Good morning. The tape's mood this morning is defined by one tension: hedge funds just dumped the most tech since SEPT 2024 (historic degrossing), yet the AI-infrastructure datapoints overnight are as strong as anything we've printed all year. MXL is the earnings story — a genuine blowout, revenue +72% Y/Y and +29% Q/Q with AI datacenter infrastructure revenue up 210%, EPS $0.62, and a Q4 guide of $240-260M that implies acceleration, not deceleration. That's the highest-conviction optical datapoint in the window, and it rhymes with Innolight/Eoptolink talking order backlogs into 2027 and customer guidance into 2028 out of Asia overnight. Speaking of Asia: the China optical chain put out coordinated market-stabilization statements (Eoptolink even announced a buyback), and semicap China revenue at ASML/LRCX surged +82% Q/Q in 3Q26 — China dollars still growing even as the mix declines. Macro is quiet; the AI trade is the macro.
Themes for the day. One: positioning vs. fundamentals — the degrossing print plus the "Perkini bubble" framing is exactly the kind of sentiment washout you see near tradable lows, and Nomura is flagging RESUMED AGGRESSIVE CALL BUYING in CRWV/NBIS while Convequity argues the neoclouds now trade inverse to AI demand because stronger demand widens the working-capital hole (CRWV: $35.6B debt vs $5.5B cash — this is a funding trade, not a demand trade). Two: optical is the cleanest way to be long AI infra right now — MXL's Keystone PAM4 ramp plus the 1.6T commentary out of China gives you confirmation at both ends of the chain. Three: the hyperscaler funding gap is widening — ORCL spent 161% of operating cash flow on capex vs AMZN at ~107% and MSFT at ~63%, and AMZN abandoning Loudoun County sites is either supply discipline or power access, but either way it sits awkwardly against the 217 GW-by-2030 consensus. Four: AI is now showing up in the P&L the other way — AXP, BCS, EBAY, NFLX all cutting jobs with AI explicitly cited, which helps the enterprise-ROI narrative and hangs over the whole tape as a sentiment tax.
We'll hit up MXL and the optical chain first, then CRWV/NBIS/IREN and the neocloud funding debate, then get to semicap (ASML, LRCX) and the hyperscalers...
SPOT delivered the quarter the setup needed. Stock +11% to ~$583, $120B market cap, and the Street spent Wednesday/Thursday chasing PTs higher. Q3 itself was a beat-and-moderate-raise affair: revenue and op income in line, but the guts were better — PREMIUM SUBS BEAT BY 1M, total gross margin +140bps y/y and 10bps above guide, and the Q4 bogey embedded the opex moderation mgmt had telegraphed. Q4 EBIT guide of €820M cleared both consensus and Raymond James' model. Guide implies 10M premium and 30M MAU net adds in Q4. Ad revenue finally inflected — +9% y/y ex-FX, the "second-half recovery" mgmt promised, driven by automated channels and podcast sponsorships (music ad pricing still soft, per Canaccord).
Eight shops moved, and the split tells you everything about the debate. The bull cluster — Canaccord $720, Morgan Stanley $700, Guggenheim $650, Citizens/Raymond James $625 — is underwriting the pricing-power and margin-pacing story. The neutrals — Piper $580, Rosenblatt $576, Cantor $550 — moved targets up but won't chase, with Cantor explicitly at 25x FY27 EBIT and Piper's raise driven by a higher FCF multiple rather than estimate revisions (their estimates actually went lower — revenue missed their bogey).
Guggenheim put the bull case cleanest, rolling to ~29x 2028 EPS:
"Results keep Spotify on track for its 2030 framework targeting mid-teens constant currency revenue growth, 35-40% gross margin, and operating margin above 20%."
Premium GM +130bps, ~100bps of total GM expansion penciled in for 2027 unchanged. MS hung the call on pricing power — Sweden and Monaco hikes landing with no churn blowback.
Bull: Pricing power is real and now proven across two more markets, opex discipline is structural not cyclical, and the ad business inflecting removes the last bear crutch. GM path to 35-40% with premium subs still growing 9% means the margin algorithm doesn't need heroics.
Bear: 9% LTM revenue growth at ~33x earnings and 29x forward is a growth-multiple price for a decelerating top line — Piper's revenue miss and lower forward estimates are the tell. Ads inflecting at 9% ex-FX with soft CPMs isn't exactly a second engine.
Positioning matters here. This was a +11% move on an in-line quarter — that's the market paying for visibility on the margin ramp, not the numbers. R/r gets tougher at $583 with the bull cluster at $625-720 (~7-23% upside) and neutrals below the tape. Stock works if Q4 net adds print and ad momentum holds into 1H27. Not sure we can read too much into Sweden/Monaco as proof of US pricing headroom — but the burden of proof just flipped to the bears.
Verdict: fine quarter, decelerating story. APPF posted its 8th straight beat-and-raise — EPS $1.82 vs $1.79, rev $295.69M vs $294.36M — but nobody's celebrating. The beats are getting smaller, existing-customer unit expansion keeps moderating, and the stock slipped after hours anyway. When you beat eight quarters running and the bogeys keep rising, "modest beat" reads as a miss. That's the setup problem at 49x P/E.
Three Buys reaffirmed, PTs nudged — no thesis change, no fireworks:
Revenue +21% y/y. Solid on its face. Underneath it, the rate-of-change story is less friendly: adjusted ARR grew ~21% y/y vs 33% last quarter (tough comp, sure, but that's a big step-down), and unit adds beat while existing-customer expansion moderated. Management tightened FY26 revenue to $1.122-1.127B — only a $2.5M midpoint raise against a $1.2M Q3 beat, and the $1.1245B midpoint sits slightly below consensus. Implied Q4 below the Street. Op margin guide trimmed 25bps at midpoint to 27.0% as they pour into sales capacity and GTM spend.
Bull case: 21% growth, 63.5% gross margins, 26% ROA, +42.7% over six months, and the margin guide-down is self-inflicted investment, not erosion. DA Davidson:
"remains constructive on AppFolio's ability to continue winning new customers and monetize through value-added services."
At 6.0x NTM recurring revenue, the bulls argue you're not paying software bubble prices for a category leader with an 8-quarter execution streak.
Bear case: the growth is rolling over in real time — aARR 33% → 21%, VAS decelerating, and Guggenheim ties the unit-expansion moderation to a macro setup (rates, inflation, multifamily starts) that doesn't fix itself next quarter. At 49x earnings, a company guiding Q4 below consensus has zero margin for error.
This is a quality name trading like the decel doesn't matter. Both things can be true — execution is genuinely good and the numbers are getting worse. R/r skews against you near-term: every analyst kept the Buy but raised PTs by $7-10, which tells you conviction is fine but enthusiasm is capped. If you're long, you're underwriting a multifamily macro recovery. If you're not, wait for the decel to trough. Not sure one more modest beat changes either side's mind.
The print was a blowout and the stock still got cut. That's the WDC story this morning. September quarter came in at $4.38B REV and $4.64 EPS against bogeys of $4.13B / $4.07 — revenue +55.5% Y/Y, +17% Q/Q. Yet two of the day's notes are PT cuts, and the stock sits at ~$353, a long way from the $800 52-week high (still +182% over the past year, so some perspective). The debate has fully shifted from "is demand real" to "what multiple does an HDD maker deserve when it starts spending again."
TD Cowen's Sankar cut to $430 FROM $540 (kept Buy), and his note frames the real argument on the tape: the capex raise — into demand strength, no less — challenges the "rational duopoly" thesis that the whole HDD re-rating was built on. He assigns a 15x multiple, which is basically where the stock trades now (14.5x trailing). The bear math is simple: if the duopoly adds supply, margins get wrecked in the next slowdown. The bull counter is also in his note — in this upturn, added supply benefits both HDD makers AND hyperscalers, and each exabyte of HDD bought instead of NAND saves roughly $500M a pop. That's a massive economic moat for nearline.
Cantor took the opposite read on the same facts. Cut PT to $625 FROM $900 — but that's a multiple/market-selloff adjustment, not a thesis break. They RAISED estimates to $30 CY27 EPS and $41.50 CY28, meaning the stock trades at ~12x and ~9x those numbers. Their line on capex:
Cantor Fitzgerald said it views capital expenditure concerns as overblown. The new price target of $625 is based on approximately 15 times the firm's updated calendar year 2028 earnings per share estimate.
For context on the rest of the Street: Baird is the high man at $900 (ePMR/HAMR-driven EB shipment growth), Morgan Stanley at $668 (enterprise strength, pricing), Evercore at $575 (pricing strength). UBS is the skeptic — Neutral, cut to $435 on the same capex worry, though even they raised 2027/2028 estimates. So the cluster is $430-900 with cuts concentrated among the multiple-compression camp, not the fundamentals camp.
Bull: LTAs run into 2030-2031 with renewals pricing HIGHER as old agreements roll, estimates keep going up across the board (6 upward revisions flagged), and management is returning cash — dividend +17%, $1B buyback in the quarter. At 9x CY28 numbers with pricing tailwinds, r/r skews hard to the upside if you believe the cycle has legs.
Bear: Capex into a peak is how every storage cycle has ended historically. If the duopoly discipline breaks, the 15x multiple the bulls are underwriting compresses fast, and UBS at $435 tells you where the stock goes if the market decides to pay trough multiples on peak earnings. Not sure we can read too much into one raise, but the 56% drawdown from the highs says the market is already pricing some version of this.
Net: fundamentals say buy the dip, tape says the multiple question isn't settled. The resolution probably comes from the next capex guide, not the next quarter.
The AI cross-sell story is finally real, and the Street just chased it. FIVE PT raises off the Q3 print — Needham to $110 (from $85), FBN to $105 (from $60, a near-double), Rosenblatt to $105, MS to $91, Evercore to $90. Stock's at ~$89-90 after-hours, up ~176% YTD (FBN had it +188% pre-print — either way, this has been one of the stealth monsters of the year and nobody wanted to own it six months ago).
Mixed headline, strong guts. EPS $1.33 vs $1.28e — beat. Revenue $653M vs $667.3M — MISS. But the miss didn't matter because the quality metrics carried: subscription revenue +6.0% y/y vs 4.8% consensus, total ARR $2.8B +6.4%, and the number everyone's now anchoring on — AI ARR >$400M, ROUGHLY 2X Y/Y, ~15% OF TOTAL ARR. Company also raised FY26 revenue guidance by MORE than the Q3 beat amount, which is the tell PMs actually care about. (One caveat: "record 24.7% next-generation operating margin" is a company-defined metric — take the framing with a grain of salt, direction is what matters.)
The collective thesis has shifted. This isn't a churn-stabilization or enterprise-renewal-cycle story anymore — Needham's Reilly explicitly said AI attach and cross-sell drove the quarter, and that's a more durable driver than the legacy bogeys. His line is the one worth quoting:
"Investors underestimate the sustainability of RingCentral's business."
Needham sees accelerating revenue growth in 2027 as the renewal headwinds roll off, and admits its own numbers are conservative. $110 PT = 18x EV/FCF on FY26 estimates. FBN is underwriting further out — $105 = ~17.6x FY28 EPS of $5.96, with FY26 FCF/share of $7.15 (+23% y/y). 12 analysts have taken estimates up. When the bear-case multiple holder (FBN was at $60) nearly doubles its target, the repositioning is broad, not idiosyncratic.
Bull: AI ARR at 15% of the base and doubling means the mix shift is doing the work — 2027 revenue acceleration plus mid-20s FCF growth at ~13x forward earnings is still cheap for a re-rated growth story. The raise-more-than-the-beat guide says management isn't sandbagging a one-off.
Bear: Top-line missed, growth is still only ~6% — this is a re-rating on an AI narrative, not a fundamental acceleration yet. Higher Q4 share count kept EPS guide flat, and buybacks slowed SHARPLY, which is a yellow flag on capital return commitment after a 176% run. Not sure we can read too much into one quarter of AI attach rates at these levels.
After 176% YTD and a re-rating from the $60s, the easy money is made — but the PT cluster at $90-110 with the two most credible voices at $105+ says the Street thinks there's another leg tied to 2027 acceleration. R/r is balanced here, better on a pullback. Watch Q4 print for AI ARR trajectory and whether buybacks re-accelerate. If AI ARR keeps doubling, the "sustainability" debate ends and this becomes a momentum name, not a value re-rating.
BMO trims CLS to $400 from $450 but keeps the Outperform — this is a valuation reset, not a thesis break. The stock has been cut in half from the highs ($474 → ~$264, DOWN 31% OVER SIX MONTHS), and BMO's read is that the market is pricing in a networking slowdown that isn't in their numbers.
The bull case rests on the 1.6T networking ramp, TPU and Jalapeño exposure, and Street estimates they view as too conservative. The valuation math is the hook: 27.2x earnings against a 0.26 PEG — you don't see that combo often in AI hardware right now.
"Street estimates remain conservative given the 1.6T networking ramp, Celestica's exposure to TPU and Jalapeño, and the scale-up opportunity."
The tell here is timing. BMO explicitly flags Q3 results and the investor day as the catalysts where "the market will gain greater confidence on the durability of growth into FY2028." That's the whole debate in this name — not FY26/FY27 numbers, but whether hyperscaler custom silicon programs hold up into the out-years. If management gives any visibility past FY27 at the investor day, the multiple should re-rate off a 44% drawdown.
Worth watching into the print. R/r looks better here than it has in months, but you need that durability proof point before the crowd believes it.
CRWV -37% IN OCTOBER, sitting at $54.75 — a hair above the 52-week low of $53.50 and miles from the $141.79 high. UBS cut its PT to $105 from $120 but held the Buy, and the language matters more than the number: they're "re-examining the credit story" and admit restoring confidence "may take multiple quarters." That's a broker telling you the bogey isn't the next print, it's the balance sheet.
The CDS market is screaming the same thing — CRWV credit default swaps at NEW HIGHS, part of the broader AI-linked debt repricing. MSFT and AMZN rallied over the same stretch, which tells you PMs aren't abandoning AI infra, they're de-risking it. Rotation within the trade, not out of it.
PT dispersion is absurd: HSBC Reduce at $39, MS cut to $70 (from $99, financing uncertainty), UBS at $105, BofA Buy at $140, Cantor OW at $176. That $39-to-$176 spread is a market that can't underwrite the funding model.
Bull case: ~$129B in RPO and demand that isn't the problem — Cantor's whole thesis. Bear case: BofA itself flags another ~$20B IN FUNDING NEEDED to meet contracted demand, and with CDS where it is, the cost of that capital is the entire question. Q3 FY26 print is coming — the RPO number will be fine. Watch anything they say on financing cost and timing. R/r at $54 is a pure credit-confidence bet, not an earnings bet. Not sure we can read too much into any single PT here — the street is guessing on the debt market, same as everyone else.
META's biggest non-AI risk just got cheaper. Argus reiterated Buy/$800 after the August settlement with nearly all U.S. states on teen/child harm claims — Meta paid up and agreed to practice changes, but it avoids the tail risk of a liability finding and caps further state-level exposure. That's the right trade. Regulatory risk was the one bear case that didn't require you to be wrong on the ad business, and it just narrowed considerably.
"Argus views regulatory and litigation risks as Meta's most salient risk factor."
The fundamentals are doing the heavy lifting anyway — Q3 revenue +26% to $64.6B, Q4 guided to ~27% growth despite FX, 81% gross margins. Street is uniformly long with PTs ranging $600-1,050 and the recent movers (Rosenblatt $900, Mizuho $950) all leaning on the same AI-driven ad monetization thesis. One thing to file away: Rosenblatt flagged a change in Meta's capital-raise language in the Q3 filing, hinting at significant raises over the coming years. With capex where it is, watch how the market digests funding the AI buildout — that's the next narrative battleground, not litigation.
JPMorgan's Brian Essex raised his PT to $140 from $118 but kept the Underweight — a rare move that basically admits the refresh cycle is real while refusing to bless the multiple. Stock sits at $177, up 104% over the past year, so even the raised target implies ~20% downside. The core bear case: the Street is overestimating Product revenue growth for next year, and at 61x earnings with a 4.7 PEG, there's no room for a decel.
"JPMorgan raised its price target on Fortinet shares to $140 from $118 while maintaining an Underweight rating... The firm believes the Street is overestimating Product revenue growth for next year."
Meanwhile the bull bench keeps building — TD Cowen at $225 (Buy, projecting 18% 2H26 revenue growth vs. consensus), AmerX initiating at $240, Yorkville Ives at $225, all leaning on OT, sovereign SASE, and AI-driven security spend. So you've got a genuine bogey fight: Essex says the refresh-cycle strength is getting extrapolated too far into next year's product line; the bulls say AI/data center tailwinds make 18% look conservative.
Print is October 28, options pricing a 10% move. R/r here is tricky — the name has momentum and 80% gross margins, but the valuation means any product growth wobble gets punished hard. If you're long into the print, you're betting the refresh cycle has another leg. Essex is betting it doesn't.
Guggenheim went to $175 (from $125) into the print — and the interesting part is why: their field checks say the bogey is de-risked, not that demand inflected. Partners report little q/q change, AI traction still muted despite management's cheerleading, and the new pricing/packaging changes are getting mixed reviews from customers. So this is a "setup" call, not a "beat the world" call.
Q3 coincides with the federal fiscal year-end, and Guggenheim's DoD trackers look strong and still revising up — that's the tailwind that makes cRPO consensus "achievable" and a Q4 guide of at least in-line likely. The read-through: r/r into Wednesday (Oct 28) leans long, because the bar got reset while the one real growth engine (gov vertical) delivered.
Consensus is clustering here — Piper at $175 OW (Security biz now >$1B ACV), BMO made it a top pick, StoneX at $165. But note what nobody's saying: that AI monetization is showing up in partner checks. Early signs on AI Control Tower, nothing more. And at 9.4x EV/NTM recurring revenue after an 85% run in six months, the multiple already prices the beat-and-in-line scenario. Stock up ~85% in six months into a "stability" print is exactly the kind of setup where an in-line guide gets sold.
Net: fine to be long into the print on de-risked numbers, but this is not the quarter where the AI narrative finally gets proven out. Fade the hype, respect the fed data.
The print beat, the stock got destroyed, and Rosenblatt says the tape is lying. Q3 came in solid — EPS of €0.102 vs. €0.07 bogey, revenue €5.83B vs. €5.73B expected, +9% constant-currency sales, 11.7% op margin. Stock still fell 15% on the print. Rosenblatt trimmed the PT to $14 from $15 (peer multiple compression, not fundamentals — they actually RAISED estimates) and kept the Buy. At $7.88, that PT implies ~78% upside, which tells you how far sentiment has swung.
The firm's read on the selloff: this was an AI infrastructure risk-off move tied to neo-/hyperscaler CDS spread worries — NOT margin pressure from the new IP router rollouts. That's the key distinction. If you buy that framing, the drawdown is a positioning/flow event, not a thesis break.
Rosenblatt described Nokia's third-quarter 2026 results as solid and said the outlook is strengthening.
The one real fundamental wrinkle: 2027 op profit guidance unchanged, but FCF guide came DOWN on €400M of added working capital to support growth. Bulls will call that growth investment; bears will say a company cutting FCF guide while the market is already punishing AI-adjacent capex stories is asking for multiple compression either way. Not sure we can read too much into one day of flow-driven selling, but the r/r at $7.88 with a Buy-tilted consensus is at least interesting if the CDS spread noise settles. (Bonus: the ICEYE/Ukraine defense satellite LOI — small, but optionality in sovereign satcom is free at this price.)
SentinelOne finally has the Street moving in one direction — and it's not on the stock, which has already run, it's on the product. Post-OneCon, SIX firms weighed in: Needham to $31, Rosenblatt and Baird to $28, Susquehanna holding $28 Positive, Jefferies at $26 Buy. Only D.A. Davidson sat at the kiddie table with a $20 Neutral. The cluster around $26-31 implies the sell-side sees the AI/agentic pivot as real, not marketing. Stock's at $22.50, UP 62% over 6 months and 50% YTD — so a lot of this is in the price, but the PT herd is still chasing it higher.
The bull case is shipping product, not slideware: 5,000+ customers have activated Purple Agentic Investigation, Singularity ADR is in early access on top of a 400-plus Prompt Security installed base, and the new credits-based consumption model gives agentic AI an actual monetization rail. Add the Google Cloud Security pedigree at CTO/CPO and you get the "endpoint vendor becomes autonomous SOC platform" narrative with credible executors.
Rosenblatt said the company "moved its autonomous security vision into shipping product" — the line that matters. Vision-to-shipping is where most security names die.
The bear case is thin but real: Baird flagged pressure from the proposed convertible offering (dilution into a 62% run — classic), and Needham had to actively dismiss cannibalization concerns, which means the question is being asked. Fundamentals do some lifting here — 21% revenue growth, 73% gross margin, and profitability expected this year. Not a name to chase at $22.50 after that move, but the r/r on pullbacks toward the convertible overhang clearing looks decent. CrowdStrike multiple-envy trade with actual product behind it now.
Beat and raise, but the growth engine is decelerating — and that's the whole debate. SSNC posted 5.2% ORGANIC GROWTH, down from 7.6% in Q2, with Q4 guided to just 4%. The miss on momentum got papered over by RECORD ADJ EBITDA of $674.8M (+9% y/y) and margin expansion to 39.9%, plus a modest raise to the FY EPS guide ($7.14 vs $7.09). UBS held Buy/$101, Needham lifted its PT to $105 from $90 — nobody's flinching on the rating, but nobody's chasing either.
The segment detail is where it gets dicey. WIT slowed hard — 1.9% VS 9.8% PRIOR — and GIDS decelerated to 6.3% from 8.9%. Healthcare accelerated, and the alternatives franchise keeps compounding: AUA at $3.116 TRILLION, +18.6% Y/Y. Buybacks took out 5% of the share count, which is doing real work under that +15% EPS print.
This is a margin-and-buyback story masquerading as a growth story. Bulls point to the EBITDA record, the alts AUA compounding near 20%, and healthcare inflecting — the r/r works if you believe the organic slowdown is transactional lumpiness, not structural. Bears say 4% Q4 organic guidance on a stock that's been re-rated on growth is a bogey problem waiting to happen, and WIT falling off a cliff (9.8% → 1.9%) needs an explanation, not a shrug. Not sure one quarter of decel breaks the thesis, but the margin story is carrying more weight than it should have to. Watch whether Q4's 4% is a floor or a trend.
NBIS — Better funded than CRWV on customer prepay (>$9B guided this year vs $8.5B gross debt / $8.0B cash), but still raising. Call buyers paid $25.7M premium for Mar 230s and Apr 260s — someone wants upside convexity into the funding debate. Bloom says servers won't be the bottleneck on the 300MW NJ project (power block ~2 months post-approval). The debate isn't demand, it's whether prepayments mask back-end-loaded unit economics.
IREN — The commitment stack is the problem. $7.6B debt vs $5.9B cash plus $13.8B in hardware/site commitments, and the Microsoft contract only carries a 20% prepay. Convequity's regime call (neoclouds trade inverse to AI demand because faster growth widens the cash hole) applies hardest here. Funding-gap trade, not a demand trade.
MXL — Best print in the window, full stop. $218M revenue, +29% Q/Q, +72% Y/Y, AI datacenter infra +210%, Q4 guide $240-260M implies acceleration, 1.6T DSP ASPs roughly double, bookings into 2028. The hair is cash conversion — negative $13.6M operating cash flow, $45.7M sucked into prepaids, revolver up $70M to $200M. Earnings real, working capital heavy. Multiple caps if that persists.
MRVL — The MXL DSP-ASP-doubling datapoint is bullish read-through, not share loss. Ara 1.6T already in mass volume and qualified supply is tight. MXL gains are an offset, not a decline signal.
ANET — 1.6T roadmap de-risked: 7060XE7 air-cooled in Q4 2026, liquid-cooled plus more 800G in Q1 2027, Tomahawk 6 publicly identified in the platform. Broader qualified optics ecosystem = deployment confidence without needing one DSP winner.
AVGO — Tomahawk 6 sits in Arista's qualified 1.6T ecosystem — Ethernet scale keeps paying. The wrinkle: Broadcom is reportedly backing a $35B debt deal buying its own chips for Anthropic to lease. Vendor financing supports demand but that's circularity risk building in the plumbing.
CRDO — Two overhangs: a class action (N.D. Cal., lead plaintiff deadline Dec 22, 2026) and MXL's Annapurna merchant AEC silicon attacking the finished-cable pool in 2027-28. Bulls dismiss the MXL threat; it's structural, not cyclical. The microLED→microVCSEL pivot shows adaptability but also roadmap uncertainty.
LITE — Can only meet 30% of demand. Extreme constraint. Bullish for pricing, bearish for revenue if capacity can't scale. 200G EML positions it in the 1.6T ramp, but watch who captures the economics — module makers vs. contract manufacturers.
COHR / MTSI — Both product claims (400G EML for 3.2T; 448G/lane linear-drive chipset) unverifiable. Treat as noise. The real MTSI signal is negative: MXL's DSP-TIA co-design at 80% attach (→100% in ~18 months) shrinks the merchant TIA market.
SMTC — Same DSP-TIA bundling read-through. Loses sockets and pricing as DSP vendors bundle.
FN — Overlapping 800G/1.6T ramps help volume and utilization. Rising semi content per module means FN gets volume but maybe not proportional margin.
ASML — Patent filings show an unannounced hybrid bonding tool — not in consensus models, but HBM adoption pushes to 2029+, so this is long-dated optionality, not a 2027 number. The near-term story is China dollars: ASML + LRCX combined $5.1B China revenue in 3Q26, +82% Q/Q. Mix declines, dollars don't.
LRCX — Fifth straight record quarter, $8.22B, +22% Q/Q, +54% Y/Y, ~$120M above guide midpoint. CY26 guided to +30%+ Y/Y — 2x the analyst-day bogey. Quality caveat: upgrades/services driving it means customers sweating installed base. China was an all-time high in dollars.
NVDA — Rubin NVL72 hit 862 tok/s/user on GLM5.3, nearly 2x interactivity in eight days via kernels — the software moat compounds at the platform ramp. Google planning up to 960K Vera Rubin GPUs (via TSM/CoWoS-L, Quanta/Wiwynn assembling). Kazakhstan sovereign deal: 45K Grace Blackwells, ~500MW planned. DSX POD cuts deployment from 6 months to 14 days but NVDA stays rack-level, which handicaps it vs. Google/Huawei full-system. Cleanest bear case: enterprises can't afford the stack, route through hyperscalers, infer on ASICs. NVDA cuts price or eats margin.
TSM — Arms-dealer perfection. Google's 960K Rubin build runs through TSMC chips and CoWoS-L. Selling so much 3nm it's building four new fabs — Samsung and Intel aren't in the merchant 3nm game at scale. Customers prepaying for capacity (MXL confirmed). North Texas as a possible second US site (ERCOT speed, cheap land) — strategic but margin-dilutive.
INTC — $3B advance payment to Taiwanese substrate makers (AMD/NVDA each $1B) — direct signal of substrate tightness. BESI says Intel's hybrid bonding tools are delivered and installed, possible expansion round in 2027. The stitching-plus-High-NA bull case (derisked by 2027 vs TSMC 2033) comes from a loud self-interested source, but it's technically specific. Packaging helps; still needs foundry customer wins.
MU — DRAM tight, NAND lagging: 32GB DDR5 server +15% in a month, NAND wafer spot -7%. Jensen says no memory oversupply by 2029; NVDA scaled back Rubin Ultra HBM on HBM4E supply/yields — supply stress, not demand. Options showed capitulation-then-call-buying ($20M premium, 27K Apr 60 calls) — possibly a local bottom after forced liquidation in memory longs. 3D DRAM/vertical HBM is the multiyear pitch; it's capex-intensive but real.
SNDK / STX — WDC's +28% HDD ASP/TB confirms the pricing cycle for STX. For SNDK: Kioxia says ~50% of US DC demand unmet and is resisting LTAs — supplier discipline can tighten contracts even with spot -7%. Mixed-to-positive.
DELL / HPE — Component inflation math is brutal: full pass-through protects gross profit dollars, not margin percentage (18.5% in the worked example). Working capital rises too. HPE has less pricing power than Dell. Both are margin-percentage losers in this tape.
SLB — DC revenue $230M, +76% Y/Y, +23% Q/Q, but ramp costs pressured margins. Growth real, profitability not keeping pace — classic DC-supplier trap. Needs scale before margins inflect.
IBM — Q3 revenue $17.1B, +5% Y/Y. Legacy-software-with-AI stays low single digit. AI not inflecting consolidated growth yet.
INFY (read-through: WIT, CTSH, EPAM, CNXC) — The most important services datapoint in the window. AI revenue +29% sequentially but FY CC guide cut to 1-2% from 1.5-3%. AI revenue rose ~$124M Q/Q while consolidated rose only ~$63M — residual non-AI declined 1.3%. AI is cannibalizing the legacy base, not adding. Revenue per unit of work compressing. Negative across the complex; CNXC's call-center model is the most directly exposed, EPAM's engineering mix may hold pricing best.
GFS — Purdue research agreement (semis, AI hardware, silicon photonics, MPW access). Not revenue. Optionality on packaging/photonics, nothing more yet.
COHU / ENTG — Investor days Nov 10 and Nov 9 respectively. Catalysts, not prints. Both can reset guidance frameworks — watch for HBM/AI test exposure language.
ORCL — One number tells the story: capex at 161% of operating cash flow ($75.7B vs $46.9B LTM). Debt-plus-leases ~$457B, 9.7x OCF vs MSFT 2.5x. RPO $664B is nearly MSFT-sized on a quarter of MSFT's cash flow, and only 13% converts in 12 months. One desk called the bottom on 2x volume — capitulation can mark a trading low, doesn't fix the mismatch. ABBX capacity plan (4GW → 12.8GW by 2029, ~$52B RPO/GW) is bullish if financed and permitted. Big if.
AMZN — Jassy: 2026 obviously sold out, 2027 largely sold out. $496B AWS-only backlog — strongest operator demand datapoint in the window. Capex at 107% of OCF — strained but not ORCL-strained. The Loudoun County exit is the tell to watch: supply discipline (capex forecasts too high) or power access (delayed, not cancelled)?
MSFT — Copilot checks positive for the first time in a year: GPT-5.6 improving UX, deployment much faster than competitors, largest multinationals buying tens of thousands of incremental seats, Azure expected to beat guide. Balance sheet is the shield — 63% capex/OCF, 2.5x leverage. Monetization per seat still needs proof, but this is the first real Copilot green shoot.
GOOGL — 960K Vera Rubin GPUs while simultaneously threatening NVDA with TPU external sales pitched at $100-250B by 2028 (cost advantage, not speed — HBM and customer financing are the constraints). Locking power aggressively: 702MW fuel-cell PPA with AEP (~$4B, 20-year) and Black Hills through 2048 (590MW grid + 2.1GW managed). Google is paying up for the binding constraint.
BE — The tell of the tape: 36% EPS beat, 15% revenue beat, revenue +138% Y/Y, guide raised to $4.4-4.6B, product GM 38.4% — and the stock did NOTHING. Expectations and positioning are full. Beat-and-no-move setups get fragile; the next miss is a trapdoor. Cash conversion (negative FCF) remains the fundamental hair.
AEP / BKH — AEP's 702MW fuel-cell PPA with Google (~$4B, 20-year) and Black Hills' ~$150M expected 2030 net income from the Google-linked project (through 2048) are the utility monetization template. Regulated return plus fees. Execution spans years.
KMI / WMB — Onsite gas generation shifts the bottleneck to firm gas delivery. KMI FID'd the $4.2B Permian Link ($1.6B KMI share, take-or-pay backed, Q3 2030). Math: 1GW at 90% utilization/60% efficiency ≈ 0.12 BCF/d. Contracted laterals win; commodity gas doesn't.
AAPL — Approaching #1 market cap from NVDA. The market is treating Apple as defensive tech with AI optionality and no capex prison — its AI exposure is others' capex, and $100B+ FCF makes it the source of funds in the complex. Foldable Duo at $1,999 in 70+ countries with resale markups — launch theater noisy, cycle real, price point limits units. Relative strength unwinds if hyperscaler capex guides clear.
CART — Exploring a sale at a reported $12B valuation, bankers hired, talks with DoorDash and Uber (Uber intermittent). Stock +10% to $56.44 prices process, not a deal. FTC is the deal killer — Uber or DASH buying the grocery leader in this political environment is a heavy lift.
VZ / T / TMUS — Starlink re-rating telecom terminal value. SpaceX reportedly talking to tower owners, targeting 80K cell sites. VZ on day three of pressure. This isn't positioning — it's terminal-value compression. D2C pricing unproven, but every headline hits the multiple. TMUS has the most to lose as the premium growth name.
PLTR — The exception to the NVDA enterprise bind. US commercial +149% Y/Y, accelerating from 133%. Monetizing AI where enterprises can't build their own stacks. Elite.
CRWD / PANW / DDOG / ZS / OKTA — Whole complex got a bid on the PHP exploit and certificate-counterfeiting headlines. CRWD named in INFY's Project QuiltWorks (Falcon assessing/remediating frontier-AI risk). Key caveat from Parekh: budgets not growing "disproportionately" — this is share-shift to platform vendors, not wallet expansion. Agent security risk supports identity (OKTA) but same budget caveat applies.
GDDY / TCX / WIX — Verisign's Nov 1 wholesale hike ($10.26 → $10.97, +6.9%) vs ~1% retail pass-through = registrar spread compression. Negative GDDY/TCX unless they push ~3.5% retail. Offset: AI-assisted site creation is a demand expander — Airo at $50M annualized bookings run rate vs $10M a quarter ago. Value accrues to whoever owns the customer relationship (WIX bullish read-through).
TTMI — Pitched as the bombed-out non-YOLO DC proxy: 9x 2027 EPS guide, 40% DC exposure, >30% organic growth. Someone liquidated yesterday — possible local capitulation. Needs follow-through.
LGN — 7.5x 2026 EBITDA, half the business data centers. Cheap DC cooling exposure via recurring HVAC services. Lower growth than equipment, but that's the price.
PALX — First commercial PO from an unnamed strategic hyperscaler for GaN MicroLED chips in 0-50m optical interconnects. Validates an alternative approach. Real but tiny.
CBRS — 2-3 year window in premium low-latency inference: wafer-scale hits high utilization at batches of 3-4 where GPUs need much larger batches. Bear case: high-volume tokens stay on GPU economics.
HAWK — $11.50 vs $26 IPO, ~$5/share net cash, two cluster launches this week. Cash floors the downside; launches are the catalyst.
RHI — 20x for LSD growth, +40% in a week. No margin of safety. Pass.
CSGP — Former senior sales director says clients recreate CoStar data via title companies and third-party MLS. Moat-erosion signal — the moat is pricing power, not exclusivity. Data recreation is possible but costly.
AXP / BCS / EBAY / NFLX — AI-cited layoffs are now a headline wave: AXP -5% (~3,500), Barclays -5,000, eBay -9%, Netflix -6% (~1,000). Supports the enterprise-ROI narrative, overhangs the whole AI trade sentiment-wise. 2.8M Americans work in call centers (CNXC holders, note).
MS / JPM / BAC / C — All blocking/restricting third-party AI agents inside consumer accounts. Unapproved agents entering with customer logins, some acting independently. Dimon's liability point is decisive: ACH has no chargeback, wires are final. Banks will allow capped virtual cards, not account-level agents.
No signal in window: SAP, TXN, PAYC, FIX, PEGA, SYM, CCI, KN, DIOD, BHE.