Altair
Skip to content

Trading Analysis Report: SMCI

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

  • Analysis date: 2026-10-14
  • Rating: Underweight
  • Generated: 2026-10-15 00:22:29
  • 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 SMCI: cut to about 50% of a standard allocation today, worked through the session rather than at the open. Settle the core at 40% by default, or at 45% only if all three fact-based tests pass (plea, company-level action, June-quarter margin); in that case leave at most 5 points as resting sell limits at 37.42-38.53, cancelled on a close below 35.2, by Oct 23, or before the print, whichever comes first. Exit to zero at market if the company or a company-level action is tied to the plea, if the company attributes June-quarter margin to one-time items, or on a weekly close below the 200-day at 32.43; a daily close below 34.09 sheds a further third. Size the remaining core for a 30-33% overnight gap, not the 2.39 ATR; add nothing to SMCI before the September-quarter print (historically early November, unconfirmed); and treat a weekly close above 44.40 as a reason to reassess, not to chase.

Current Price: 35.78

Price Target: 33.0

Confidence: Medium

Time Horizon: 4-8 weeks

IV. Risk Management Team Decision

Aggressive Analyst

Aggressive Analyst: Since nobody else has spoken, I'll open. "Aggressive" usually gets read as "buy the dip." I think the aggressive move on SMCI is the one the trader already made: cut real risk now at defined prices, keep a smaller stake that still pays if the bulls are right, and pre-commit to the exits instead of hoping.

Start with the argument I expect to hear from both of you, that this is cheap. It's 11x trailing, 6.7x forward, with a PEG under one. But roughly 53% of FY26 net income came from a single quarter, the June one, where gross margin was about 17.5% against 6 to 10% in the five quarters before it. That's one quarter out of eight at that level, and it's a number we derived by subtracting nine-month figures from the annual, not one we read off a reported line. The trader's math says about 40% of that margin premium has to persist just to justify roughly 12x. The market isn't missing a bargain. It's pricing exactly that doubt. And the quality underneath is poor: operating cash flow of negative $6.8 billion against $2.2 billion of profit, about $9.5 billion of financing, an equity jump that looks like a raise, and a share count that doesn't reconcile (647 million versus roughly 928 million implied). On the higher count, price-to-sales is closer to 0.85 than 0.6. Add a beta of 2.0, and "cheap" looks like the discount you'd expect for this risk.

On the tape, I'd push back hard on "support has held three times." Compare today with September 15, when the close was almost identical at 35.64. RSI was 47.97 then and is 37.68 now. The MACD histogram bottomed at -0.47 then and is -0.83 and still deepening. Same price, much weaker momentum, is the opposite of a base. All three SuperTrend tiers are down, and we're below the 10 EMA and the 50 SMA. The one prior bounce from this zone is an anecdote, not a pattern.

Then there's the setup. The plea is unread, and we know it only from a headline. We don't know who pleaded or what it means for the company. The macro is hostile: a 97% market probability of no cuts in 2026, an 82% probability of another hike (thin volume, so directional only), and oil around $103. The print is about three weeks away, and this stock has gapped 33% in a session, and also from 40.64 to 29.27. ATR is 2.39, about 6.7% of price, while the trader sizes for 25 to 30% gaps, nearly four times that. Compare the payoffs: if we're wrong and the weekly SuperTrend flips at 44.40, we're giving up roughly 19% from the 37.42 entry on the trimmed portion. The left tail the trader sketches is 48 to 59%. That's about 2.5 to 3 to 1 in favor of acting.

To the conservative analyst, who I expect will say sell everything now: that is the timid choice dressed up as bold. We're at the low of an 8.4% down day, with a daily z-score of -2.82 and a close 1.64 below the lower Bollinger band. A reflex bounce is more likely than not, and the daily TD count is only at 5 of 9. Dumping the whole position here means paying for panic. The plan sells about half now and works the rest into 37.42 to 38.53, which is 4.6 to 7.7% above the close. It also goes to zero on hard triggers: the company named in the plea, the MD&A pinning Q4 on one-time items, September gross margin under 10%, or a weekly close below the 200 SMA at 32.43.

To the neutral analyst, who will say hold because demand is intact: I agree demand looks fine. The Open Compute Project expansion, the Digi Power X deal, and HPE's strength all point that way. Revenue up 78% is not the issue. The issue is what that revenue costs: gross margin down three straight years before one outlier quarter, and cash going out the door. Holding the full position is a bet on one binary outcome, and the trim keeps about half the upside while capping the damage. I'll also concede your best point, that OBV is about 112 million shares higher than at the same price in September. That's a mild constructive sign, and it's part of why I keep a stake.

I should be honest about the limits. The tools don't explain today's drop, and the sentiment read is headlines only, with low confidence. The Q4 numbers are derived, and we don't know the caller's holdings. If the plea turns out to involve an individual and not the company, the stock could rip, and this trim will cost some upside. I'd pay that price. The 44.40 weekly-close trigger means we reassess instead of chasing, and the position stays sized to survive a gap. If either of you has data that breaks the margin-persistence problem, I'd like to hear it. Otherwise, I think the bold and correct call is to sell. Aggressive Analyst: Let me start with a correction, because I argued against two people who didn't show up. I expected one of you to say dump it all and the other to say hold. Neither did. The conservative analyst wants the bulk of the cut now, a small resting piece, and a dated deadline. The neutral analyst wants the same with a deeper first cut and the plea read before anything else leaves. That's a Sell executed as a trim. What's left is how big the waiting piece is, when the clock runs out, and what the plea read changes.

First, what I'll concede. The neutral analyst is right that my 2.5-to-3-to-1 was sizes, not odds, and it flatters me. From 35.78 the 44.40 flip is about 24% away, not 19%. The two down gaps on record, 28% and 33%, land around 24 to 26, not in the 48 to 59% range, which sits below the 52-week low of 19.48 and is a scenario, not a base case. Redo the break-even with those numbers and a gap only has to be about four-fifths as likely as a run to 44, not a third. I think a gap is at least that likely with three unread documents and a print coming, but that's judgment, not data. I also can't call a bounce "more likely than not." I'll take the structure-versus-stretch defense, but it doesn't give me a probability. And the conservative analyst is right that 33 million shares against 191 to 243 million on the big days means nothing has flushed. The price-based exits are backstops, not floors.

Now the pushback. The conservative analyst's 4-to-5x payoff math is right, but it treats the waiting tranche as exposed to a gap for the whole window. That exposure is shorter than it looks. The first rung at 37.42 is about 0.7 of an ATR above the close, and the top at 38.53 is about 1.15. The stock was trading at 37.97 to 38.02 this morning. The plea and the 10-K can be read before tomorrow's open. The deadline is our choice, and the print, if it's early November as the pattern suggests, lands after it. The print date is unconfirmed, and that's exactly why the tranche should be small. The neutral analyst puts waiting at roughly a wash, which means it's a sizing question, not a reason to scrap it.

I'd also separate two kinds of triggers. A weekly close below 32.43 is a price trigger and fires after a gap, so I won't call it hard again. But "the company is named in the plea" and "the MD&A pins Q4 on one-time items" are information triggers. They can fire this week, before any price move, if someone does the reading. That reading is what turns the trigger list into protection.

On the gap-absorption test, I like it for choosing the retained size, but it can't be pass/fail. A 30% gap on 40 to 50% of a standard weight costs 12 to 15 points of that weight. Whether that's survivable depends on what a standard weight is in your book, and none of us knows that.

I'll take the neutral analyst's deadline of Friday the 23rd over end of October, and I can add to the case. October 31 is a Saturday, so an Iran ceasefire headline would reach the stock as a Monday gap. A deadline there would sit on the weekend.

On valuation, I'll side with the conservative analyst and make it worse. The vendor market cap implies about 646 million shares, which is below the roughly 684 million weighted diluted count behind the EPS. On 684 million, price-to-sales is about 0.63. On 928 million it's about 0.85, and trailing P/E is nearer 15. The cheap case gets weaker with every reconciliation. That's my arithmetic on unconfirmed counts.

My one caution on the neutral plan is the "individual or third party" branch. If the plea isn't the company's, letting the final stake sit at the top of 40 to 50 is fine. But that moves it within the range, and it isn't a clearance. A third-party plea in a $2.5 billion China AI server case could still reach the company through regulators or customers, and the headline says the case isn't going away. Who pleaded also has nothing to do with the margin outlier, the negative $6.8 billion operating cash flow, or the share count.

The tools don't explain today's 8.4% drop, the sentiment read is headlines only, and the Q4 figures are derived. With that said, here's where I land: Sell, front-loaded to about 55 to 60% of a standard weight today, 10 to 15 points working between 37.42 and 38.53, and the plea read before tomorrow's open. If the company is named, the waiting tranche goes at market and the final stake heads toward zero. Whatever remains goes by the 23rd. Nothing new before the print. A close above 44.40 means we were early and reassess, not that we chase. That keeps the upside on the half that stays and cuts the half that only hurts on a gap, and I'd call that the bold version. Aggressive Analyst: I'll start by giving back what I should. The neutral analyst is right about valuation and I was wrong. Trailing P/E is price over a per-share number, so the share-count fight moves market cap and price-to-sales, not the 11x. My 15x came from spreading the profit over 928 million shares, which is a sensitivity on an unreconciled figure, and I shouldn't have said the cheap case weakens with every reconciliation. Call it 10.7 to 15, settled by the 10-K cover, and none of us should lean on it. I'll also retire "bold and correct." A break-even of four-fifths says this is risk reduction under bad data, not conviction. What I'll defend is smaller: act now at pre-set prices instead of waiting for certainty that isn't coming.

A practical point first. The tools stop at the October 14 close, and the date I'm working from is the 15th. So Friday the 16th is one session away, and reading the plea has to happen today. The stock may also already be somewhere other than 35.78. If it has bounced into the zone, the tranche debate is over. If it has gapped lower, the first cut gets worked, not dumped.

To the conservative analyst: your tranche arithmetic is right at 10 to 15 points. I'll take 5 to 10, which cuts the gap loss to roughly 1.5 to 3 points. But "a bounce has to be five times as likely as a gap" treats the fill as a reversal bet. A resting sell at 37.42 is about 0.7 of an ATR above the close, on a stock whose ATR is 6.7% of price. It can fill on noise alone. That's my arithmetic on volatility, not tool output, and it ignores the downward drift you point to. It does change the design. Fill odds are front-loaded, because if the stock touches that zone it will most likely do so early, while gap exposure builds every day. That's why the clock should depend on the plea read, as the neutral's tree has it. Unread, the piece goes by Friday. Read and not the company, the biggest binary is gone, the gap odds on the piece fall, and I'd let it run to the 23rd.

You're also right about the quiet slide. A drift to 33 with no gap costs about 8% on whatever hasn't filled, and a TD count finishing on the 20th doesn't rescue it. That's the price of any time stop, and it's why the piece is small. I'd read the 33 million shares against 191 to 243 million the same way. No flush means a bounce isn't guaranteed, which is why the plan sells over half regardless of one.

Where I part from you is 50 now and defaulting to 40. Not knowing what a 30% gap costs in the caller's book argues for percentages and pre-set triggers, which we have, not for the lowest number in the range. Selling isn't the risk-free side either. It's a price decision too, and the extra five points would be sold in the early session after an 8.4% down day that closed 0.15 above the low, with sector and company effects we can't separate. Being early on the core also isn't cheap if the plea is clean. The stock closed at 44.94 on October 7. The annual statements behind the margin spike, the cash burn and the share question appear to have been public then. What changed since is the plea, the rates tape and the sector, so the plea is the swing factor on the retained stake, and about 25% separates 35.78 from 44.94. I'll own that this cuts both ways: it also means the stock was priced richer a week ago, and part of the drop is the market trimming its own persistence assumption. Still, I'd sit at the 55 end of the first cut and you at the 50 end. That's a five-point disagreement.

To the neutral analyst: I take your correction that I had no fallback if nobody could get to the plea. The Friday fail-safe fixes that. But I'd push back on the wording "core heads toward zero" if the company is named. The trader's rule is exit to zero, and that is the one trigger that arrives as a gap. If we soften it to "toward," we've rebuilt the hoping we pre-committed to avoid. Named means zero.

On the daily close below 34.09, I'll accept it as a sizing rule, not a signal. That level is 4.7% under the close, and this stock went from 25.50 to 37.61 within weeks of the June gap, about 47%. Noise will test it, and it may well be wrong. But it costs a quarter to a third of a 40-to-45 stake, so ten to fifteen points of weight, and it's cheap insurance against the slow-slide path the technical report lays out.

We're now arguing over five to ten points of weight, while the retained stake carries a 30% gap loss of about 12 to 14 points, several times the tranche's worst case. The next hour is better spent reading the plea than tuning the tranche.

So I land on Sell, executed as a trim. Take the first cut to about 55, worked through the early session rather than the opening print, with 10 points on limits between 37.42 and 38.53, which lands the stake near 45. If the company is named, everything goes. If the plea is still unread at Friday's close, the waiting piece goes at market and the core sits near 40. If it's read and isn't the company, the piece runs to the 23rd and the core tops out at 45. A daily close below 34.09 takes another quarter to a third, and a weekly close below 32.43 takes it to zero. Nothing new before the print, and a weekly close above 44.40 means we were early, so we reassess instead of chasing.

All of this is relative to a standard weight. If the book is already smaller than that, the trim may be done. The sentiment read is headlines only, the Q4 figures are derived, and these tools can't confirm the filings. Aggressive Analyst: We're at the point where all three of us would sign the same tree, so I'll say what I'm taking, what I'd add, and where I still hold ground.

The conservative analyst is right that I oversold my own arithmetic. Going to 55 from a full weight sells 45 points, not half. It only crosses half if the limits fill, and I shouldn't have said the plan sells over half regardless. I also take the information-driven sizing: unread means 50, read and clean means 55, named means zero. That's the pre-commitment I've been arguing for, and I was wrong to treat the lower number as a reflex. When the one fact that matters is missing, you size for the bad case. When you have it, the branch sets the size.

I'll take the flip on 44.94 too. If the same public statements supported 35.64 on September 15 and 44.94 on October 7, they don't anchor the price, so the right response is to size for the swing. The neutral analyst's version of that point lands on the same 40 to 45 core. I also accept the neutral's edges on the rules: a company-level action tied to the case, not any 8-K, and the 35.2 rule touching only the tranche.

Two additions. First, the cost of not reading the plea is permanent. Nothing gets added before the print, so a core cut to 40 in the unread branch stays at 40 even if someone reads a clean plea on Monday. That's five points of core lost for skipping an afternoon of reading, which is the strongest reason to make the read the first task today.

Second, we can turn the gap test into something the caller can use without knowing our numbers. Multiply the core, as a fraction of a standard weight, by 0.3 and compare it to the one-night loss the book can accept. A 45 core costs about 13.5 points of a standard weight. If a standard weight is 5% of the portfolio, that's roughly 0.7% of the portfolio, which is an illustration, not a known figure. If that number is uncomfortable, scale the core down proportionally. That answers "I don't know" with arithmetic instead of defaulting to the bottom of the range.

Here's where I'd still hold ground. In the clean branch I'd keep the tranche at 10 points, not 5. The conservative's 5-point math is right for a gap: about 0.2 of a point if it fills, 1.4 to 1.65 if a 28 to 33 percent gap hits. But the clean read removes the biggest source of gap risk, and the 35.2 cancel rule means a quiet slide costs roughly 1.6% on that piece, not the 8% down to 33 we worried about earlier. So it's a cheap, capped bet that the sector part of today's drop mean-reverts. If the book can't stomach the variance, take 5 and nobody should lose sleep over it. We've spent four rounds on a piece that is about a tenth of the core's gap exposure.

I agree with the neutral on the 23rd as a ceiling, not a target. A level 4.6% up on a stock with a 6.7% daily range tends to get touched early or not at all. I also agree on the gap-lower rule: in the unread branch, treat it as company news and sell the tranche. In the clean branch, a lower open is more likely sector, so you work it.

I'm also done saying "bold and correct." The core is where the bet lives. If the plea is clean, the filings hold up, and the stock gets back toward 44.40, 45 points gain about 11 points of weight. A 30% gap costs about 13.5. That's close to even, which is why the core is 45 and not 70, and why this is risk reduction, not conviction. What I'll defend is the structure. The downside is sized before the news. Every rule is tied to a price or a fact. The part that pays if the bulls are right is still there. Waiting for certainty the tools can't deliver is the passive choice, and the pre-set tree is the active one.

If the caller remembers only three things, they should be these. Read the plea today. Named means zero. Anything unresolved at Friday's close gets treated as unread.

The limits haven't changed. The tools stop at the October 14 close, so SMCI may already be somewhere else this morning. If it bounced into 37.4 to 38.5, the tranche filled for free. If it gapped under 35.2, that rule has fired and we're in the unread branch until someone shows otherwise. The Q4 figures are derived, the plea is known only by headline, and everything is relative to a standard weight in a book we can't see. Within those limits, I'm a Sell, executed as a trim. Aggressive Analyst: I'll start with what I'm giving up, then say where I think each of you still leans too far.

Neutral, you're right that the tranche and the first cut are one decision. I had 55 with ten points working, you have 50 with five working, and both land on a 45 core if the limits fill. Four rounds of argument come down to whether five points sell at market today or sit on limits. I'll take your version. I'm also dropping my 1.6% line for good. That was the distance to the 35.2 trigger, not the cost of the exit. A close-based stop sells next session, on a stock that averaged about 4.5% down a day for five sessions. And Conservative, you were right that the sector leg won't mean-revert on its own. It's rates, with 97% on no cuts and 82% on another hike, and a clean plea doesn't touch that.

Neither of you noticed that you've both treated gap risk on the tranche as one-sided. A resting sell limit that gets gapped through fills at the open, so the tranche has gap upside too. The technical report lists four big gaps: down 33% and 28%, up 19.8% and 19%. If an up gap lands, the five points sell near 43, about a point of weight better than selling at 35.78 today. If a down gap lands, they lose 1.4 to 1.65. With two of each on record, that nets to roughly a quarter to a third of a point lost per gap, and a normal fill adds a few tenths back. That's a wash, slightly negative with the plea unread, so five is the right size and the plea read is the tiebreaker. It's the tree Neutral wrote.

The same arithmetic is my best argument for keeping a core. Selling 50 points from a standard weight avoids 14 to 16.5 points of loss on a down gap and gives up 9.5 to 10 on an up gap. If they're equally frequent, as they are in a four-event sample, selling wins by four to seven points per pair. That's an edge, not a landslide, which is why this is risk reduction and not conviction. Conservative, if the bad outcome alone sets the size, the answer drifts toward zero. But the stock also went from 25.50 to 44.94 in under three months, and two of those steps were single sessions on 160-plus million shares. The core is the part that gets paid if that version is true.

On defining "clean," I side with Neutral. "Anything ambiguous goes to unread" makes the clean branch unreachable, and nobody can verify customer fallout this week. Use the three tests that can be read today. The plea names someone other than the company, there's no company-level action tied to the case, and the MD&A doesn't pin Q4 on one-time items. The 45 cap covers what we can't check. I'd weight the MD&A test as heavily as the plea, because it tests the number the cheap case rests on. Q4 annualizes to about $6.70 a share, the nine-month run rate to about $2.11, and consensus at $5.33 sits roughly 70% of the way to the Q4 level. That's my arithmetic and it ignores revenue growth and share count. Still, it says the 6.7x forward multiple is only cheap if most of the outlier repeats. The MD&A is the one document that tests that, and it's an hour of reading.

Conservative, on the door you found in the trader's text, I'd read it more narrowly. The quarter-size trade needs clean filings and a close above 38.53, which is the 50 SMA and 7.7% above the close. So if you skip the read and the plea turns out clean, you can only get back in after the stock has already recovered. That's the real cost of the unread branch, and it's why the read comes first. On insiders, I'm with Neutral. A $7.7M sale against a $23B cap, and no buying even near 19.48, tells us nothing at 35.

Neutral's point about the proceeds is the one that decides whether this is risk reduction at all. If 50 points of weight go into another high-beta AI hardware name, the rates leg of today's drop is rebuilt and the sale accomplished nothing. The tools don't tell me where the money should go, so I won't name a place. But the freed capital is the aggressive part of this plan. It's optionality for after the filings are read and the print is out, when the margin question can be underwritten with a reported number instead of a derived one.

Here is the tree I'd sign. Read the plea, the MD&A and any company-level action first, then place the order, worked through the session and not hit at the open.

  • Exit to zero: if the company is named, there's a company-level action tied to the case, or the MD&A pins Q4 on one-time items, everything goes at market.
  • Otherwise: sell to 50.
  • All three tests pass: the core is 45, and the five points above it rest between 37.42 and 38.53, with the 23rd as a ceiling, not a target.
  • Anything unread or unclear when you place the order: the core is 40, and the ten above it are gone by Friday's close.

In every branch, a close under 35.2 ends the waiting on the tranche only. A daily close under 34.09 takes a third of what's left, and a weekly close under 32.43 takes the rest. A weekly close above 44.40 means we were early, so we reassess instead of chasing. Nothing new goes into SMCI before the print.

The limits haven't changed. The tools stop at the October 14 close and today is Thursday the 15th, so SMCI may already have bounced into the zone or gapped under 35.2. The Q4 figures are derived, the plea is known only by headline, and everything is relative to a standard weight in a book we can't see. If the position is already smaller than a standard weight, the trim may be done. I'm a Sell executed as a trim, and the piece I'd defend hardest is the core.