Good morning. SOX -2.3% into the weekend despite NVDA GUIDING 70% CY27 GROWTH vs 47% CONSENSUS — momentum unwind, not demand signal. Volume, reliability, prepaid supply all intact. Rotation is the story: IGV +25% vs SOX -22% since June 22. TEAM +35%, CRM +23%, WDAY +18% on recent prints; software is the crowded long now.
OpenAI's Jalapeño on Samsung HBM4 is the first credible CUDA challenge on tokens-per-megawatt — a 2027–2028 problem, not 2026. Cerebras is the exposed one.
Power is the binding constraint. Turbine blades sold out to 2030, three global casting suppliers, Musk vertically integrating. Underwrite GW, not $-per-GW — 11–16GW likely 2026 additions vs 17–29GW theoretical absorption. Slippage, not destruction.
Asia: China compute complex printing — CXMT ¥77.6B H1 profit, multimode fiber doubled YTD, RTX Spark AI PCs sold out.
Three themes: NVDA's 70-vs-47 is the number; the fade is positioning. Software rotation works until it hits commodity multiples. Energization gates infra, not demand. And China substitution runs hot.
We'll hit up NVDA, META, and TEAM first, then get to power/optical — ANET, COHR, AAOI.
Good quarter. Better guide. The tape sold it off anyway. MRVL printed $2.739B REVENUE and $0.94 EPS — modest bogeys — raised Q3 guidance TO $3.15B/$1.10, and still fell 7.5% after hours to ~$223. PMs wanted a knockout; they got a solid decision on points. After a 196% six-month run and +28% in the month before the print, the bar sat at perfection. Benchmark called it bluntly.
"The guidance increase was insufficient given the stock's recent performance." — Benchmark
The quarter itself was spotless. Data center +46% YoY, +18% QoQ, with Q3 guidance implying +74% YoY. Management raised FY27 revenue TO $12B FROM $11.5B and FY28 TO $18B FROM $16.5B. Custom XPUs run >$2B this year, on pace to more than double next year. Maia contributes $700M next year; Trainium stays on schedule. Gross margin holds at 58% through CY27, and Oppenheimer models operating margin at 40% by end-CY27. Nobody is arguing with the fundamentals. The argument is price and timing.
This stock stopped being a quarterly earnings story. It's a Google warrant story with a P&L attached. Full warrant vesting implies $120B IN CUMULATIVE GOOGLE PURCHASES over roughly seven years — TEN TIMES the current run rate. Oppenheimer models a $25B/YEAR AVERAGE from CY28-33, with material upside starting in 2028. KeyBanc explicitly notes FY27/FY28 guidance ALREADY embeds Google. Read that carefully: the raise we just saw is the pre-Google base, not the Google tax. The warrant upside lands on top of a fiscal 2029 custom XPU base above $10B. (Assuming milestones vest, which is the whole ballgame.)
"Adding one year of a new Google agreement would support $20 in earnings power, excluding wins elsewhere." — Cantor Fitzgerald
Even the Neutral guy on the tape hands you the bull case. Cantor's $300 PT is 15x that $20 EPS stretch number. They admit risk/reward skews up near-term — they just won't pay 77x trailing earnings for a 2028 option. You can argue with the rating. You can't argue with the math.
Bull: Two compounding engines — optical interconnect and custom XPUs — both accelerating. Data center growth steps up from 46% to 74%. Google converts a product cycle into a seven-year annuity, and the Oct 6 analyst day should give the Street the milestone detail it wants. Street PTs span $246 (Morgan Stanley, Equalweight) to $400 (KeyBanc, Overweight), median ~$300, with the overwhelming majority at Buy. KeyBanc's $400 says the market still under-models the custom XPU ramp.
Bear: Paying for perfection in a stock up 213% in a year. The "beat" came in roughly $40M above Stifel's $2.70B revenue bogey — marginal. The raise added $500M to FY27 and $1.5B to FY28 — incremental, not transformative. And the transformative event, the Google warrant, doesn't hit the P&L in a material way until 2028. That's a long-dated call option, not next quarter's earnings. Valuation: 29x CY28 EPS with NVDA cheaper in the same sector. If Oct 6 underwhelms, there's no near-term fuel.
Call the after-hours fade exhaustion, not thesis breakage. NVDA set the bar one day earlier; MRVL couldn't clear it in the tape. But the custom silicon narrative remains the cleanest secular trade in semis — the hyperscaler custom XPU mania feeds ALAB, AVGO, and the whole ASIC complex, not just Marvell. MRVL is just the purest expression of it, with Google as anchor tenant. Long-dated bulls hold. Fresh money can wait for the Oct 6 dip. The next binary catalyst is the analyst day, not the December print.
THE EPS BLOWOUT IS A TAX ARTIFACT. THE OPERATING BEAT IS THE REAL STORY. AFRM finished FY26 with a clean F4Q: GMV +4pts, RLTC +8%, adjusted operating income +10pts vs estimates. The $4.62 GAAP EPS against $0.35 consensus carries a large tax benefit — strip it out. What matters is the guide: FY27 came in ABOVE Street, with the Walmart volume loss and Prime Day shift already embedded. That's operating leverage, not accounting.
Targets moved up across the board — the new cluster is $85-$124, up from roughly $80-$117. TD Cowen leads at $124 (Buy), BofA at $104, BMO at $101 (Outperform), Needham at $100 (Buy), RBC at $96 (Sector Perform), UBS at $90 (Neutral), Cantor at $88 (Overweight). The bull thesis is NOT the BNPL category — it's the Affirm card. LESS THAN 1 IN 5 AFFIRM CUSTOMERS HAS THE CARD. TD Cowen's $124 is an underwrite of card attach plus UK merchant adds plus Australia/ShopPay as the compounding engine. BofA thinks FY27 guidance is outright conservative.
BMO's Andrew Bauch had the cleanest framing:
"AFRM closed out FY26 on a high note with a strong F4Q beat, and provided an upbeat start to FY27 with guidance ahead of Street expectations. Results continue to reinforce the durability of its model, on top of pristine execution driving top-line growth and operating leverage consistency. While the bar was elevated, results and guidance give us added conviction in Affirm as the highest-quality growth story in our coverage to date."
Bull: Structural separation from the BNPL pack. RBC highlights stable delinquencies and improving funding costs despite the higher-rate tape — the capital markets funding model works at scale. Card penetration under 20% is a multi-year runway: more card users → more in-store GMV → better unit economics. Walmart revenue gets replaced by higher-margin card economics.
Bear: The easy money came in the $50s. At $84.32 (a $28.2B market cap) and +65% in six months, r/r is compressed. UBS sits Neutral, RBC sits Sector Perform — they see the quality, they just won't chase the multiple. The tax benefit flatters the profitability print, and the bar sits high for F1Q.
Best asset in the space, clean print, guide above Street. None of that is a secret anymore. The next leg comes from Q1 card-penetration metrics — card share of GMV, weekly active cardholders. If those print strong, TD Cowen's $124 looks sane. Until then, this is a buy-the-dip name, not a chase-it-at-$84 story.
Verdict: The margin-repair trade is ~80% done, and the stock knows it. WDAY trades at $201.56 — ABOVE DA Davidson's fresh $190 target, which is exactly why they stay Neutral. The quarter was clean, the guide is soft, and the AI story is real but still running on training wheels. The PT dispersion tells you everything: $160 (Stifel) to $238 (Bernstein). That's a $78 spread on a mega-cap software name. The model is in flux and nobody's sure which metric matters yet.
Seven PT moves post-print: six up, one down. One downgrade. Not a single analyst is arguing the quarter was bad — the fight is about what the deceleration means.
Don't chase it; own it on a dip. The 45% six-month rally was the margin-repair repricing. The next leg needs AI ARR to convert into net-new ACV, and we won't see proof until FY28 guidance firms up. That makes $190 the line in the sand (DA Davidson's target, also roughly the pre-rally breakout level) and $215-238 the ceiling (KeyBanc to Bernstein). Cantor's $205 cut is the honest middle: this is a show-me stock now. Range-bound is the base case until the AI monetization model stops needing subsidies.
Sell the story, buy the sell-off. Rubrik beat net new subscription ARR by $20M (printed $96M vs $76M consensus) and STILL traded down. That's an expectations problem, not a fundamentals problem — the Street just wanted more after a 106% six-month run.
PT cluster says it all: $114-120 across the board. Piper to $114 (from $91), KeyBanc to $120 (from $112), BMO to $115 (from $98). Cantor, Scotiabank, Wolfe all sitting at $120. That's eight firms, one message: the quarter was legit, the guide is sandbagged, and the dip is the setup.
Piper frames the guide as deliberately conservative, especially cloud in 2H:
"Rubrik is not seeing any material impact to annual recurring revenue from rising hardware costs or supply constraints... The company's 14% net new annual recurring revenue guide for the year appears conservative."
KeyBanc is even more direct about the trade:
"The analyst firm described the after-hours sell-off as a buying opportunity for a category leader with growth above 30% and accelerating, along with approximately 20% free cash flow margin."
Bull: Cyber resilience is a secular budget line. Elevated AI threat landscape makes recovery capability table stakes, not optional. 33% ARR growth accelerating, 81% gross margins, 20% FCF margin, Mythos driving new-customer adoption — that's a scarce combination at $22B. Net new ARR acceleration (35% vs 16%) says the go-to-market is compounding, not fading.
Bear: You're paying for perfection after 106% in six months. The post-print fade despite a $20M beat tells you expectations have outrun even good execution. BMO notes RAC is early innings and the expanding portfolio needs time to gain traction. Still unprofitable. The multiple already embeds the bull thesis.
My take: 14% net new ARR guide vs 35% actual acceleration is a massive gap. Either 2H decelerates hard or management is sandbagging — I'd bet on the latter. Beat-and-raise cadence, supply chain fear disproven, pullback after a run. This is the entry, not the exit.
Beat-and-raise, guide hike bigger than the beat, cloud +20%, sales-led sub +17% — and the street STILL can't fully commit. That's the setup in one breath. The stock popped ~24% post-print, now $99.84, $10.7B cap, +61% in six months. The PT cluster tells the real story: one convicted bull at $130 (Piper, up from $85), a quiet middle at $120 (Canaccord), and a $100-108 wall of "prove it again" from everyone else. The market owns the good news. The debate is whether 2H acceleration is real or hope dressed up in a beat.
F1Q (fiscal 2027): revenue $478.1M (+15% cc), sales-led subscription $398.5M (+17%), cloud $235M (+20%). cRPO held at 20% — good, but steady, not inflecting. GM 76%, which is the quiet flex: that's pricing power in a market where search and security fight for every seat.
Management raised full-year guidance ABOVE the size of the beat. That's the single strongest signal in the print — you don't do that unless the pipeline is real.
Piper Sandler (PT $130 from $85, Overweight) makes the bull case:
"F'1Q results were strong, showing a slight acceleration across key top-line metrics (revenue, sales-led subscription, cloud) and consistent 20% cRPO growth. Management highlighted continued success in security and AI search fueling strong commitment trends, while strong consumption and forward pipeline build underpinned confidence in further 2H acceleration."
Piper's conviction: valuation isn't reflecting what the company can demonstrate through the year. 53% PT hike says he means it.
The skeptics cluster tight. DA Davidson (Neutral, $100 from $80) wants "more consistent execution" — hedge-speak for "one quarter doesn't make a trend." Cantor (Neutral, $100 from $91) actually moved its multiple up to 4.4x CY27E EV/Rev from 4.0x, flags Splunk displacement potential in SIEM, and still needs "a clearer path to sustained acceleration." BofA at $108, Stifel at $107, UBS at $105. Needham: Hold.
Bull: Security demand is structurally elevated (threat landscape, not cyclical), AI search is a legit second vector, Splunk displacement is real SIEM share-shift, and 76% GM means pricing isn't the problem. The guide raise above the beat is management telling you the forward pipeline converts. At $99.84, Piper's $130 target is ~30% upside on a name compounding cRPO at 20%. R/r skews favorable.
Bear: You've already banked 61% in six months. The bull's own language flags only a "SLIGHT acceleration" — and cRPO is steady at 20%, not re-rating. GM ticked down on cloud mix and AI inference costs — the AI economics giveth and taketh. The $100-108 crowd says the stock already prices this quarter; they need a second consecutive beat-and-raise before paying up. Fair.
This is a first-derivative trade now. The print settled "is it working?" — it is. The next leg depends on cRPO re-accelerating past 20% and cloud consumption staying strong into 2H. Watch for Splunk displacement mentions on the next call — that's the share-shift evidence Cantor's sniffing around. Bottom line: the street's skepticism after a 24% pop is the bull's fuel, but it's also the honest check on a stock that's not cheap at 4.4x forward revenue. If you believe the guide raise over muscle memory, Piper's your guy. If you need to see it twice, $100-108 says you're not early — you're on time.
DA Davidson sticks with Buy and $325 PT (22x FY28 FCF), and the stock's at $261 — sitting ~24% below InvestingPro's $342.68 fair value. The near-term EPS trim spooks nobody. That's just net financing costs from the MaintainX deal closing August 3rd, not operational bleed.
The real signal in the updated guide: ~1pp HIGHER organic growth, plus another ~1pp inorganic from MaintainX. Deal extends growth capabilities into FY28. Q2 already proved the engine works — $3.30 vs $3.12 est, revenue $2.05B vs $2.01B. Full-year billings and revenue raised.
The Street's converging on the same thesis. UBS calls it "durable low-teens organic growth" and jumps to $325. Stifel and Guggenheim hold Buy at $285/$283. Nobody's anchoring on the acquisition noise — they're underwriting the core.
Perfect Piotroski (9) and 92.5% gross margins give the fundamentals crowd cover. The r/r here is straightforward: buy a low-teens organic compounder at a discount while the market fixates on deal mechanics. Momentum's the story and it's not slowing.
Benchmark keeps the Buy and $9 PT (100%+ upside from $4.29) after ROC's federal ABIS contract win. Don't get distracted by the headline number — $5.1M over five years is a row of desks at the Pentagon. The REAL signal is the transition: ROC moved from selling biometric algorithms to owning the full platform in a mission-critical federal deployment. First tangible proof point. That's the narrative that gets you to $9.
"The first tangible proof point that ROC can successfully transition from a biometric algorithm provider into a full-platform Automated Biometric Identification System (ABIS) vendor competing for mission-critical federal deployments at scale."
But steelman the bear too. $5.1M is tiny. Q2 revenue of $5.1M was only +2% YoY (though nearly 2x Q1) and the company swung to a $0.8M net loss vs +$0.6M a year ago. MANAGEMENT BLAMES THE COMPLETION OF A LARGE ROC WATCH DEPLOYMENT — i.e., the growth engine just stepped off a cliff and we're waiting for the next one. Stock down 36% in six months says the market isn't paying for the story yet.
One thing I like: executives voluntarily extended lock-ups to February 2027 covering ~54% of shares outstanding. Insiders locking up through a platform transition tells you they think the re-rating comes later, not sooner. The reference-account angle is the real optionality — FBI, DHS, DoD, State. That's a pipeline, not a contract. Buy-and-hold with patience, not a momentum name.
Verdict: Great quarter, wrong price. S beat and raised, delivered $56M net new ARR vs street — FIVE STRAIGHT QUARTERS OF NET NEW ARR BEATS — and the one Neutral holdout still says the curve hasn't bottomed. That's the whole debate in one sentence.
DA Davidson (Neutral, $20 PT vs $21.10 close) isn't disputing the execution. Revenue $292M (+21% YoY) topped consensus, EBIT came in strong, and the bulls all piled on — five firms pushed PTs to $24-26 (Scotiabank and Cantor at $26, Canaccord and Citizens at $25, UBS at $24). The bear case is purely about slope: net new ARR keeps beating but at lower absolute levels, and Kessinger thinks the deceleration lacks a hard floor. At $20, that's ~30x EV/FY28 FCF vs mid-teens growth SaaS comps at ~20x — good company, wrong price in his book.
"Shares remain range bound until SentinelOne establishes a hard floor on annual recurring revenue growth."
Bull steelman: five straight beats isn't luck, and the margin trajectory means this isn't the old land-grab S. Bear steelman: the stock is +73% IN SIX MONTHS, +51% YTD. The re-rating already happened. Paying 30x FY28 FCF for decelerating ARR leaves zero room for error — and with shares already above the $20 target, DA Davidson is saying you get paid nothing for that patience.
Benchmark keeps the Buy and $29 PT post-Q2 — roughly 75% upside from a $16.61 tape pinned near 52-week lows. Stock still slipped premarket after the print. The analyst is holding the line; the market is not buying the story yet.
Ads are the engine. +28% YoY in a soft macro environment is a number that makes PMs lean in.
"Advertising grew 28% year-over-year despite macro pressure, supported by AI-driven efficiency, expanding monetization scenarios and a maturing user base."
Gaming remains the drag against a difficult comp base. Management expects YoY growth to resume in Q4, with new titles supporting momentum into 2027. AI investment creates near-term R&D pressure — Benchmark frames it as targeted spend already showing up in content creation, recommendation, and monetization.
The quarter itself: revenue RMB 7.9B (+8% YoY), basically on the RMB 7.92B bogey. Net profit RMB 339M, +55%. Margin trajectory is real on the bottom line. But the tape wants to see gaming turn before granting a re-rating. That Q4 guide is the catalyst to watch. Until then, this is a "right story, wrong tape" setup. $29 PT argues either Benchmark is early or the market is too cold on ad durability. No rating changes this book; this is a hold-for-the-inflection name.
All feed reads, no ARW-grade coverage. Use for position sizing and theme mapping, not primary conviction.