Hi everyone,
Here is my portfolio update for the end of December 2025.
2025 YTD Return: Slightly over 100%.
While the final number is positive, the journey to get here was incredibly volatile. As I discuss below, a significant portion of this return feels attributed to luck rather than pure skill—a realization coming from some self-reflection.
Current Portfolio (as of 12/31/2025):
| Ticker | Dec Allocation | Nov Allocation |
|---|---|---|
| APP | 22.99% | 19.11% |
| IREN | 17.80% | 20.92% |
| NVDA | 14.89% | 13.70% |
| NBIS | 10.96% | 11.76% |
| ALAB | 10.54% | 9.84% |
| EOSE | 9.10% | 11.45% |
| RBRK | 8.40% | 6.86% |
| NET | 4.79% | 4.69% |
| HIVE | 0.51% | 0.00% |
| HIT | 0.00% | 1.66% |
Sold Out completely since Nov: HIT (was 1.66% in Nov)
Reflections on 2025
1. Cleaning House: Exiting HIT
I sold out of HIT completely. This position was always a low-confidence holding for me, and I realized I should have sold it much earlier. Keeping a low-confidence stock with no immediate catalyst for growth in a high-conviction portfolio was a mistake.
2. Acknowledging Luck
I was very fortunate this year. My returns were largely driven by getting into AI infrastructure stocks early. However, looking back, my entry into IREN in August involved significant timing luck.
If I had strictly followed the standard “wait for confirmation” approach, I might have waited until September or October to buy. Given the recent volatility, that timing difference—driven by luck rather than strict rules or skills—would have resulted in a loss on IREN rather than a gain. I have to admit that a few weeks of difference determined my year’s success.
3. The HIVE Saga: Operational Risk vs. Valuation
I have flip-flopped on HIVE: I bought in September, sold in October due to a lack of confidence, and just re-entered with a tiny “watch position” (0.5%) in December.
My confidence in HIVE has always been lower than in IREN. While HIVE trades at a significantly lower EV/S ratio, its AI/HPC business appears less robust than IREN’s, suggesting higher operational risk.
This highlights a key realization for me: My goal should be earning a “certain 30%” rather than chasing an “uncertain 50%.”
In a portfolio that is already volatile, adding lower-confidence companies just because they look “cheap” is unnecessary. I will keep HIVE as a tiny position for now, and will only add capital if the company proves it can secure new contracts and stably expand its AI/HPC capacity.
4. Risk vs. Reward: A Comparison with Bert
Despite my 100% return, I don’t feel I did a great job this year. My portfolio experienced extreme volatility and stress.
When I look at Bert’s high-growth portfolio, he achieved a ~90% return this year—almost the same result—but with much lower volatility and “heart attack” risk. This forces me to ask: Did I take on a massive amount of meaningless risk? The data suggests I did. I need to tighten up my criteria to ensure that the risks I take actually contribute to potential outperformance.
5. Changing the Mindset
For years, I took excessive risks because I was fantasizing about retiring early. I would constantly compare my YTD or monthly returns with others on the board, hoping to “win.” This was the wrong approach.
I have found a new mindset that is much more effective for me. Instead of asking:
-
“Will this stock have a higher return?”
-
“Will I beat X investor this month or YTD?”
I now ask myself:
-
“Does this stock fit the Saul investment philosophy?”
-
“What would Saul do in this situation?”
Focusing on the process rather than the result seems to lead to better decision-making.
6. 2026 Outlook: Bullish on AI Infrastructure
Heading into 2026, AI Infrastructure remains my highest conviction theme.
The market has been spooked recently by concerns over OpenAI’s cash flow and the sustainability of LLM training costs. Personally, I share some doubts about whether OpenAI specifically will be profitable in the long run.
However, this does not shake my confidence in the infrastructure play.
The training of Large Language Models might end up being a “Winner Takes All” (or “Winner Takes Few”) market, dominated by giants like Google (Gemini), Anthropic (Claude), OpenAI, and Alibaba (Qwen). It is possible OpenAI could lose its lead or be acquired at a lower valuation.
But AI Inference is different.
Inference is not a winner-take-all game. I believe inference workloads will continue to grow exponentially. As long as that demand exists, the compute providers will win, regardless of which specific model is the most popular.
Therefore, if the market drags down the entire sector because of OpenAI fears, I view it as a buying opportunity for the infrastructure providers.
Happy New Year to everyone, and thank you for the incredible discussions in 2025.