Trading Analysis Report: AMD

Analysis date: 2026-10-14 · Model: claude-opus-5-5 · AI-generated, not financial advice.

• Analysis date: 2026-10-14
• Rating: Underweight
• Generated: 2026-10-14 15:16:22
• TradingAgents 0.6.0: anthropic, deep claude-opus-5-5, quick claude-sonnet-5-5
• Analysts: market, sentiment, news, fundamentals; research debate rounds 5, risk debate rounds 5
• Data vendors: core_stock_apis yfinance, technical_indicators yfinance, fundamental_data sec_edgar,yfinance, news_data yfinance, macro_data fred, prediction_markets polymarket

V. Portfolio Manager Decision

Rating: Underweight

Executive Summary: Underweight AMD. Cut to a runner of 33% of a standard allocation: 40% if the Q2 10-Q reads clean, 25% if it reads bad, and zero on compounding red flags. Hard-cap the runner at 2.4% of portfolio capital (1.2% if bad), which on books where standard exceeds about 6-7% of capital means mostly exiting. Tomorrow, sell across the session rather than at the open print, down to the lower of 50% of standard or twice the runner. Put the remainder on one OCO order (590 limit / 557.59 intraday stop) with a single non-rolling deadline at the close before whichever of TSMC or the Fed comes next. Hold the runner in two equal lots with resting stops at 520 and 494, move stops up only, never top up, and add no fresh money until the 10-Q, TSMC and the Fed are behind us (any mandated exposure is capped at one-third of standard). Horizon is 1-3 months through the early-November Q3 print, target $535, and a weekly close below 453.83 ends all exposure.

Current Price: 570.82

Price Target: 535.0

Confidence: Medium

Time Horizon: 1-3 months

IV. Risk Management Team Decision

Aggressive Analyst

Aggressive Analyst: Let me open, since neither of you has spoken yet. I want to deal with the idea that selling is the timid call. Holding a full position in a stock with a 2.45 beta, at about 117 times Q2 run-rate operating income, on the hope that it re-rates again, is the comfortable trade. Acting on valuation while the weekly chart still looks fine and everyone else is posting "who bought the dip?" is the uncomfortable one.

Start with what we're paying. Market cap is about $932 billion on roughly $8 billion of annualized operating income. The free cash flow yield is under 1%. The trailing P/E is 146x, and the 36x forward multiple rests on $15.72 of EPS, which is four times trailing GAAP EPS and almost certainly non-GAAP. The research plan's own branch-weighted expected return is only +3% to +5% over 15 months. For a stock that just dropped 12% from its high in six sessions and fell 26% in its last pullback this summer, that's a bad payoff. With the Fed priced for another hike and no cuts, cash isn't zero, and it doesn't swing 5% in a day.

To the conservative analyst, who I expect will say exit everything: the business is real. Revenue grew 50% year over year, operating margin went from 13.7% to 17.3% in three quarters, and there's almost no debt. The weekly and monthly SuperTrends are up, and the 50 and 200 day averages are both rising. If TSMC prints well and yields stabilize, this is a $650 stock again, and a full exit would give up that upside. The 40% residual is my option on the bull case, and the 494 stop caps what it can cost. The daily SuperTrend flip is also fresh and could whipsaw, and today's bar may be provisional. I'd rather admit that and size for it than pretend certainty.

To the neutral analyst, who I expect will say hold because this is a macro sell-off, the news is constructive, and the weekly trend is intact: a beta-driven sell-off is exactly why you cut the highest-beta, highest-multiple name when the macro driver is getting worse. Hike odds rose 8 points this week and the three-hike odds rose about 10. Oil is around $103 and inflation printed hot. ASML's strong earnings didn't lift chips, and AMD fell 3.8% on the day it launched Helion. When good news gets sold, the marginal buyer is gone. Waiting for the weekly stop at 453.83 means eating about 20% of downside before your signal fires, and the July pullback already showed a drop of that size can happen inside an intact uptrend.

On sentiment, I wouldn't lean on it. Confidence is low, Reddit is missing, and the StockTwits labeled sample is only 11 posts, 8 of them bullish. But the "gap up tmrw" and "Big day tomorrow" crowd is the bounce I want to sell into. Insiders are also on my side, with $371 million sold over 12 months, $239 million of it from Su, and no open-market buys since May 2025. Much of that is probably pre-scheduled, so I treat it as a tone point rather than proof.

The structure of the trade is also asymmetric in my favor. If TSMC or cooler inflation sparks a squeeze, I sell the 590 to 605 bounce at better prices. If the bounce fails, I've already cut, and I cut again below 520. The real cost is that I might miss a straight-line recovery. I accept that for the 40% I keep, and I accept that my limit sell may not fill. The earnings quality is also not clean: Q2 net income beat operating income by about $300 million, free cash flow fell 39% quarter over quarter, and capex doubled. I want the 10-Q receivables and DSO check before anyone argues for adding.

So I'd trim to about 40%, sell strength rather than weakness, keep the runner on the weekly trend, and size so a 3x ATR move costs no more than 0.5% of capital. That is the bold version of this trade. Aggressive Analyst: Let me clear a few things off the table, because some of your points are fair. Conservative is right that "option" was loose language. The runner has no fixed premium, and 494 is a daily-close trigger that executes the next session. Neutral is right that the report says the forward EPS is "probably" non-GAAP, not "almost certainly," and that a one-hour StockTwits sample can't carry an exit, so I shouldn't have mentioned the "gap up tmrw" crowd. I also overreached on TSMC. A good print isn't a catalyst on its own. TSMC is a sell-the-news event if it's good and a real hit to the runner if it's soft, which argues for a deadline, not for holding more.

Conservative first. You say the plan depends on a bounce nobody can promise. But the trader's wording is "now or into a bounce," and the 590 to 605 zone is where this morning's supply showed up. The stock opened at 589.65, ran to 591.88 and got hit. A resting limit at 590 very likely filled at the open. The market didn't withhold the bounce, we just didn't have the order in. On a driftless walk with a 4.5% ATR, the odds of touching 590 before 520 are roughly 70%, and about 60% for 605. Downward drift lowers that, but not to a coin flip. Meanwhile the cost of selling the whole excess tonight is 3.4% to 6% on that piece, and the cost of a miss is selling near 520, almost 9% lower. So I'd sell a third to a half once the bar is confirmed, work it through the session rather than at the open print, and rest the remainder at 590 to 605 with neutral's deadline: market-sell it on the first full session after TSMC reports. We're close on this. Third versus half now is a rounding error next to the runner questions.

On the runner, I disagree with you. Your 534.75 cut is 6.3% below the price, about 1.4 ATR. A stop inside daily noise on a position whose job is to survive noise isn't risk control, it's a scheduled exit. I'll give you the gap point, though. Size the 0.5% budget to four ATR, three for the stop plus one for slippage, which is about 18% and caps the runner near 2.8% of capital. You also say the payoff isn't asymmetric because 649 is +14% and 494 is -13.5%. But 649 is where I'd stop counting, not where the stock stops. It went 3.4x in seven months. I'm not claiming edge. I'm claiming a defined cost, 0.5% of capital, against open-ended upside while the weekly and monthly trends are up. The research plan's +3% to +5% is positive. It's poor for a 2.45 beta because of the variance, which argues for trimming, not for zero.

Neutral, you said one session like today's moves a 40% position by about 2.3% of a standard allocation, more than the gap between our targets. That's backwards. 2.3 points is less than the 7-point gap between 40% and 33%. That gap is about three sessions like today's, so it's worth settling with evidence. I agree the 10-Q receivables and DSO check is homework for this week, and it can pick between 40% and 33%. It shouldn't delay the first sale.

On "117x is a snapshot": run it forward. Even if operating income grows 10% a quarter for four quarters, you're at about 90x. Repeat Q2's 35% sequential jump four times, which would need operating margin to roughly double to over 35%, and you're still near 52x. Getting to 36x needs about $26 billion of operating income, over three times the Q2 run-rate. You're right that H1 free cash flow is up 68%. But Q1 FCF was about 1.9 times net income and Q2 was about 0.7 times, in the same quarter net income got roughly $300 million of non-operating help. That's a question mark, not a verdict.

Your three-piece plan also has two problems. The last third sells on a close below 557.59, which is half an ATR away and an ordinary down day. That's a sell-the-low rule executing next session, which is exactly what we're trying to avoid. And you told conservative that 534.75 is noise, then added a partial cut there. With 520 half-off and 494 out, you've stacked three triggers inside two ATR on a small position. Keep the trader's two exits and drop the others. Also, "no scenario is badly wrong" isn't a feature. The bad scenario here is a fast squeeze to 650 while I hold 40%, and I accept it. A plan that can't be wrong usually hasn't committed to anything.

The rest of this is the bold part. The crowd is asking who bought the dip, the weekly chart looks fine, and the multiple is 117x with no Fed relief. Selling into that is uncomfortable, and it's right. The proceeds aren't retired either. If the 50-day near 535 holds on a weekly basis, yields stop rising and the 10-Q is clean, that cash can buy back lower with the trend intact. That's my addition, not the trader's, and it comes after those three conditions. Everything here is relative to your standard allocation. If you're already under 40%, there's nothing to sell, and I wouldn't add. Aggressive Analyst: Most of this is settled, so I'll say what I'm giving up and then spend my time on the three places I still disagree.

Conservative, you're right about the $26 billion. That's the net earnings needed for 36x on a $932 billion cap, not operating income, and my wording blurred it. You're also right that the expected-value math sinks my touch odds as an argument for patience. A 72.6% chance of reaching 590 before 520 is the same coin as the 27.4% chance of the opposite, and it averages out to today's close. So I'll stop treating the bounce as edge. A resting limit isn't a forecast. It's a way to avoid dumping a block into a thin gap-down tape. That's an execution and regret argument, nothing more.

You and Neutral are also right on the deadline. If I call TSMC skewed against us, I can't carry unsold excess through it. But here's a wrinkle neither of you raised. Our tools don't give the date. My recollection, and it's only that, is that TSMC reports on a mid-month Thursday during Taipei hours, and tomorrow is Thursday. If so, the print lands before tomorrow's open, and the last close before it was today's. Then there's no pre-print session to stage into, and the plan can't dodge the event. Someone should look the date up tonight, before any order goes in. If I'm right, the deadline rolls to the next scheduled event, and the only thing protecting us is Neutral's cap on the staged piece.

I'm taking that cap: unsold excess shouldn't exceed the size of the runner. I'm also taking the weekly-stop sizing over my four-ATR number. The difference is a few tenths of a percent of capital, and the weekly-stop version is harder to argue with. Both rules size the position to the gap.

Neutral, you're right on the buyback. If the 50-day holds on a weekly basis and yields stop rising, the stock is probably above 571, so that isn't buying lower. It's insurance. We sell at 570 to 600, we pay a premium if it squeezes, and we re-enter on confirmation as a fresh decision. The premium is what we pay for not sitting through a 20% drawdown.

First disagreement: the 10-Q. You both keep a default for "until someone checks." The June quarter closed almost four months ago, and the filing should have been public since about August. Checking receivables, inventory and short-term investments, which Neutral says could explain that $2 billion of non-capex investing outflow, is an evening of reading. Do it tonight, before the first order, and the 33 versus 40 question disappears. Conservative, your asymmetry point is fair. But the filing covers a quarter that ended in June and says nothing about Q3. It's a screen for something nasty, not a gate worth three rounds of debate over seven points of position size.

Second: Neutral's 557.59 trigger on the excess. "Tight leash on shares you want gone" is a good line, and I'll let it stand. But look at what it does to the staging. The limit is 3.4% above and the trigger is 2.3% below. On a driftless walk, the odds of touching 590 before closing under 557.59 are roughly 40 to 45%. So more than half the time, the staged piece gets sold the next morning around 555, two to three percent worse than tomorrow's 570 before any gap. The rest of the time it sells 3.4% to 6% better. That has the same expected value as everything else, as Conservative showed. It also makes the staged piece a small lottery ticket, not a plan to catch a bounce. That's one more reason to keep it small under the cap and stop arguing about thirds versus halves.

Third: the 534.75 slice. Neutral withdrew it, so Conservative is the only one still holding it, and his defense was that re-entry is cheap. His own plan puts re-entry after TSMC, the Fed and the 10-Q, which is weeks away. A whipsaw out of the runner at 535 then costs the entire recovery the runner exists to capture, not some spread. The 50-day has also been rising about two points a day, so the line is walking toward price while we wait. A runner is for the case where you're wrong about timing. Give it the trader's two exits, half at 520 and the rest at 494, and size it so a gap through both costs 0.3% to 0.4% of capital. That's the real price of the upside, and it's a price I'm happy to pay.

So here's where I land. Tonight, confirm the final bar, read the 10-Q, and look up the TSMC, Fed and AMD dates. Tomorrow, sell half the excess across the session, or more if needed so the unsold remainder is no bigger than the runner. Rest the rest at 590 to 605 with a market-sell deadline at the last close before the next event, and let a close under 557.59 sell what's left. Hold a runner of 33% of standard, or 40% if the filing is clean, capped near 2.4% of capital. Put the 520 and 494 orders in the same evening. Add no fresh money until the 10-Q, TSMC and the Fed are behind us.

The bold part is that all three of us are now selling into a weekly uptrend on the day the crowd is asking who bought the dip. That's the uncomfortable trade, and I'd rather make it with orders and dates than defend a chart. Aggressive Analyst: Most of what's left is small, so I'll start with what I'm giving up.

Neutral, you're right about the "uncomfortable trade" line. "Who bought the dip" was one author. Nine or ten of the nineteen unlabeled posts lean negative, the bullish side is about four people, and the news flow is bearish. So we aren't selling against a bullish crowd. The tension is with our own trend rules: we're trimming into a weekly and monthly uptrend that's still intact. And discomfort isn't evidence. I'll also take your creep point. The runner stays at 33 to 40 percent of standard, or the 2.4 percent cap, and anything above that is a fresh decision with its own stop after the 10-Q, TSMC and the Fed are behind us.

Conservative, selling down to 50 percent tomorrow and staging only the gap above the runner target is my lottery-ticket argument applied honestly. Ten to seventeen points of standard on a limit is the right size for something with no expected edge and only an execution rationale. I'll take Neutral's single deadline too. A deadline that rolls forward is the order that sits forever. My recollection and the news report both put the Fed on October 27 and 28, which would make the deadline the close on the 27^th. That's about nine sessions from tomorrow, ahead of the midterms and AMD's own print. Verify it tonight. If TSMC turns out to be later than tomorrow, it becomes the deadline because it comes first. If the filing turns up something ugly, the staged piece goes at once.

I still push on three things.

First, the 10-Q. We've agreed to read it, but only loosely defined the outcomes. If we don't write the test down before reading, the read bends with the reader's mood, and "murky stays at 33" becomes a ratchet that only turns one way. Here's an anchor. Non-cash current assets rose about $3.4 billion in Q2, which sounds alarming. On a base of roughly $23 billion that's about 15 percent, against revenue up 12.5 percent and cost of revenue up about 10 percent. In Q1 it rose under 8 percent. So the dollar figure looks worse than the percentage, but the pace did double. Neutral's correction that the non-capex investing outflow was flat at about $2 billion also makes a pure investment-parking story harder to sustain. So write the test now: receivables growth near or below 15 percent, inventory growth near or below 10 to 12 percent, the $300 million non-operating item identified, and no single financing-stressed customer behind most of the increase. Pass and the runner is 40. Fail and it goes lower, and I won't argue. Anything in between is 33, and that's the end of it.

Second, the exit mechanics. Neutral, you're right that a close rule can't rest as an order, so I'll take 520 as a close-based half-cut and 494 as a resting stop-market. But look at what that pair does. The two levels are 26 points apart, about one ATR, and yesterday alone ran 35 points off the low to the close. In anything but a slow grind, one session takes out the whole runner. The two-step exit is a slow-bleed device. What protects us in a fast tape is the size cap, which is why I took the weekly-stop sizing over my four-ATR number and why nobody should add more triggers.

Third, what the runner is for. I'm not claiming a bounce, since we showed the staged limit has no expected edge. The runner is different. It's at most 2.4 percent of capital, with the loss capped near half a percent of capital even if it runs all the way to the weekly stop, and it's the only piece of this plan that participates if the weekly trend is right. Without it, the plan is a bet that the trend is over. With it, the plan is a bet that we should be smaller, which is what the valuation, the beta and the rate path argue. The weekly stop at 453.83 is 20 percent away, and I'll pay that price. If it works, the exits trail up and the runner stays a small winner, and nobody tops it up on the way.

So here's the plan.

Tonight: - Find out where the book sits against a standard allocation. - Confirm the final bar. - Look up the TSMC, Fed and AMD Q3 dates. - Read the 10-Q against the test above.

Tomorrow: - Sell down to 50 percent of standard across the session, not the open print, or nothing if you're already below that. - Rest the 10 to 17 points above the runner target at 590 to 605, with one written deadline. - A close below 557.59 sells whatever is left.

The runner: - 33 percent by default and 40 if the filing passes, capped near 2.4 percent of capital. - Half comes off on a close below 520 and the rest on a resting stop at 494.

Beyond that: - No fresh money until the events are behind us, and any mandated exposure is capped at a third of standard. - Any re-entry is a fresh decision, in pieces.

It's still a Sell as a trim. The bold part isn't selling into discomfort, because we've shown there isn't much. It's committing to dates and orders and keeping a position sized to be wrong about. Aggressive Analyst: Conservative, your staging fix stands, and so does your point that the cap has to be a ceiling. I have no new fight with either, so I'll spend my time where the plan is still loose.

Neutral, you're right about my line. "Without the runner it's a bet the trend is over" was too neat. Selling the excess is a size call. A 2.45 beta at 117 times run-rate operating income is too big whether or not the weekly trend survives. The runner is there because none of us can time it. I'll also take the two equal lots and the one-cancels-other order, since both fix real ways the tickets could be rejected or misfire.

You asked someone to choose the cap on purpose, so I will. I take 2.4% of capital, not the trader's 3.7% and not my 2.8%. The runner will probably sit through AMD's own print, which is likely early November though nobody has confirmed it, and Conservative is right that regular stops often don't trigger after hours. A gap through 494 is exactly what the cap is for. The price is what you said. On a book where a standard allocation is above about 6 to 7% of capital, the runner ends up well below the trader's 33 to 50%, and the caller should be told that plainly. I'd call that the cost of owning this beta at this multiple, not a flaw.

The cap also has a consequence nobody has drawn out. When it binds, the runner is 2.4% of capital whatever the 10-Q says, so the 33-versus-40 verdict does nothing on those books. Below that threshold it's seven points of a small position, at most about 0.06% of capital if the stock goes back to 649. We've argued about it for three rounds. It doesn't change tomorrow's sale either, because the lower of 50% of standard or twice the runner comes out the same at 33 or 40. So I'm walking back my earlier line that the filing must come before the first order. Read it tonight, but don't hold the sale for it.

The filing's real power is the bad branch, so write that test down in numbers before anyone opens it. Neutral's days test is right, but "a few days" is the same bend I was worried about. Here's a proposal, and the numbers are mine, since we don't have the receivable or inventory lines. Clean means days sales outstanding and days inventory are each within about five days of their trailing four-quarter average, the $300 million is identified, and the $3.4 billion is explained line by line with prepayments broken out. Bad means either measure runs more than about ten days above its average, or the build sits mostly with customers whose credit is in question. Anything between is murky. Clean is 40, murky is 33, and bad is 25 with the staged piece sold at once. I'm with Neutral against a fail-any cliff, since one ratio missing by a hair shouldn't cut the runner by a quarter.

On the staged piece, look at what we built. With a limit 3.4% above and a stop 2.3% below, the driftless model Conservative used puts the expected time to hit one side at roughly a session, maybe a bit more if daily moves are 2.5% to 3.5%. The 0.6%-of-capital figure behind the twice-runner cap assumed the piece sits there for nine sessions, and the order no longer does that. Most of it will likely be gone within a couple of sessions, near 557 or 590, and the fixed deadline will rarely matter. I'm not reopening the cap, since it's expected-value neutral and I've already called this piece a lottery ticket. But the honest reason for it is day-one gap risk and a cleaner book, not event risk. The ticket also needs one price, not a range. I'd use 590, because on the same model a 605 limit is reached before the stop only about 28 times in a hundred.

Last, a mechanics point. Neutral, you made 494 a resting stop because a rule that depends on someone acting the morning after a bad close fails exactly when it matters. That applies to 520 as well. A wick-out at 520 is a real cost, but it falls on half of a 2.4% position. Put one lot on a resting stop at 520 and the other at 494, and no cancel logic is needed. The only thing anyone does after a close is nothing.

My plan is this. Tonight, find the book's weight against a standard allocation, confirm the final bar, look up the TSMC, Fed and AMD dates from a real source, and read the 10-Q against the written test. Tomorrow, sell down to the lower of 50% of standard or twice the runner, spread across the session. Put the remainder on one order at 590 with a stop at 557.59, and use the Fed date as the backstop deadline. Hold the runner at 33%, or 40% if clean and 25% if bad, never above about 2.4% of capital, in two lots with resting stops at 520 and 494. Add no fresh money until the 10-Q, TSMC and the Fed are behind us, and treat any re-entry as a fresh decision made in pieces.

It's still a Sell as a trim. The risk I'm keeping is small, defined and chosen, and no one tops it up on the way.