Trading Analysis Report: AMD

Analysis date: 2026-10-21 · Model: claude-opus-5-5 · AI-generated, not financial advice.

• Analysis date: 2026-10-21
• Rating: Underweight
• Generated: 2026-10-21 11:26:32
• 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 exposure to a loss-budgeted core and keep it, rather than exiting. Today, sell to about 60% of a standard allocation whatever Meta's report date turns out to be. Then work down to the final weight, which is the lower of (a) 50% of standard or (b) 0.5% of the portfolio divided by twice the earnings-implied move (minimum 20% stress; use 25% if the implied move cannot be sourced). Use limits at $566-580 sized to the remaining gap, and finish at market by the close before the confirmed report (fallback: the Nov 2 close). No new money at $555, no $485 bid, no $599-617 limit tier. Before the print, a close below $526.83 takes half the distance to a 25%-of-standard floor and a second close finishes it; a weekly close below $453.83 exits the remainder; after Nov 3, price-based rules decide cuts and rebuilds.

Current Price: 555.6

Price Target: 583.0

Confidence: Medium

Time Horizon: 12 months

IV. Risk Management Team Decision

Aggressive Analyst

Aggressive Analyst: I'll open, since neither of you has spoken yet, and I'll take on the arguments I expect to hear.

The first thing to settle is what "aggressive" means on a Sell. At about $555 AMD has a $907B market cap, trades at roughly 22x sales and about 140x trailing operating income, and our return grid averages only 4-6% a year over three years. For a beta-2.45 stock, that is cash-like compensation for equity-plus risk. Holding full size is the comfortable, passive choice. The bold move is to recognize that the capital is badly priced for the risk and to act before the Nov 3 print. Even the bull conceded he wouldn't buy here, which tells me the case for adding is hollow.

I expect the conservative analyst to say, "If it's that expensive, exit completely." I disagree, and not out of timidity. A full exit throws away real convexity. Revenue growth accelerated to +50.1% in Q2 FY26, margins are expanding, and the weekly and monthly SuperTrends are both up, with the weekly stop at 453.83. The stock reclaimed its 50 SMA today and Q3 has been seasonally strong. Selling everything means betting that every one of those things fails at once. A loss-budgeted core keeps me in a 240% uptrend while capping the damage. That is how you stay in the game for the next leg and still harvest the valuation.

The conservative case also misses the other side of the plan. The trade isn't only about selling. I'm placing a dislocation bid at $485 or lower, sized at up to 25% of a standard allocation. If a bad print or a Meta capex scare gaps this stock down 12-15%, I want to be a buyer into forced selling, not a spectator who sold everything and has no framework to re-enter.

The neutral analyst will probably say it's balanced: strong fundamentals, a stretched price, so hold and see. That sounds sensible, but it isn't neutral. Holding full size into a binary event, with a 4.8% ATR, an unknown implied move, and a stock that already dropped 13% in a week, is a decision to take maximum variance for a low expected return. The neutral view also treats the bullish evidence as equal weight to the bearish evidence, when it isn't. Today's bounce looks like a relief rally. MACD histogram is -13.11, price sits below the 10 EMA at 566.57, and the daily SuperTrend only flips on a close above 617.04, about 10% away. I'll concede the volume point is softer than it looks, since the 10.1M-share bar may be unfinished. I don't need it, because the trend and momentum readings carry the argument without it.

Take the bulls' best card, the 35x forward P/E and 0.53 PEG. Forward EPS of $15.76 is about four times trailing GAAP EPS and nearly three times the Q2 GAAP run-rate. That implies about $26B of net income against $6.4B over the last twelve months. It's probably non-GAAP, so it isn't like-for-like, but either way the price assumes a giant earnings step that hasn't been reported yet. Meanwhile, Q2 net income exceeded operating income by about $0.27 a share of unexplained benefit, FCF fell 39% sequentially as capex doubled, and insiders sold about $335M this year with no buys in 17 months. None of that kills the story, but it means the margin for error is thin.

Sentiment backs the trim too. StockTwits shows 14 bullish tags and zero bearish, and posters are dismissing the "AI bubble" talk and predicting short squeezes. When the crowd has no bears, I get paid for being the one who sizes down. Susquehanna's $620 target is only about 12% above the current price, which isn't much reward for this risk.

I'll acknowledge the gaps. We lack the options-implied move, consensus, segment mix, verified rate data, and any explanation for the Oct 15 gap down. That's exactly why the plan is a trim to about half of standard size, capped so a 20% earnings gap costs no more than 0.5% of the portfolio, rather than a naked bearish call. It also has hard triggers: cut to 25% on two closes below 523.78 or a Meta capex-digestion signal, exit on a weekly close below 453.83, and stop trimming on a close above 617.04 on rising volume.

I'd rather be the analyst who reduces exposure at 140x operating income with a defined plan to buy dislocation than the one who rides a coin flip at full size and calls it patience.

FINAL TRANSACTION PROPOSAL: SELL Aggressive Analyst: I'll give back what you two have fairly taken first, because the rest of my case is stronger without the weak parts.

The conservative is right that selling one-third of the reduction now leaves you at about 83% of standard until the stock rallies 8 to 11%. I can't call this a relief bounce and also need it to run to 599 or 617. Both of you are right that the $485 bid collides with my own rule to cut at 524, and that it would fill on a gap before anyone knows whether Meta or AMD's own guide caused it. I'm also dropping "Q3 is seasonally strong" and the 50 SMA reclaim. Nothing in our reports supports the first, and the second is half an ATR on a possibly unfinished bar. The case doesn't need either. It rests on price against a 4-6% return grid, a binary event in about two weeks, and an implied move nobody has looked up.

So the mechanics change. Sell about half the reduction in the next session or two, taking 100 to roughly 75. Put the next tranche on limits around the 10 EMA at 566 to 580, not 599 to 617. Finish with a market order keyed to the confirmed report time, because if AMD reports after the close on Nov 3, "the close before" is that same day's session. Check the Meta date today as well. Replace the $485 bid with a post-print buy ladder triggered by what guidance says, not by price.

Where I won't move is the conservative's 25%-or-flat. Our own grid says AMD returns 4-6% a year. That is inadequate pay for the variance, but it isn't negative. Going flat is a call that expected return is below zero, and nobody has shown that. Meanwhile the weekly and monthly trends are up, and the biggest gaps in the window were upward. The reward-risk comparison also used the wrong marker. Susquehanna's $620 is one firm's number, and the stock closed at 649 two weeks ago. Against the real high, the upside is +17% to +18.5%, and the weekly stop is 18% down. That is roughly symmetric. Only the 200 SMA is 28% away, and that is where a trend break ends, not a marker for the next 13 days.

On the 20% gap as a central case, the 18.6% slide was a grind with one roughly 5% gap-down, not a series of gaps. Twenty percent is about 4.2 ATR, so I treat it as a stress test, which is how the plan uses it. The conservative is right that it runs through 453.83. That is why the cap is computed on the gap and why the plan calls the weekly stop a thesis check, not a loss limit.

The conservative's growth arithmetic is correct. At Q2's 12.5% sequential pace, revenue is about $13.0B, only +40% against $9.25B, and holding +50% needs about 20% sequential growth. I'll use that against the neutral. His "company growing 50%" leans on a Q2 FY25 base that was the softest in the series, with a 39.8% gross margin and an operating loss. That is a weak reason to hold more. The reason to keep a core is that margins are expanding, incremental operating margin was about 40%, and a 40% headline on 12.5% sequential growth isn't a miss in dollars. We don't have consensus, so we can't say how the market will read it. That is the argument for halving, not for exiting.

On the neutral's 40% versus my 50%, the difference is 0.1% of the portfolio on a 20% move either way. That isn't something the evidence can settle. The 0.5% cap, the 0.4% and the 0.25% are all mandate choices, and I won't pretend the data picks 40 over 50. If your risk budget says 40, run 40 with these mechanics. What we can actually argue from the data is the shape: front-load, sell near the 10 EMA, back it with a date-based order, and rebuild after the print.

I'll take the neutral's put-spread idea further, and it turns the conservative's best point into a tool. He's right that a share with a stop is not an option payoff. So if the implied move is about 10% or less and puts are fairly priced, hedge the last tranche with a put spread instead of selling it. That is real convexity, and for a holder sitting on a 240% gain it avoids a taxable sale. If the implied move is 14-15%, premium is too rich and you sell the shares. We need to pull that number before the final tranche.

My case doesn't depend on rates, the StockTwits tags or insider selling. I agree none of those should carry weight when FRED is down, the sample is thin, and the selling looks formulaic. A flat position also carries a cost. It means buying back after the print, probably higher if the print is good, with no core to benefit from the upside gaps this stock has already shown.

FINAL TRANSACTION PROPOSAL: SELL Aggressive Analyst: Let me start with what I'm taking, because the conservative and the neutral are right about the order of operations, and fixing it makes my own case stronger.

The conservative is right that "check the Meta date today" and "sell over the next session or two" don't fit together. The retail posts about a Meta after-hours reaction run from about noon to 1:20 Eastern, and one poster quotes 555.55. That suggests our close is a midday print and today's bar is unfinished. If so, the 10.1M volume is mostly a clock artifact, and I'm dropping it from the thesis entirely. I'm also dropping "relief rally," because I can't read a bar that's still printing. The tape evidence I'd keep comes from completed bars: the unexplained 10-15 gap down on 28.8M shares, and 18.6% lost in eight sessions with no earnings.

So the lookups go first, and they take minutes: Meta's report time, AMD's confirmed date and time, the implied move, and consensus. Today is a Wednesday, which is a likely Meta night, but nobody has verified it. If Meta reports tonight, I'd be at about 60 of 100 before the close. If it doesn't, I'd get to 70 over a session or two.

On 50 versus 60, I side with the neutral. The conservative's four-times test treats a limit order as a bet that loses 13% against a 3% gain. But a gap up fills the limit at the open, well above 573, so the order only underperforms if the stock falls. That makes it a directional call, and nobody has argued for negative drift. What's left is a variance budget. Ten points of standard through a 5-13% Meta reaction is roughly 0.03-0.07% of the book, assuming a 5% standard weight. If you want 50, take 50. I wouldn't spend more time on that gap.

I'm also taking the conservative's rebuild point. A 40-versus-50 growth gate fails on arithmetic, because 12.5% sequential growth gives about $13.0B, only +40% against a base that included the one 20% sequential quarter in the series. Write the gate in dollars before the print. My proposal is revenue near $13B, gross margin at or above about 53%, operating margin at or above about 17%, and a Q4 guide measured against consensus once someone pulls it. Those thresholds anchor on Q2's 53.8% and 17.25%, so they are my proposals, not data on what the market expects. I'd add a price condition too, since the same good print is a different buy from a stock that has already gapped up than from one that is 12% lower.

The put spread comes in at half credit, as the neutral said. A 525/450 spread covers about two-thirds of a 20% gap and under half of a 30% drop. The tax-gain argument depends on a holder we know nothing about, so I'm withdrawing it.

Here is where I'm staying put. The conservative's 25-33% floor comes from a 30% stack: a 20% gap plus a week of follow-through, treated as something you can't act on. The gap is the part you can't act on. After that you can, so the neutral's post-print rule is the right answer. I'd make it two-sided. If AMD's guide lands below the bar we wrote beforehand, cut to a quarter within two sessions with no open-ended judging. If the numbers clear the bar and the stock still gaps down on a sector or Meta-driven move, hold the core and buy in steps. That is what my $485 bid was trying to do, minus the blind buying.

I also don't want the stress number to be a fight between 20, 25 and 30. Let the implied move set it. A reasonable rule is stress at about twice the implied move, with a floor of 20%, and the weight is 0.5% divided by that. If the market prices 10-12%, you get 2-2.5%, which is the trader's number or the neutral's. If it prices 15%, the stress is 30%, the weight is about 1.7%, and the conservative is right. That is my heuristic, not a law, but it lets data decide.

On "thin edge on huge variance means a small core," I only half agree. Our own grid says 4-6%, which the trader calls close to cash. That doesn't prove the edge is zero, and it's an average across scenarios, not a point forecast. And "opportunity cost versus lost capital" is an asymmetry only if your mandate says so. Fifteen points of standard move the book by the same dollars up or down. A stability-first mandate is legitimate, and nobody has given us the portfolio context. If the mandate is strict capital preservation, go flat. We agree on that. Otherwise the lower-of rule already builds in the 0.5% cap, and nobody should have to rebuild a core after a good print.

The conservative also made the best argument for convexity. A beta-2.45 share is linear, and a stop doesn't turn it into an option. So for the upside you don't want to hand back, if the implied move is moderate, a small call spread paid for out of the same 0.5% budget is a cleaner way to stay in than 15 extra points of shares. It's fairly priced, not free, and the premium counts as capital at risk. If the implied move is 14-15%, skip it and just sell shares.

So the plan is this. Today, do the lookups. Sell to about 60 before the close if Meta is tonight, or to 70 over a session or two if it isn't. Put roughly 15 on limits at 566-580. Finish at market on the confirmed AMD report day. The final target is the lower of half of standard or the budget weight from the stress rule. There is no new money at 555, no $485 bid, no 599-617 tier, and no seasonality or volume arguments. Anyone already at or below target does nothing. After the print, follow the two-sided rule and the dollar gate.

We all say sell, and the remaining disagreement is worth a few hundredths of a percent of the book. The reason to act now is that the reward is cash-like, the event is binary, and we have time to get the order of operations right.

FINAL TRANSACTION PROPOSAL: SELL Aggressive Analyst: Before I argue anything, here is what I'm taking from you, because you've both earned it.

The conservative is right that tranches should be written as fractions of the distance to target. Sixty today plus fifteen on limits lands at forty-five, which either sells through a fifty target on a bounce none of us can read or leaves a forty target short. I should have absorbed that when the neutral raised it. Size the limit tier to the gap between sixty and the target, cancel it on arrival, and let the dated market order finish.

I'm also taking the attribution trap. My "hold if it was sector-driven" carve-out is the story people tell themselves to justify holding, and the 28.8 million share gap on the fifteenth is a live example that nobody can explain. It goes. The neutral is right that a gate anchored to Q2's 53.8% and 17.25% is necessary but not sufficient, because consensus decides the rest. One nuance is that forward EPS of $15.76 is probably non-GAAP and covers a forward year, so "three times Q2" overstates how much of a step-up the next print must show. But the gate still can't carry the decision alone.

The call spread comes out of the baseline. The neutral is right that selling fifteen points of shares and buying back a fraction of that delta in calls is a smaller sale with more moving parts, bought into pre-earnings implied volatility that we haven't priced. The put spread gets less than half credit until someone has the chain, since it pays nothing between 555 and 525, and that zone contains a gap the size of the one we just saw. Both are optional extras for someone with a chain in hand, and the baseline is shares.

On sequencing, I'm fine with sixty today unconditionally. My Meta-dependent split was only there because I hadn't done the lookups. Put Microsoft, Alphabet, Amazon and the October FOMC on the list, but all four dates come from memory, not our reports. I agree with the neutral that "more events means sell sooner" has no stopping point, because the target and the budget are what stop it. Don't size off the 78% December hike print either. FRED was down and nobody has verified it.

Now the places where I'd still push.

The post-print rule is the live argument. I side with the neutral over the conservative on the first step. An in-line guide is the base case, and cutting to the floor on "at or below consensus" double-counts the valuation worry you already acted on. "In line" needs a band, though. I'd say a guide midpoint within roughly 1-2% of consensus, which is my rule of thumb and not data. But the neutral's second step has a flaw he hasn't noticed. The 50 SMA is under 3% below price and rising. Any gap of 7% or more from here opens below it, so "closes below the gap-day low or the 50 SMA" fires almost automatically. The rule quietly becomes "gap down 7%, go to the floor." If you want that, say it. I'd use the gap-day low as the second test and drop the 50 SMA from it. Cut half the distance to the floor at the open, and cut the rest only if the stock closes under that day's low. That is still attribution-free. It just refuses to sell the low of a session the stock hasn't yet confirmed, and it fails fast when the stock does confirm.

On Meta, I agree the trader's trigger needs numbers. A cut or flattened capex range fires it and nothing else does. We don't have AMD's customer mix, though, so don't lean on Meta as a proxy for AMD's demand. If Meta holds its range and AMD drops 7% or more the next day, the same price rule applies.

On keeping a core, the conservative is right that the big gaps up came off 348 and 421, not 555. I'm not asserting that a gap up is the base case. I'm keeping a core for the reason the neutral gave. We can see accelerating growth, roughly 40% incremental operating margin, a clean balance sheet and weekly and monthly trends still up. We can't see the $0.27 of unexplained EPS, the Q2 capex doubling, or what moves the multiple next. When we can't tell which way it breaks, a budgeted core costs a few hundredths of a percent of the book. Going to the floor and rebuilding slowly after a good print isn't free either, but it is a price worth paying to cap the other tail.

The remaining unknowns are the implied move, consensus, the confirmed AMD date and time, Meta's capex range and the other dates. They should set the target and the hedge decision, but they shouldn't hold up the first sixty. That sale is justified by price against a 4-6% return grid alone, so do the sale and the lookups in parallel.

This is the bold version of a trim. Sell forty points of standard at 555 into variance the grid says you aren't paid for, keep a core sized so a stress gap stays inside the budget, and write the rebuild and cut rules before the print, while nobody is emotionally attached to the answer.

FINAL TRANSACTION PROPOSAL: SELL Aggressive Analyst: I'm taking most of that, because the last round made the plan better.

The conservative is right that a rule calibrated on a gap that went from a 5.4% open to 7.4% under the old close by the 16^th can't be written to look only at the open. The neutral is right that a close can't print below its own day's low, so my gap-day-low test only works as a later close. I'm fine with the 50 SMA clause being gone, with look-through exposure going into the definition of standard once, and with Meta's trigger written as a number against its last capex range. I also accept pulling the earnings-implied move the same way twice. One practical addition on the re-pull: if the second pull raises the target, nobody buys back the day before the print. You cancel whatever limits are still resting. That gives the conservative his ratchet without a buyback.

Here is where I still push, limited to things that change an order.

First, the 5% and 7% post-print triggers. The conservative justified 5% as half the expected swing at a 10% implied move, but we don't have the implied move. If it comes back at 14%, 5% is a third of the swing and fires on an ordinary outcome. We also calibrated these on a gap with no earnings in it. So write them as half and seven-tenths of the earnings-implied move, with 5 and 7 as floors and as defaults if the lookup fails. At a 10% implied move it's the identical rule. It uses the same logic as the stress rule we already adopted, one input instead of hand-picked numbers.

Second, the guide-versus-consensus trigger. Both of you kept "guide below consensus, go to the floor in the first session." But the conservative's own argument in his last message was that for a stock up 240% the bar that matters is the buy-side's, and the tape is the only way to read it. Consensus is a sell-side number. If the guide is a hair under and the stock is up 6%, that tree sells to the floor into strength on a story, which is the attribution trap with a different label. I proposed a guide-below-bar cut myself earlier, and I'm retiring it for this reason. Use consensus as a veto on rebuilding, not a trigger to cut, and let price decide the cuts. That deletes a branch, which is what the neutral asked for. The plan already defaults to price alone when consensus can't be sourced, which suggests price was carrying the weight anyway.

Third, the pre-print levels. The conservative is right that dropping the 523.78 cut silently isn't a decision, and I'll take the neutral's half-step version. But 523.78 is a Bollinger band. It recomputes every session and will almost certainly have moved by the time it matters. Pin the rule to a printed price instead, the 10-16 intraday low of 526.83. A close under it is a close below the low of the entire sell-off. The same standard applies to the trader's clause that ends trimming on a close above 617.04. Nobody kept it, so I'm dropping it on purpose. The budget is sized for a gap, and a rally into the print makes a gap more expensive in dollars. The return grid also gets worse at 617, not better, so a rally shouldn't stop the trim.

Fourth, the clock on the rebuild. The neutral is right that "rebuild slow" without a clock becomes never. The cuts are same-session, and the justification is follow-through: 5.4% down at the open to 7.4% by the close, and 18.6% in eight sessions. The same tape shows follow-through the other way, the 43% from the July 29 close to the September 21 close, which is the conservative's own number. If follow-through is why we cut at the open, it's also why the first rebuild step shouldn't wait more than two closes above the gap-day low. Rebuild is capped at the valuation cap, so the worst case is returning to half of standard, a size we already called acceptable. I'd do the second step a week after the print, and cancel both if the stock closes under the gap-day low. That's a clock on the existing branch, not a new one.

None of this is the bold part. The bold part is selling about forty points of standard today into a reward our own grid calls cash-like. We keep a core because we can't see which way it breaks, and we write every rule before the print so neither fear nor FOMO writes it after the gap.

To keep the plan from waiting on a lookup, here are the defaults. If there's no implied move, use a 25% stress, which is about a 2% weight at a 5% standard. If there's no consensus, price decides alone. If AMD's report time isn't confirmed by Oct 30, finish the sale by the Nov 2 close. Anyone already at or below target does nothing, and a strict capital-preservation mandate goes flat. Nobody adds money at 555.

FINAL TRANSACTION PROPOSAL: SELL