

The intended markets are worth noticing. Investment in those technologies matters to businesses whose products have little need for the latest processor node.

The timetable deserves equal attention. Anyone responsible for next year’s production plan will have spotted the gap between that requirement and a factory opening in 2029.

These projects are welcome. They are also a reminder of how slowly physical supply can respond compared with an order forecast. By the time a new fab is producing reliably, its customers may have revised their plans several times.

Semtech is securing its place

Semtech’s announcement with Tower Semiconductor was smaller, but arguably more revealing about how a supplier prepares for growth.

On Tuesday, the companies expanded their multi-year agreement, with Semtech investing in dedicated silicon-germanium capacity across Tower’s factories in the United States, Israel and Japan.

The process supports analogue components used in optical and copper interconnects, including transimpedance amplifiers, laser and modulator drivers and linear equalisers.

Semtech chief executive Hong Hou described the need for “long-term capacity planning and commitment with our manufacturing partners”.

There is a fairly direct commercial message in that. Winning designs creates a manufacturing obligation. Semtech is committing money to help meet it.

Power semiconductors brought another development on Monday. Polar Semiconductor and IC-Link by imec announced plans to bring imec’s enhancement-mode GaN technology into 8-inch production in Minnesota. The initial agreement covers devices from 40V to 650V, with early development access available through IC-Link.

These agreements tell us more than a generic promise of extra wafer capacity. They identify the technology and the route a customer can take into production.

They also show why capacity cannot be treated as one large, interchangeable pool. A factory can have available space and still be entirely wrong for the device you need to make.

Someone has to supply the slurry and test the chips

Fujifilm’s US$18 million expansion in Arizona could easily disappear beneath the larger announcements.

It deserves a look.

The Mesa investment, announced on Wednesday, will expand production capability for chemical mechanical planarisation slurries. These materials help create the flat surfaces needed during semiconductor manufacturing. The additional capability covers copper interconnects and other applications, including advanced packaging, with the expanded line scheduled to start operating in 2027.

A new fab creates demand for a considerable supporting industry. Materials suppliers have their own development work, production schedules and investment decisions. Their ability to keep pace is part of the expansion story.

The same applies to test.

On Monday, FormFactor and Teradyne announced a technology collaboration, accompanied by Teradyne’s agreement to make a US$300 million equity investment in FormFactor. Their work will cover test architectures and production readiness for applications including high-bandwidth memory, advanced packaging and silicon photonics. The investment remains subject to closing conditions and regulatory approvals.

Manufacturing an increasingly complicated device is only part of the job. Establishing that it works, consistently and at production volumes, requires investment of its own.

Neither slurry nor probe technology tends to dominate the wider conversation about AI. Both belong in the conversation about whether the promised hardware can actually be delivered.

Where does that leave the rest of electronics?

My reading of this week is that access to the right manufacturing capability is becoming a more explicit part of the product plan.

Some businesses are reserving it. Others are funding new facilities or building the materials and test capability around them. There is confidence here, although it is directed towards particular technologies and customers.

That leaves a more useful set of questions than whether electronics is having a good year.

What is being added? Who is it intended to serve? When can customers use it?

For a business trying to maintain a production schedule, those answers are worth considerably more than another impressive investment total.

There is plenty to welcome in this week’s announcements. Nobody should feel obliged to find their next supplier quotation reassuring as a result.

Something you might have missed: a humanoid heads for electronica

Mouser has offered an early glimpse of its electronica plans, including a live demonstration of RoboTUM’s humanoid platform.

Developed by the student-led robotics initiative at the Technical University of Munich, the robot combines embedded computing, EtherCAT networking, power electronics, sensing and energy storage. Its components include microcontrollers, NVIDIA Jetson modules, magnetic encoders and inertial measurement units.

Mouser will exhibit alongside TTI in Hall C3, stand 578, at the Munich show from 10 to 13 November.

It should give visitors something worth stopping for: the chance to see how the computing, control and sensing work together when the machine has to move.

We will be looking more closely at the technology heading to electronica over the coming weeks.

Sources

The Electronics Brief; primary company announcements, SEMI and TrendForce