Why My Biggest Position Will Get Bigger
Design wins, AI inference contracts & improving margins. The last uncrowded AI name in the market.
The AI hardware trade is overcrowded.
Everyone knows the story, from memory to networking to “optimization plays”, and has a list of proxies to ride it. Most of the sector is fully priced and the risk/reward isn’t that interesting.
All, but one.
Silicom is the only AI hardware name I could find that nobody was talking about. The stock is up ~150% since I shared my first thesis in March, and yet, we’re still very far from where it should trade.
Back in March, the thesis was simple: the bottleneck moved from raw compute to networking. Inference overtook training, GPUs became powerful enough but sat idle, waiting on data because of copper’s physical limits. Photonics is the long‑term fix, but we’re years away from industrial scale while the industry needs a solution today. That near‑term solution comes from customizable hardware - FPGAs, SmartNICs and more. Configurable, low‑latency silicon that can be tuned to any workload.
Silicom was - still is, the overlooked pure‑play: a vendor‑agnostic SmartNIC assembler for low‑latency niches (cybersecurity, algo trading, defense, telecom…). Hyperscalers started approaching them to test whether that hardware could handle AI inference, and initial Proofs of Concept (PoCs) followed.
We were looking at a ~$100M cap, ~1x EV/sales, ~50% of the market cap in net cash, double‑digit growth guidance, and the inference opportunity as a free call option on top. Since then, the story changed enough that my old base case became a bear case. The last few weeks erased that bear case and forced a new one:
What used to be my March bull case.
In this write‑up, I’ll walk through what’s changed since my first thesis:
The AI infrastructure situation - CPO delays & the emerging memory crunch.
The unexpected strength of Silicom’s core business and FPGA ecosystem.
The details of the AI inference deal(s) and what I expect from them.
What comes next for Silicom across AI inference and its other verticals.
Future catalysts, stock price action, and how I plan to size the position.
Wealth is made by sizing on our best ideas.
Silicom is one of mine.
