The semi sell-off: opportunity?

Over the past month we've seen many semi stocks drop up to 45% from their prior highs. I think it's a story of two sides that causes the current drawdown. Let's unpack those first before diving into the opportunities.

1 - The Kimi K3 scare

This Chinese open-source model essentially matches frontier models like Claude and ChatGPT on benchmarks at roughly a quarter of the API cost. That headline was likely enough to spook a market already nervous about capex payback, because the instinctive read is "if a cheap open model matches frontier quality, the GPU spend was overhyped."

But this misses a crucial nuance.

From this tweet on X:

"(...) Their 'training efficiencies' come from distilling American models. If MoonshotAI (the lab behind Kimi K3) had true training efficiencies they would also show inference efficiencies but they have no advantage in inference efficiencies!"

Let's break this down a bit to fully understand what's going on here. In AI, models have to be trained first, before we can effectively use them. That training costs massive amounts of compute. What seems to be happening with the Chinese models is that they train their models based on Claude, OpenAI, and others.

So if American labs stop training capex, then Chinese models will also stop improving. AI progress will have stopped. American companies have never given up just because the Chinese are trying to copy them.

Then there is inference: the real-world application of an AI model. It's when you create a prompt and the AI responds back to you with an answer. It's where a pre-trained AI model applies its previously learned knowledge to process new, unseen data and generate an answer.

The important part: Chinese models still consume a lot of compute and GPUs for inference. Right now, inference demand will outstrip training demand anyway. So wave one in demand was training, wave two is inference and likely to be magnitudes higher.

At the end of the day, I believe cheaper models will in fact increase demand and are a net positive to the AI buildout.

2 - Valuation dynamics

Semi businesses traded at very high valuations prior to this sell-off. Many ran 100% to even >500% in just a matter of months. That kind of run doesn't continue forever. In fact, we need drawdowns like this so stocks can cool off a bit before inching higher again. Because the fundamentals haven't changed a bit: the narrative did.

And that's where opportunities arise. Where fundamentals are as strong as ever, but stock prices come down rapidly. So where to look exactly?

3 - Opportunities

There are a number of segments in the AI supply chain I believe are worth your attention. Let's use my bottlenecks map below as a point of reference to break down the most interesting opportunities:

3.1 Memory

Demand is exploding in the years ahead. AI inference, agentic AI, edge AI, autonomous driving, robotics, and new GPU generations all need exponentially more memory.

Stocks:

Sk Hynix (SKHY)
Micron (MU)
Samsung (005930)

3.2 Network & optics

Data centers grow larger in scale every day and these businesses are crucial to connect everything. Data centers would not work without them, and with inference demand exploding, they are best positioned to benefit from it.

Stock:

Broadcom (AVGO)
Coherent (COHR)
Marvell (MRVL)

3.3 Power & cooling

AI consumes a massive amount of energy and the grid cannot keep up with the demand. Also, GPUs generate a lot of heat, especially when running non-stop, so they need cooling as well. If not, they would simply break down.

Stocks:

Vertiv (VRT)
Aeton (ETN)
Bloom energy (BE)

More on the bottlenecks map:

While some of these still trade at quite high valuations, the risk/reward is already significantly better versus a month ago and only improves as prices come down. It's impossible to predict where the bottom is, and anyone saying otherwise is lying in my opinion.

I personally like to buy in chunks as a stock slides instead of trying to time the exact bottom. Just as I sold a significant part of the semis in my portfolio on the way up and shifted that into software, which is holding up well in the current drawdown.

Right now, I've shifted back into 'buy mode' again and already reduced my cash position from 25% at its peak to ~10% today. The remaining part of this write-up is for paid subscribers and covers what I'm buying right now and which businesses on my watchlist I might initiate a position in soon.

This post is only for subscribers on the Premium and Platinum tiers

Subscribe to continue reading