There was also much to say about AI at TSMC’s annual Technology Symposium in Amsterdam. However, the Taiwanese chip giant wants to look beyond the current hype, with robotics, a new design center in Germany, and (much) more efficient chip process nodes.
TSMC began its presentation as you would expect from a chip company that is flourishing largely thanks to mega-investments in AI. It stated that 40 percent of its work activities are being transformed by technology, while 15 percent of daily decisions at work are being taken over by AI. However, Kevin Zhang, SVP Business Development & Overseas Operations Office at TSMC, is more concerned with improvements in the here and now that go far beyond AI.
No fuss, just secrets
Now that we have visited TSMC a few more times, it is clear once again that the company has a unique vision that it continues to espouse. What other company would start a series of slides with the comment “what you see here is boring, but predictable”? Zhang is referring to TSMC’s largely stable roadmap for the coming years. We say largely, because a new process node has been added: 14A. In 2028, the latest iPhones will probably be manufactured using this technology, which will deliver 10-15 percent higher performance for the same usage compared to 2 nanometers (or N2, as TSMC calls it, given the nanometer measurement means nothing anymore). It will also save 25-30 percent in energy consumption at the same performance level. The transistors and other logic will also be placed roughly 20 percent closer together. This is compared to N2, which has not yet been used for mass production.
N2, planned for the second half of 2026, in turn offers an 18 percent performance gain over the current N3 node. It promises 36 percent lower power consumption and a chip built 15-20 percent closer together. Once again, Apple is likely to be the first to adopt this technology, as it was with N3. In addition, advanced packaging is increasingly being used to build chips in 3D. Silicon photonics is also being mentioned more often than before, which involves chips communicating internally via light signals instead of copper.
TSMC has plenty of copper, by the way. According to Zhang, the many copper cables in data centers are something to be “horrified” by. This method of communication is expensive and extremely inefficient, he says. That is why light must also be used here. This will come in handy for large-scale AI workloads with multiple GPUs, to name one use case.
Greater ambitions
Zhang does not stop at these rather technical details. For people with less sharp eyesight than ASML machines, developments in chip design are already clearly visible. Where current GPUs combine chip components that are already 3.5 times the size of what a chip machine can manufacture, some AI players are already moving to 7 times or even 40 times this size. In short: giant processors that need to be cooled better than ever, which the new chip nodes will counterbalance by improving efficiency.
There is “strong AI data center demand,” according to Zhang, which, oddly enough, only became really noticeable to TSMC in 2024. That’s just how slow the chip supply chain moves. According to him, advanced packaging is so well known and popular (as in Nvidia’s Blackwell chips) that CoWoS (Chip-on-Wafer-on-Substrate) can apparently be explained by any random Taiwanese person on the street. We’re not entirely sure if this is meant sarcastically, but we wouldn’t be surprised if it’s simply true.
There is also mild growth in chip production for smartphones, PCs, and IoT devices. TSMC is repeating a claim we heard from Samsung earlier this year: the premium category is growing relatively quickly, while the rest of the phone market is lagging behind. Efficiency is also a priority for the most expensive phones; the latest chip processes are not intended to load apps even faster, for example.
Otherwise, there are no surprises in the overview of the chip sector. The automotive sector is still “soft” (exactly the same term used as at an earlier TSMC event), even though car manufacturers are moving toward increasingly advanced chip processes to bring their “software-defined vehicles” to life.
Tariff barriers are “noise”
What is striking is that TSMC is predicting uncomplicated 10 percent growth. This is remarkable because, even after questions on the subject, Zhang considers tariff barriers to be nothing more than “noise” at this point. Of course, TSMC does not have a say in these geopolitical policies, so it can only deal with the reality it’s facing. The company’s forecast can therefore only be made based on what can be estimated, and the mood in Washington is not part of that. In any case, TSMC has not seen any real change in demand as a result of these tensions.
Although TSMC is expanding significantly in Arizona (with six factories in the pipeline), Zhang is positive about Europe. According to him, the chip company is certainly willing to meet European demand, although he first wants to see the German factory being built. The ground was only prepared for the first time last year. Another achievement is that a design center will be established in Munich, Germany. TSMC hopes this will attract the large pool of talented personnel within Europe. It is good to hear some positive news about the European chip industry, and this is a (albeit cautious) positive story.
Robotics
In recent years, we have heard a lot about robotics from Nvidia. It seems that the GPU manufacturer is already thinking about what should follow the AI hype: robots in every factory with human-like skills and sometimes even human-like appearances. TSMC, which is less fond of marketing talk than Nvidia, is also talking more about robotics than ever. According to Zhang, this is also out of necessity. According to him, TSMC itself does not have enough staff to keep all its chip factories running now and in the future, so it needs advanced robots to reduce the workload. The same applies to TSMC’s customers. So if the focus on AI wanes, no one should be surprised if a robot craze follows.
The latest nodes are not enough to power robots, the fastest AI chips, and automotive technology. Another transistor revolution is needed. Once a PowerPoint slide, now a reality in test labs: the CFET. This Complementary Field-Effect Transistor can be described as a stacked transistor. It doubles the density of computing power. However, we will have to wait a while longer: TSMC has not provided a timeline for this. This shows that the chip manufacturer continues to innovate even in the smallest domain.
Conclusion: (dis)continuity
To the casual observer, TSMC’s output may not seem particularly relevant. After all, it is the software that organizations run on a daily basis that matters on that same daily basis, and the exact hardware is almost never really important. The exact construction process of these processors is a further step away from everyday reality: why do we need to know about its inner workings? This is all somewhat true, but TSMC is a critical component in turning all the big promises of the vendors into reality. More efficient, smaller, and smarter processors are needed to maintain progress in the tech world, and this Symposium once again shows us what the coming years have in store. In other words: what’s announced here gives a pointer to what’s possible in the first place. Without TSMC to create the chips, there’s no AI boom, just like Nvidia, ASML and the various server builders are critical links in the chain.
However, TSMC is above all a stable player with little interest in disruption. It works with partners in an ecosystem to develop smaller chips, higher production yields, and new ways to circumvent the laws of physics. For the time being, the economic turbulence seems to pose surprisingly little challenge to this approach.