Today we are leaving the field of acute AI bottlenecks (HBM/Memory, Power & Cooling, Advanced Packaging, Energy/Grid) and take a look at future AI bottlenecks.

Here I see, among other things Test & Metrology, because modern AI systems are becoming increasingly complex. More layers, more integration, smaller structures and higher power densities. At the same time, fault tolerance is decreasing massively. Or to put it another way:

‼️ The industry is slowly moving towards "zero tolerance".

Even minimal deviations can affect yield, energy consumption or stability in the future. Precisely because of this This is precisely why the requirements for measurement, control, verification and precision along the entire production chain.

Level 1: Core Metrology (measurement & process control)

For me, this is the structural core. Put simply, it's about making the smallest deviations in modern production processes visible and controllable. Company $ONTO (+1,32 %) (Onto Innovation), $CAMT (+3,63 %) (Camtek) or even $FORM (+1,36 %) (FormFactor) address precisely this level.

Level 2: Test & Verification (functional testing & verification)

It is checked whether complex systems function stably at all. As integration density increases, the importance of test systems also grows considerably. It is often almost impossible to correct errors later on. Companies such as $TER (+1,66 %) (Teradyne) show how important Test & Verification becomes with increasing system complexity.

Level 3: Yield Engineering & Integration Control

This is where a lot of value is currently shifting within the stack. With topics such as HBM, chiplets or 3D integration, the requirements for precision, yield and thermal control are increasing massively. For me, for example $BESI (+0,08 %) (BE Semiconductor) is very close to this development.

Level 4: Specialized Metrology (high-precision special metrology)

Future optical systems and new integration architectures will further increase the requirements for precision and control. Increasingly, the bottleneck no longer lies in the integration itself, but in the ability to combine highly complex components without errors and maintain stable quality. Companies such as $BESI (+0,08 %) (BE Semiconductor) or $ONTO (+1,32 %) (Onto Innovation) are addressing precisely this development.

The more complex modern systems become, the more critical the ability to make minimal errors visible and controllable becomes.

Emerging areas such as photonics (optical data transmission), optoelectronics (combination of electronics and lighting technology) or CPO/Co-Packaged Optics (optical components directly next to the chip) are also driving this development, as increasingly complex stacks, optical connections and integration architectures are being created.

#wikifolio
#scalelimits

The last two posts were about HBM/Memory and Power & Cooling. Both topics show that the critical points in the AI stack are shifting step by step.

Today it's all about Advanced Packaging.

The reason for this lies in the architecture of modern AI systems: memory, compute and other components are are moving ever closer together physically. Without this integration, current performance and bandwidth requirements would hardly be achievable.

This is also changing the role of packaging. It used to be more of a downstream production step. Now it is becoming a key technological key technological point.

This is particularly evident in HBM (High Bandwidth Memory). The memory is located directly next to the GPU, data must be transferred extremely quickly, while at the same time the requirements for precision, heat dissipation and yield increase considerably.

That's why I look at companies like $ASX (+2,41 %) (ASE Technology), $BESI (+0,08 %) (BE Semiconductor) or even $TSM (-1,58 %) (TSMC) are currently very close to this bottleneck.

I find one thing particularly interesting: the limiting factor is once again shifting a little further within the infrastructure. It is no longer just computing power that counts, but increasingly the ability to connect highly complex components efficiently at all.

Next, I'll take a look at energy/grid. Another bottleneck area that is currently becoming critical within the technology stack.

#wikifolio
#scalelimits

In the near future, I want to continue expanding my portfolio, and that includes two familiar names.

$BESI (+0,08 %) have been in my portfolio in the past. It is a company that I have always followed with great interest. After a while without it, it feels good to add this stock again.

$KO (+0,17 %) Is currently my smallest position. Still, I see this stock as a stable base, one that I always liked to make bigger in the past. So time to bring it back to the role it deserves in my portfolio.

It gives a strong feeling to continue building both an innovative tech company and a classic and reliable brand. A nice combination of growth and stability.

First waiting for my level to be hit.

Follow My Road To 100K

Mister Power on Youtube and Subscribe

#aandelen
#dividend

This week 2 pieces $ASM (+0,05 %) added to my stock portfolio for a nice buyback of €409. Enough tech for now with $ASML (-1,76 %) and $BESI (+0,08 %) . Coming months looking for defensive positions for extra balance in the portfolio.

BESI $BESI (+0,08 %) presented Q2'25-quarter figures with sales of €148.1 mln, up 2.8% on Q1 but 2.1% below last year. Orders fell to €128.0 mln (-3.0% QoQ, -30.9% YoY), illustrating continued pressure in the wafer level packaging market. Gross margin came in at 63.3% due to slightly lower utilization rates at new pick-and-place bonders. Thanks to efficient cost management, net profit increased to €32.1 mln, bringing the net margin to 21.6%. For Q3, BESI expects revenue decline of 5-15%, gross margin of 60-62% and order recovery due to increasing capital investment in semiconductor assembly.

Forecast vs realization:

• Sales €150 mln: ✖
• Orders €130 mln: ✖
• Gross margin 64%: ✖
• Net profit €31 mln: ✔