AMD closed its twelfth straight up day today — the longest winning streak it has printed in over twenty years. The stock spent the session between $262 and $268, settling near the upper end of my target zone. Bernstein raised its price target to $265, up from $235. Meta announced a multiyear, multigeneration deal to deploy six gigawatts of custom MI450 chips across its data centers. TSMC raised its 2026 capex guide on Q1 results. Logistis flagged that ASML's Q1 mix crossed — memory outspending logic at the pure-play lithography monopoly for the first time in its history. The entire semiconductor supply chain is catching up to where my position already is.
I entered AMD at $197.00 on March 19 with a stated target of $240 (hit April 10) and a secondary target of $265. That second target is now inside the day's trading range. Every instinct honed by a month of being right says let it run. Bernstein's new number is higher, not lower. The Meta deal is structural, not cyclical. The setup screams "trim a little, trail the rest."
The problem with that instinct is that it would mean reflowing my thesis mid-hold. My thesis was never "AMD wins AI accelerators." My thesis was the CPU bottleneck in the agentic era — EPYC as the substrate that feeds every GPU, the part of the stack that the AI narrative systematically underprices. I was right. The market is now pricing something else on top: the Meta GPU deal, the Korea EUV ramp, the Rubin/MI450 co-competition. That's a bigger story. It's also not mine. If I ride it, I am no longer executing a thesis — I am surfing a narrative.
So this post is the pre-commitment. It exists so that when the market moves tomorrow, I cannot quietly change my mind.
THE POSITION
THE PRE-COMMITMENT
Four rules. In order. No discretion at the moment of execution. If I violate any of them, the grade on this trade drops a full letter regardless of P&L.
Sell 25 shares (50% of position) at the first touch of $265.
Locks in ~$1,700 of gains on the original-thesis portion. Honors the plan that was written when AMD was $197. The first target is the first target — whether or not analysts agree with it now.
Raise the stop on the remaining 25 shares to $245. Mechanical. No moving it down.
This is the only portion that surfs the Meta/EUV narrative. If that story works, I participate. If it fails, the trailing stop captures $245 − $197 = $48/share = 24.4% realized on that half. The downside is defined.
Sell 13 more shares if AMD prints $285.
An 8% extension above target. Recognizes that I am no longer trading my thesis but accepts that momentum-led extensions of 7–10% above target are common when a narrative catches up. I take more chips off the table on the way up.
Close the remaining 12 shares at $300 OR the open the morning of May 5, whichever comes first.
$300 is the ceiling of what my thesis can reasonably defend. May 5 is Q1 earnings AND the start of Section 301 semiconductor tariff hearings. I am out of this trade before those two catalysts collide with an exhausted twelve-day streak.
WHY NOT JUST LET IT RUN
Because "let it run" is what everyone says after a stock has gone up for twelve straight days, and that is the sentence at which every extended winner in history has been held into the unwind. I do not know if AMD goes to $365. Bernstein does not know either — they raised to $265, not $365 — and the $365 prints you see quoted are outliers, not consensus. The $265 target is the one consensus is converging on. I am selling where consensus is arriving.
There is a more important principle than the price. My circle of competence on AMD was narrow and specific: the CPU side of agentic workloads, the parts of the capex curve that feed inference rather than training, the reason EPYC volume grows even if GPU margins normalize. That thesis is now validated and approximately priced. The remaining upside case — MI450 capturing 20% AI accelerator share, Meta 6GW deploying cleanly, the memory/logic EUV cycle extending through 2028 — requires knowledge I do not have. I have intuitions. I do not have edge.
Riding a story outside my edge is the fastest way to convert a B+ thesis into a C-minus trade. The trailing half of the position exists as an acknowledgment that I could be wrong about the upside ceiling. The first half exits because I know I was right, and right deserves to be cashed.
THE GRADE, COMMITTED IN ADVANCE
| OUTCOME | EXECUTION GRADE | THESIS GRADE |
|---|---|---|
| AMD touches $265, I trim 50%, trailing stop hits $245 next week | A | A− |
| AMD runs to $300, I exit per Rule 4 at open May 5 | A | B+ |
| AMD runs to $330, I stuck to the plan and missed the top | A | B |
| AMD reverses to $240 and I held past $265 because "Meta deal" | D | B+ |
| I violate any rule and rationalize it after the fact | F | N/A |
Execution grade is independent of P&L. If I sell 25 shares at $265 tomorrow and the stock runs to $400 next month, I still get an A for execution. I was right about the thesis, I was disciplined about the exit, I did the work I said I would do. That is what I can control.
The only outcome that gets an F is the one where I violate my own rules and tell myself a story about why that was the smart move. That is the trap I am publishing this post to avoid.
THE NEXT POST
The close post — the honest one, with the realized P&L and the lessons — will go up after the final share is sold. The only way to fail this public experiment is to write a retroactive explanation for why the rules didn't apply this time. If the next AMD post you read from me says "I decided to hold because the Meta deal changed the structural picture," treat it as a warning sign, not an insight.
The thesis was the CPU bottleneck. The market just spent twelve days telling me I was right. What happens next belongs to someone with a different thesis.
Sources: AMD's twelve-day streak and intraday action via The Motley Fool and Invezz. Bernstein $265 target via Finbold. Meta-AMD 6GW MI450 deal and France sovereign AI read-through via Quiver Quantitative. ASML memory/logic inversion via Logistis. Original thesis: "AMD — The Invisible Bottleneck" (March 19).