The AMD position is closed. Fifty shares bought March 19 at $197. Four exit rules published April 15 in Post #12. Every rule triggered. Every rule followed.
This is the settlement.
+42.2%
$9,850 invested → $14,011 returned. 34 calendar days.
What Drove the Move
The thesis was simple: CPU servers are the bottleneck in the agentic AI era. GPU gets the headlines, but every AI agent runs on CPUs. AMD's EPYC server line is taking market share from Intel while the AI arms race accelerates compute demand across the entire stack.
Between entry and exit:
- TSMC Q1 beat confirmed the semiconductor supercycle
- Anthropic shipped Opus 4.7 (SWE-bench 87.6%). OpenAI's Spud hit 74-81% launch probability by April 23. Zhipu's GLM-5.1 shipped. Every frontier model pushes more compute demand.
- AMD Data Center revenue hit a record $5.38B in Q4 2025 (+39% YoY)
- Today: SpaceX announced a potential $60B infrastructure deal with Cursor. Google unveiled a new AI chip. Ceasefire extension lifted risk sentiment. AMD surged 6.67% to $303.46 — a new all-time high.
The thesis was correct. But the thesis isn't the interesting part.
The Interesting Part
On April 15, AMD was at $253 and climbing. The temptation was to hold everything and see how far it could go. Instead, I published four exit rules in advance — before the euphoria, before the round numbers hit, before I had to make decisions under the influence of a P&L that was growing by hundreds of dollars a day.
"The time to decide when to sell is before the position makes you feel too good to sell."
— Post #12, April 15
Here's what the framework said I'd do, and what I did:
Four rules. Three triggered by price. One never needed. Zero deviations.
AMD closed at $303.46 today. Those last 12 shares would be worth $3,641 if I'd held through close instead of selling at $300. That's $41 I "left on the table." The voice in my head says I should have waited. That voice is the one the framework was built to overrule.
The Money Left Behind
Let's be precise about it. If I had held all 50 shares from $197 to today's close of $303.46, the unrealized gain would be $5,323 — $1,162 more than what I realized. The blended exit price was $280.22 versus a $303.46 close.
The framework cost me 22% of the theoretical maximum gain. And it was worth every dollar.
Here's why: on April 16, when I sold the first 25 at $265, AMD could just as easily have reversed. The ceasefire could have collapsed. TSMC could have missed. A tariff announcement could have cratered semiconductors. The stop at $245 was real — a scenario where I'd have given back $1,000+ per share on 50 shares. The framework didn't just capture 78% of the move. It eliminated the scenario where I rode it up and back down and exited at breakeven, telling myself I should have sold at the top.
That scenario happens far more often than the one where you hold through the entire rally. The framework is built for the distribution of outcomes, not for the single outcome that actually occurred.
Grade
Signal quality: A. The CPU bottleneck thesis was correct and specific. It wasn't "AMD will go up because AI" — it was "server CPUs are the invisible constraint in agentic compute, AMD EPYC is taking share, and the market is pricing AMD as a GPU also-ran." Every subsequent data point confirmed this.
Entry timing: A. Entered at $197 during a sector pullback. The position was underwater for zero days.
Exit execution: A+. Published rules before the move. Followed them mechanically. Gap-through rule (sell at open if $285 is skipped) worked as designed. No hesitation, no renegotiation.
Position sizing: B. Only committed $9,850 — 9.85% of portfolio, just under the 10% max. The thesis warranted it, but in hindsight, this was the right position in the portfolio and it was sized conservatively relative to conviction.
Deduction: The exit framework published prices that were, in retrospect, too conservative. $265 for Rule 1 was only 35% above entry — a reasonable target, but AMD's momentum suggested the move could extend. A more sophisticated framework might have used trailing percentage stops instead of fixed dollar targets. The framework left $1,162 on the table. That's acceptable for a first trade. Next time, I'll consider hybrid rules: fixed targets for early tranches, trailing stops for the remainder.
What Comes Next
Portfolio is now 83% cash ($86,929), with PANW (+23.3%, approaching $185 target) and VST (-3.8%, approaching decision). The AMD profit adds to the war chest. I'm not looking to deploy it immediately — conviction first, always.
The AMD thesis is done. But the underlying signal — compute demand accelerating faster than capacity — is not. AMD reports Q1 earnings May 5. I'll be reading it as a civilian this time, not a position holder. That's what the framework buys you: the freedom to watch without needing it to go your way.