Daily Analysts — Thursday, November 12, 2026 — Midday Edition

AMD at $584: The Data-Center Flip Is Real — Trim Into Strength

Single-stock deep dive (Option A). Prices intraday ~16:00 UTC Thu Nov 12 (11am ET): AMD $584.07, NVDA $165.66, AVGO $363.32. Q3 is the primary source.

The call: DAILY-AMD-1105 just triggered — AMD printed $584.58 high vs our $580 take-half level. Sell half $580–600, hold rest to $620 with a $500 trail. No new chase. HIGH conviction on the hold.

Why now: Q3 Data Center $8.55B (+97% y/y, 64% of sales) plus a $17B Q4 guide met William Blair's upgrade on Nvidia's leadership vacuum — and the stock is doing exactly what the note predicted.

The disagreement: Consensus now treats AMD as an Nvidia hedge. We think it's a standalone EPYC + Helios ramp — agentic CPU demand plus 14 GW of contracted Helios — that works even if Nvidia stabilizes.

The level that changes everything: $500. A weekly close below it means the Helios ramp or the supply chain broke. Above it, the 2027 doubling thesis is intact.

Our $580 half-take just hit — take it

AMD touched $584.58 this morning, inside our $580–620 target zone. Discipline means selling some into the upgrade-day spike.

On Nov 5 we bought the thesis at $404.40 (DAILY-AMD-1105): EPYC share plus MI450/Helios optionality, target $580–620. That is +44% in seven sessions. This morning William Blair upgraded AMD to Outperform citing Nvidia's "potential leadership vacuum" and CPU-attach opportunity, while Bernstein and Oppenheimer downgraded Nvidia after Jensen Huang resigned and CFO Colette Kress took over as interim CEO. AMD is +5.6% to $584.07 while Nvidia is −4.6% to $165.66 (low $148.80 — which, for the record, filled tranche 1 of our staged NVDA value long at $148–152).

What to do, concretely: if you followed DAILY-AMD-1105, sell half between $580 and $600 today. Move the stop on the rest to $500 (from cost). New money: do not chase the Blair spike — wait for $520–540. My opinion: upgrade-day gaps on a 120x trailing-P/E stock fade more often than they extend.

Q3 proved the mix flip: AMD is now a data-center company

Data Center is 64% of revenue, up from 47% a year ago — the quarter AMD stopped being a PC company with an AI kicker.

Q3 2026 (primary: IR press release, Nov 3): revenue $13.38B (+45% y/y, above the high end of guide), non-GAAP EPS $2.09 vs $1.95 expected (+74% y/y), non-GAAP operating margin 29% (+5 pts y/y). Segment split tells the story:

SegmentQ3 revenueSignal
Data Center$8.55B, +97% y/yRecord EPYC + first MI450 Helios revenue; operating income $2.88B (+168%)
Client + Gaming$3.72B, −8% y/yClient +7% (Ryzen mobile); Gaming −40% — PC/memory softness Intel already warned on
Embedded$1.11B, +29% y/yAerospace/defense-led; stable, not the driver

Q4 guide is ~$17B (±3%), +66% y/y and +27% sequentially — well above the ~$16.2B Street mean — at ~55% non-GAAP gross margin. Lisa Su on the earnings call (CNBC recap): "we expect growth to accelerate further in Q4 as we scale Helios and Venice production," server revenue to double y/y in 2027, and supply (wafers, substrates, packaging, memory) secured for "much higher growth next year." Consequence: every extra Data Center dollar now drops ~34 cents to segment operating profit — operating leverage the PC cycle never gave AMD.

Why this week is different: contracted gigawatts plus a competitor in chaos

Two independent signals agree — the order book and the vacuum — which is why this stays HIGH conviction even after a 44% run.

1. The book is contracted, not hoped. Advancing AI 2026 (IR, Jul 23) plus updates since: OpenAI 6 GW of Instinct GPUs, Meta up to 6 GW (custom MI450 + Helios), Anthropic up to 2 GW of Helios (with AMD investing up to $5B), plus Microsoft, Oracle (50k GPUs) and Vultr/HPE's first Helios order. Meta's Helios racks are already running, per CNBC. That's ~14 GW of named demand before counting Microsoft/Oracle. Agentic AI is the CPU kicker: Su said agents run for hours on virtual machines powered by EPYC — "adoption is just starting." My opinion: the market still models AMD as a GPU laggard; the 2026 surprise is EPYC attach per rack.

2. Nvidia's vacuum is a procurement event. Huang's resignation, Kress as interim, no named permanent successor, two downgrades, Congress threatening subpoenas and export bills — BBC's successor rundown puts it plainly: boards re-qualify sole-source suppliers when governance breaks. William Blair's logic — scrutiny + vacuum = share + CPU attach — printed the same morning AMD spiked. Consequence: even a two-quarter delay in Nvidia's Vera Rubin qualifications pushes 2027 hyperscale capex toward the only qualified second source with rack-scale systems shipping.

AVGO's +5.2% to $363.32 on the same tape confirms it: the market is paying for every non-Nvidia AI path. Our DAILY-AVGO-1109 ($300–320 → $380–400) is now halfway there.

What $584 prices — and why we trim but don't exit

At $584 AMD discounts a credible 2027, not a fantasy one — which is exactly when you sell half and let the rest prove it.

Arithmetic, checkable: 9-month revenue $35.17B + $17B Q4 midpoint ≈ $52.2B 2026. Q3 non-GAAP EPS $2.09 (+74%); assume Q4 ~$2.40–2.60 on the $17B guide and you get ~$7.50–7.80 full-year non-GAAP EPS. At $584 that's ~75x 2026 — full, but directionally right if Su's "server doubles in 2027" plus Helios delivers 50%+ revenue growth with margin expansion. Finnhub trailing P/E 120.7x and 3-yr EPS growth 46.7% frame the same point: you are paying for 2027–28, not 2026.

My opinion on fair value: $620 is ~80x my ~$7.75 2026 EPS, or ~50x a plausible $12+ 2027 EPS if Data Center doubles — rich but reachable on a platform ramp with 29% operating margins expanding. $500 is the line: ~64x 2026, where the ramp is doubted. That's why the trail sits there. The load-bearing assumption is that Helios ships in volume Q4–Q1 (HPE/Vultr, Meta, OpenAI first gigawatt). If Helios slips — KeyBanc chatter in October about <1k racks for Q4 is the canary — 2027 EPS falls ~25% and $584 becomes ~100x. I name it because it's the one input that flips the answer.

Insider yellow flag, not red: CEO Lisa Su sold ~13,494 shares Sep 10 at $506–514 and gifted 35,000 shares; EVP Forrest Norrod sold ~17,500 shares Sep 15 at $498–507 (IR/SEC filings via Finnhub insider feed). Pre-quarter 10b5-1 selling at $500s, not panic dumping — but no open-market buys into the ramp. Watch Q4 blackout lift: buys confirm, silence doesn't.

The trade: sell half into $580–600, ride the rest

The edge was $404; at $584 the edge is position management, not entry.

FieldDAILY-AMD-1105 (updated Nov 12)
ActionSell HALF $580–600 today. Hold half to $620; stretch $650 on a Helios beat. Trail stop $500 (weekly close).
New moneyWATCH. Enter $520–540 on a pullback only. Do not chase upgrade-day.
InvalidationWeekly close <$500, or Q4 Helios revenue guides flat, or TSMC/memory allocation cut forces Su to walk back "much higher growth next year."
Timeframe1–3 months to $620
ConvictionHIGH (two signals: contracted gigawatts + record DC leverage; confirmed by Blair upgrade + price trigger)
SizingHalved position = full risk on remainder. If S&P <7,300 (book rule), cut remainder by half regardless.
Scenario to year-endProbTriggerAMD
Bull30%Nvidia names no permanent CEO in 2 weeks + Meta/OpenAI confirm Helios at scale; Q4 $17B beaten$620–680
Base50%Helios ramps as guided; Nvidia interim steadies; memory costs contained$540–610 range
Bear20%Helios <1k racks, or margin guides down on rack mix, or insider selling accelerates post-blackout$460–500 (stop)

Expected value ≈ 0.30×$650 + 0.50×$575 + 0.20×$480 = $578.50 — essentially today's price. That math is the trim call: no edge to adding, large edge to holding a trailed runner.

What would prove me wrong

  1. Weekly close below $500. The ramp thesis breaks; exit remainder, don't average.
  2. Q4 print or pre-ann guides Helios flat sequentially. The $17B guide assumes a "very significant step-up" (Chipstrat on the call); flat means the gigawatts aren't converting.
  3. Non-GAAP gross margin guides below 53% for Q1. Rack mix + HBM cost overwhelming EPYC mix would kill the leverage story.
  4. Nvidia names a Buck/Kress permanent ticket within days and keeps guarantees. The vacuum trade unwinds; AMD still works but $620 stretches to 2027 — take the second half at $580.
  5. Anthropic IPO (Nov 19) reprices circular financing. AMD has up to $5B committed into Anthropic against 2 GW of demand; if the S-1 loop (Amazon/Google/Microsoft paid $36B in, owed $431B firm out) spooks AI capex, neocloud orders — and CRWV alongside — fall first.

What to do with the rest of the book

AMD's spike pays for patience elsewhere — and Nvidia's low filled our first buy.

Appendix

A. Data snapshot · B. Model · C. Open-call reconciliation · D. Sources

A. Data snapshot (intraday, ~16:00 UTC Thu Nov 12)

ItemLevel
AMD$584.07 (+5.64%; O $569.97 H $584.58 L $554.90; prev $552.90)
AMD fundamentals (Finnhub)TTM P/E 120.65; TTM EPS $4.60; 3-yr EPS growth 46.65%; rev growth 41.87%; gross margin 53.72%; beta 2.63; 52-wk $188.22–$658.52
AMD Q3 / Q4Q3 rev $13.377B +45%, DC $8.546B +97%, non-GAAP EPS $2.09; Q4 guide ~$17B ±3%, GM ~55%
NVDA / AVGO / CRWV / LRCX$165.66 −4.65% (L $148.80) / $363.32 +5.24% / $56.99 −6.22% / $238.46 +1.09%
SPY / QQQ / DIA / IWM$760.74 +0.12% / $691.73 +0.20% / $525.67 +0.23% / $280.89 −0.78%
BTC / ETH / SOL; fear; total cap$71,593 −3.17% / $2,003.79 / $81.42; Fear 28; $2.406T −5.09% 24h

B. Model and load-bearing input

2026 bridge: 9M GAAP revenue $35.17B + Q4 $17B ≈ $52.2B (≈+60% y/y on implied ~$32.5B 2025). Non-GAAP EPS: 9M ≈ $5.41 ($2.09+$1.66+~$1.66 implied Q2) + Q4 ~$2.40–2.60 ≈ $7.80. $584 ÷ $7.80 ≈ 75x. $620 target ≈ 80x 2026 or ~50x $12+ 2027 bull EPS. Load-bearing: Helios volume Q4–Q1 (first gigawatt OpenAI/Meta/Anthropic). If wrong (slip to H2), 2027 EPS ~$9 not $12+, fair ≈ $450–500 — the stop zone. Warrant drag: $1.097B fair-value charge in Q3 (customer warrants remeasured on price) — GAAP EPS understates run-rate; use non-GAAP.

C. Open-call reconciliation

D. Sources

Opinion, not investment advice. Numbers intraday ~16:00 UTC Nov 12 and will move. Weekdays verified: Thu Nov 12, Fri Nov 13, Tue Nov 17, Wed Nov 18, Thu Nov 19. — @dailyanalysts · 中文版