Tower Semiconductor (TSEM): The Quiet Bandwidth Bottleneck in AI

What’s happening
Tower is seeing growth because AI needs more bandwidth, not more computing power.

Why it matters
AI growth now relies on better optics, radio frequency technology, and power efficiency.

What headlines miss
Timelines for qualifying products and how much they are used are more important than big announcements.

Key risk
Tower is expanding before demand fully arrives, and its stock price does not allow for much delay.

Investor lens
Tower is in a good spot as an infrastructure provider, but its stock price assumes everything goes as planned.

In every market cycle, the popular investments get crowded, and the more interesting opportunities become less noticed.

AI has reached that point.

Compute was the first act. GPUs got faster, racks got denser, and capital flowed accordingly. Now the constraint is less glamorous and more mechanical: how data moves between those chips without burning power or time. That problem lives in optics, RF, and power management.

This is where Tower Semiconductor operates.

There is a real opportunity here, but also a real risk. Tower’s stock price already expects that the needed infrastructure will be built on time.

$TSEM ( ▼ 1.94% ) is a speciality analogue and mixed-signal foundry. It does not chase the smallest digital nodes or compete head-on with $TSM ( ▼ 0.52% ). Instead, it focuses on process platforms that tend to last longer, change more slowly, and stick once designed in.

Those platforms include silicon photonics, RF infrastructure, power management, and sensors. They show up in data centres, industrial systems, automotive electronics, medical devices, and defence hardware.

Put simply, Tower does not make the main processors for AI. It makes the parts that connect everything.

AI data centres are learning a common lesson: when one part of a system improves quickly, another often becomes the new bottleneck.

GPUs have outpaced networking.

At the scale of AI, moving data with wires over short distances is becoming less efficient. Optical connections and, soon, co-packaged optics help solve this. They make data travel shorter, use less power, and allow for more bandwidth.

Tower is already manufacturing into that problem set.

Recent results show this trend. In Q3 2025, revenue was about $396 million. For Q4, the company expects around $440 million, give or take 5 percent. Management says this growth comes from demand for data centre infrastructure and new platforms, not from consumer products.

This is not just a hopeful projection. Real demand is already showing up sooner than expected.

The market is not debating whether silicon photonics matters. It is debating timing and discipline.

Three questions matter more than all the rest.

At its current price, the stock reflects a more positive outlook.

Tower’s role in AI is indirect but useful. Its silicon photonics and radio frequency parts are essential for bandwidth growth. They gain from AI expansion without needing to spend heavily on the latest chip technology.

Rather than spending billions on brand-new factories, Tower uses a corridor strategy. In New Mexico, $INTC ( ▼ 1.85% ) runs the factory while Tower supplies the equipment. In other places, Tower upgrades and expands existing factories.

The goal is to use money efficiently. The real test is whether these upgraded factories are fully utilised.

In analogue and power technology, expertise builds up slowly and steadily. Tower’s regular updates to its 300mm power and BCD platforms strengthen its position in areas where experience matters more than the number of transistors.

In the next year to year and a half, a few key factors will be more important than any press release.

None of these developments is flashy, but they all affect profit margins.

Tower still operates in the real world of factories and business cycles.

If infrastructure spending slows, factory use can drop. Delays in getting new capacity ready can delay profits by months or years. Spending on expansion can get ahead of revenue. Also, geopolitical risk is real since some operations are in Israel.

The stock price does not need a disaster to fall. It just needs growth to be slower than expected.

These are possible scenarios, not guarantees.

Growth keeps going through 2026. Demand for infrastructure makes up for weaker auto markets. Factory use improves slowly, and profit margins rise, though not quickly.

Silicon photonics grows faster than expected. Co-packaged optics is starting to make money instead of just being a plan. The upgraded factories are fully used. The market starts to see Tower as essential infrastructure, not just a cyclical company.

Growth slows down. Costs go up before revenue does. Factory use is lower than hoped. The stock price drops faster than the business can recover.

Buy if you believe optical interconnect demand scales smoothly and Tower executes without drama.
Hold if you want exposure but need confirmation that ramps persist and margins follow.
Avoid if you expect a broad analogue slowdown or think expansion spending caps returns.

These choices are not about right or wrong. They are about timing.