Good morning.
Chip selloff day three, but the tape's finding footing — Asia didn't cascade, Nasdaq futures back in the green. 30Y AT 5.3%+ (19-year high) stays the bogey; everything AI trades as a duration function until it doesn't.
Overnight's biggest signal is corporate behavior, not earnings. SK hynix (HXSCL) dropped a KRW 40T (~$28.6B) buyback/cancellation — Korean history — plus a 50%+ FCF shareholder-return pledge. Management's buying the dip while public markets de-gross. That's the tell. Samsung Foundry raising advanced-node prices 10–15% pushes the same way; if TSM follows, leading-edge reprices.
Model-layer headlines flipped again — now OpenAI trails Anthropic, Altman pauses frontier training. Read it as compute reallocation, not capex destruction; Anthropic's $11.6B quarter needs hardware. NBIS: Pennsylvania GRID order is a moderate negative on Highridge timing, but the first 260MW looks defensible. CEG picks up clean-firm scarcity either way. China: H200s landing at ByteDance/Tencent (~10k each), Unitree +542% day one, Kuaishou's worst print in five years on AI costs — same de-grossing, different time zones.
Fed minutes + Warsh tonight. Hawkish on term premium = 30Y retests 5.35%. Neutral = this selloff was a clearing event.
We'll hit up HXSCL, NVDA, and NBIS first, then get to semis and AI infra.
THE SETUP IS GOOD, BUT THIS IS A CROWDED LONG INTO THE AUG 26 PRINT. Five firms mark PTs to $230-250 post-split and the street is aligned, not dispersed. That unanimity IS the risk.
Cantor's headline is the $725 to $250 cut — but that's the 4-for-1 split. Split-adjusted, it's a RAISE from $181. Ruykhaver stays Overweight with strong checks, platform consolidation, and core EDR acceleration. Benchmark matches at $250 saying CRWD beats across ARR, revenue, op income, and FCF. TD Cowen at $235, Wells Fargo and Stifel at $230, Citizens at $230. The PT cluster isn't the story — the agreement is.
Channel checks are the real signal. Stifel's 25-VAR survey: 44% of resellers beat their quarter — THE HIGHEST PRINT IN NINE QUARTERS. 84% expect 10-40% growth in FY27, up from 76%. Cantor's 26-partner survey shows 58% ahead of plan. Two independent surveys, same trajectory. Reseller bullishness runs 12-to-1.
AI is the next leg. Stifel sees AI Security and Next-Gen SIEM as the biggest adoption drivers over three years, with Mythos already hitting the pipeline. Benchmark flags AIDR adoption and Project QuiltWorks monetization as the watch-items. The Cerebras partnership gives CRWD an AI inference story it can actually sell, not just defend.
"Investors will increasingly look for evidence that AI Security can complement and potentially accelerate the company's core growth algorithm." — Benchmark
Flex consumption clears the bar too — 89% at least in line per Stifel. That matters because Flex was the overhang last year. Falcon now runs 33 modules vs 30 a year ago. Consolidation math keeps working.
Key level: $227.50 — the 52-week high. Break that on a beat and this runs. But the market already prices in a good quarter at $212. You need a raise to get paid.
THE WHOLE SETUP IS ONE NUMBER: CONSUMPTION. SNOW IS UP ~85% IN SIX MONTHS AND THE STREET HAS COLLECTIVELY DECIDED Q2 IS A BEAT-AND-RAISE. Stifel frames it best — investors expect revenue growth acceleration ABOVE 34%, implying a beat of >3%. Morgan Stanley's checks support ~34% product revenue growth and an FY27 raise to 32-33%. The print is days away and everyone is long the same trade.
The target dispersion tells you everything about the debate. UBS is the high at $425 on AI demand. Citizens at $408, Evercore $360, Stifel $350, Jefferies $310. Morgan Stanley holds the line at $300 — roughly 67x CY27 FCF — and explicitly flags that "the significant rally over the past three months has created a high bar heading into results." That's the bull/bear in one sentence: momentum vs multiple.
Bull: Consumption is inflecting on multiple cylinders. Core data warehouse demand is healthy, CoCo (Cortex) usage spiked quarter-over-quarter and WAS NOT in Q1 guidance — it's now in Q2 guidance, but with conservative assumptions baked in given limited operating history. Citizens' checks flipped to 88% POSITIVE from 53% last quarter. The AI narrative isn't hypothetical here — it's agents generating queries, queries generating data, and data generating consumption. That's the flywheel.
Bear: The stock has run 80% since the Q1 print. AI adoption is "mixed" — the core warehouse is doing the heavy lifting, not the new stuff. MS's $300 PT is the tell: if the quarter merely matches the bar, the multiple does the work on the downside. A beat that doesn't accelerate growth is a sell-the-news event at this valuation.
"As long as Snowflake serves as the primary data repository, agents will create more queries. As people do more queries there is more data, which is how the company charges." — Partner checks via Citizens' Patrick Walravens
That's the cleanest articulation of the agent-driven consumption loop we've seen all season. It also explains why Citizens is at $408 — they're underwriting a structural shift, not a quarter.
MS says the demand environment for infrastructure is "the strongest backdrop since 2022" — SEVEN OF EIGHT covered infra companies accelerated core KPIs in the June cohort. SNOW is the highest-beta liquid way to play that tape. 31 analysts have revised FY estimates upward. The positioning is crowded, the narrative is hot, and the setup is genuinely good. Just remember what happened the last time everyone agreed SNOW was about to inflect. (It didn't.)
The floor: another upside quarter is likely. The ceiling: this stock is priced like the AI data layer winner — and if CoCo doesn't show follow-through, the floor gets lower in a hurry.
DEAL ARB. THAT'S THE NAME. Francisco Partners takes Weave private at $7.40/SHARE CASH — ~$650M aggregate, 34% PREMIUM, board unanimous, Q4 2026 CLOSE. Stock trades $7.28-29, a ~1.5% carry to close. Stifel and Loop both cut to Hold, PTs pinned to $7.40. And then B.Riley starts coverage with a BUY and an $8.25 PT — ABOVE the takeout. That's the only genuinely interesting thing in this name today.
Stifel went Buy→Hold, $8.00→$7.40. Loop did the same. Both land on the same point: no competing bid coming. Stifel said it best:
"The business was well shopped and does not anticipate a competing bid to Francisco Partners’ offer."
That's the whole ballgame. At 1.8-2.1x forward EV/S (pick your analyst), FP didn't overpay. But the strategic universe for a 40k-location dental/vet/optometry comms+payments platform is thin. Who else shows up? Nobody obvious. Loop's framing is right too — this trades on arb dynamics now, not fundamentals.
Initiated Buy, PT $8.25 — deliberately above the deal. They use 2.0x 2027 EV/S on their $310M revenue estimate vs. a 2.8x peer median. Current stock: 1.2x that same number. The gap between deal price and B.Riley's fair value is the whole bull case in one line.
But their own model tells you why FP won this. 2027 EBITDA margin: 8.6%. Growth: 13.2%. TTM NRR: 92% — as in, the existing book SHRINKS without new logos. Monthly retention ticked up two quarters straight and embedded payments grew 2x total revenue, so the trajectory is real. 72% gross margin, too. But this is still sub-100 NRR, sub-10% EBITDA, post-guidance-cut, mid-sales-force-transition. That's exactly what a take-private at 34% premium looks like.
Bull: B.Riley's $8.25 says standalone intrinsic value sits ABOVE the offer. NRR inflection + payments attach + 72% gross margin = 2027 estimates beatable. If FP financing hiccups or closing slips, the floor isn't $5 — it's $7-plus on fundamentals.
Bear: 92% NRR. 8.6% EBITDA in 2027. Growth decelerating into the low teens. B.Riley's "discount to peers" assumes a re-rating to 2x sales for a business that loses customers unless it sells new ones. Deal closes, you get $7.40. A ~1.5% spread pays you to wait three months.
B.Riley's $8.25 is effectively a call on a topping bid or a broken deal with a fundamental floor. The spread says the market expects neither. Longs are long Francisco Partners' ability to fund and close — that's a merger-arb desk's game, not a TMT PM's. The standalone story resets at $8.25 if the deal dies; until then, the ticker trades at $7.40 minus time to close. Take the carry or find another fight.
Verdict: APPROVAL TRADE DONE. The stock ripped 39% in a week, cleared every price target on the street, and now sits at $269.80 — ABOVE the highest fresh PT on the tape ($260 Cantor). DA Davidson just hoisted their target to $250 from $175 and STILL says Neutral. That's the whole story in two numbers.
The Vineland data center approval is the catalyst. DA Davidson had this flagged as the key risk — construction freeze, regulatory limbo, the whole "is Nebius a poster child for data center delays?" narrative. That's dead now. Approval means they resume construction, and the analyst made the point that this removes NBIS from the danger zone as the industry sweats through PJM interconnection and power availability issues.
PA Governor Shapiro's executive order is noise, not signal, for the company-specific story. Cantor read it as "not a ban, codifies the process" for new builds — a procedural step, not a political gut punch. 20 proposed data centers in the state, 14 with active exemptions. Ratepayers eat 46% of capacity charges, which is why the order exists. But Nebius's Highridge site is what matters, and Cantor keeps Overweight/$260.
"The company no longer faces the risk of becoming a focal point for such delays."
That's the de-risking event. Full stop.
Q2 was a monster — revenue $582.3M, +454% YoY, ~2% above consensus. Annualized run-rate $3.0B, up from $1.9B sequentially. Adjusted EPS -$0.68 vs. -$51.46 expected (that prior number is absurd, but the improvement is the point). Spot pricing for AI compute is doing a lot of the heavy lifting, and DA Davidson's caution is partly "how much of this is spot vs. contracted?" Good question. The right question. Compass Point is at Buy/$300 — the bull case is that run-rate trajectories this steep get re-rated on scarcity, not on next quarter's multiple.
Bull: The market is pricing NBIS as the pure-play AI cloud scarcity asset. $3B run-rate, +454% growth, approvals cleared, and Tier 2/3 hyperscaler demand still outstrips supply. $300 is the near-term bogey.
Bear: You are paying 100%+ premium to CRWV (per DA Davidson's comp) for a company that JUST reported negative EPS, has power/regulatory overhangs in PA, and just ran 39% in a week. Neutral at $250 with the tape at $270 is the analyst saying the easy money got made. If spot pricing rolls over, the re-rating unwinds fast.
Rating check: Compass Point $300 Buy / Cantor $260 Overweight / DA Davidson $250 Neutral. Stock at $269.80. The street is NOT ahead of this move. That's the tell — you need the next leg of contracted revenue visibility to justify the leap. It isn't here yet.
Verdict: The optical trade is doing the work, and the bear case just lost its best data point. TD Cowen's bump to $225 (from $200) is mostly a chase — the tape is already there — but the buried admission matters more: Marvell's optical DSP share is NOT eroding at the pace previously feared. That was the #1 short thesis. It's cracking.
Stifel is the loud bull: Buy, $350 PT, expects July quarter ABOVE the $2.70B bogey and October guide ABOVE $3.00B — back-to-back ~11% sequential growth quarters. The bigger deal: management just raised INTERCONNECT GROWTH TO 70% YOY FOR FY27, up from a prior 50% guide. Data Center overall now 50% this year and 55% in FY28. UBS sits nearby at $340, playing the CXL angle. The bull camp is large, and it's getting larger.
"Management now expects Interconnects to grow 70% year-over-year in fiscal 2027, up from a prior 50% forecast." — Stifel
That's the rate-of-change signal you want at 80x trailing earnings: guidance accelerating, not decelerating. Optical strength showed up across the entire value chain this earnings season — ZR/ZR+ coherent pluggables are the common thread. This isn't a one-name story.
Bull: Data Center is already 76% of revenue, and it's growing faster than anyone modeled 90 days ago. Optical + lead XPU production + XPU-Attach expansion = three growth legs, not one. Amazon printing $25B annualized AI/chips revenue just reinforces the capex supercycle narrative. At 60x CY27 earnings (Stifel's frame), you're paying a premium for 50%+ DC growth — that's a growth-adjusted price, not a delusional one.
Bear: The spread between Stifel's $350 and Goldman's $195 Neutral IS the debate. Goldman sees the datacenter story; they just won't pay the multiple. Erste downgraded to Hold on valuation. Custom ASIC uncertainty persists too — if the lead XPU slips even a quarter, the derating will be fast and violent. At $234, this is a $205B market cap. The base gets big in a hurry.
The 206% one-year return means the easy money is made. But the guidance trajectory keeps stepping up, and the October guide is the next catalyst. If Stifel's $3.00B+ read is right, the bears lose their near-term argument entirely. I lean bull until the rate of change rolls over.
THE STREET JUST SWARMED THE LONG SIDE INTO THE SEPT 1 PRINT. FOUR FIRMS RAISED TARGETS THIS WEEK — Stifel to $415 (from $330), Cantor to $425 (from $340), Citizens to $415 (from $320), TD Cowen to $400 (from $360). All Buy/Overweight equivalents. The cluster speaks for itself: nobody on the sell-side wants to be caught flat-footed if FQ4 prints like the last nine did.
Channel checks matter more than the PT moves here. Stifel talked to five Tier-1 VARs/SIs representing $2B+ in combined PANW spend — four beat, one in-line. The largest US partner "performed well above expectations" and closed several mid-eight-figure deals. Cantor's independent read: 63% of partners now selling ahead of plan, up from 57% last quarter, with below-plan responses easing from 18% to 15%. That's sequential momentum in the channel, not just a one-off beat.
Buy-side is modeling the September 1 print at 35% revenue growth, mid-to-high 20s product revenue, 63.5% NGS ARR growth, and 36% RPO growth. Cantor pegs organic net-new NGS ARR at ~$573M. Note the word "organic" — the Street is forcing the platformization story to carry water without acquisition accounting for CyberArk and Chronosphere yet.
The CyberArk narrative is interesting: partners say customers are "buying into the vision" but describe the integration as early. That's bullish for the multiple, not necessarily for the cash flows. PMs should watch the FQ4 call for any commentary on the fiscal 2027 AI security budget inflection — that's the catalyst that takes this from "great company, expensive stock" to "institutionally must-own."
Bull: This is a platform share-shift story — SASE, XSIAM, firewalls, plus AI security exposure across network, endpoint, cloud, browser, and identity. Goldman's view that security budgets inflect higher on AI spend starting late 2026/early 2027 gives PANW a durable TAM expansion narrative, and the channel data supports acceleration, not deceleration.
Bear: You're paying 59.8x forward EV/FCF against a peer average of 21.4x. That's not a premium, that's a different asset class. InvestingPro has the stock on its "Most Overvalued" list. Nine straight beats are priced in. Any whiff of NGS ARR deceleration or CyberArk digestion issues on Sept 1 gets this thing cut in half — there is zero room for a "good, not great" quarter.
The 52-week high is $398.88. Stock closed at $374.14 — that's ~6% shy. If the Sept 1 print lands anywhere near buy-side expectations and FY27 guidance implies an AI security inflection, the path to $425 is open. But the multiple compression risk into that print is real.
"We view Palo Alto as one of the better-positioned names to capitalize on AI tailwinds because of its broad visibility across network, endpoint, cloud, browser, and identity, where AI traffic and agent activity must be inspected and secured in real-time." > — Cantor Fitzgerald
That's the core long thesis in one sentence. The question isn't whether PANW is the best platform player — it clearly is. The question is whether you can hold a 3x peer multiple into a print where expectations are already sky-high. Today's answer from the Street: yes. September 2 will tell you if they were right.
Earnings Aug 26 is the whole show. Stifel and TD Cowen both reiterate Buy, both slap fresh PTs on the tape ($282 and $275 — stock sits at $219.74, $5.32T mcap), and nobody on the sell-side is stepping out of line on the AI story. The consensus base case: beat AND raise. The interesting part isn't the ratings — it's the supply chain dirty data that backs them up.
This is what actually matters going into the print:
Stifel's sharpest take cuts through the bear case. Memory cost inflation and inference competition are the two big shorts du jour — but don't look for them in demand. Watch the margin line instead. NVDA still prints 74.15% gross margin. If the bears are right, that's where the crack shows first.
"Debates around memory cost and inference competition are likely to be expressed in gross margin rather than demand."
That's the honest battleground. Revenue is beaten to death — UBS already calls for a $3-4B beat, potentially hitting $95B. The debate is whether pricing power holds.
TD Cowen flagged that shares are driven by non-fundamental factors right now. Fair — this stock trades on narrative heat as much as the P&L. And the narrative keeps expanding:
BULL: This is a multi-year capex supercycle with visible demand into 2028. Foxconn and Supermicro prints confirm rack demand is accelerating, not peaking. Open-weight model proliferation creates a new inference demand layer, not just frontier-lab training.
BEAR: At 34.55x P/E and $5.3T, perfection is priced in. Memory costs hit margins, ASIC competition erodes pricing power, and any guide below the $95B street bogey gets ugly fast at this multiple. The stock trades on headlines — that cuts both ways.
Bottom line: beat-and-raise is the bogey, so the bar is high. Watch gross margin like a hawk. The demand side is the easy call. The margin side is where this trade lives or dies.
Don't chase the September print. Stock's up 31.5% QTD into a quarter where partner checks deteriorated — and the Street's high-water PT is $225. This is a great company at the right price and the wrong entry.
The PT cluster tells you everything: Stifel raised to $200 from $175, TD Cowen at $200, Truist at $200, Cantor the lone bull at $225. All Buy/Overweight. Nobody's stretching. That's a Street that likes the story at the right price, not a street pounding the table.
The bogey: Stifel expects a 2% revenue beat and 0.5% ARR beat when Q4 FY26 prints Sept 3. They also expect FY27 guidance reaffirmed — the fiscal year just started, so the bar is "don't screw it up" not "raise the roof." 41 analysts revised estimates UP. Company turns profitable this year. GPM 77%. The fundamentals are real.
The checks are the interesting part — and they're not uniformly good.
Stifel calls the quarter "uneven": mixed Tier-1 VAR checks, positive federal. Cantor's 21-partner survey shows real deterioration: 43% in line vs 53% in Q3, and 29% BELOW PLAN vs just 5% the prior quarter. That's a 6x jump in missers. Only 29% exceeded plan vs 42% before. That's a step-down however you slice it.
Cantor's not sweating it. They see Red Canary synergies, Z-Flex driving organic net new ARR, and three growth pillars — Zero Trust Everywhere, Data Security Everywhere, AI Security. Their strategic thesis is the strongest line in either note:
"SASE evolving into a control layer for enterprise AI governance given its in-line visibility into user and application traffic, alongside its ability to limit exposure to AI-driven threats through Zero Trust segmentation and policy enforcement."
That's the long-term story. AI traffic is the next big security wedge and ZS sits directly in its path. (The Hugging Face incident backdrop only reinforces it — cyber threats are evolving and AI is the new attack surface.)
Bull: Platform story extends well beyond legacy ZIA/ZPA. Red Canary closes the MDR gap. Z-Flex drives efficient net new ARR. October analyst day gives management a platform to lay out the multi-year AI Security framework — that's the real catalyst, not this print. Options flow confirms the setup: 39,437 contracts with calls dominating. Everybody's long.
Bear: The bar moved. 29% of partners below plan is a step-down, full stop. Sales leadership changes mid-quarter never feel great. If FY27 guidance comes in as "reaffirmed" rather than raised, the stock already banked the good news. Stifel's own words: "whether the upcoming results will be sufficient to drive shares higher." That's the whole ballgame. The stock doesn't need to miss to go down — it just needs to be merely good.
1. Q4 revenue/ARR vs the 2%/0.5% beat consensus — in-line with the beat is base case; anything less and the multiple compresses 2. FY27 guide — reaffirm is priced; guide-up sends it, guide-down is ugly 3. Sales leadership changes — why now, what's the narrative? Management will get asked 4. Analyst day teases — they'll preview October; listen for Zero Trust Everywhere / Data Security / AI Security traction numbers
Net: I want to own ZS on the AI Security thesis long-term. But at $185, into a print that needs to be great not good, with partner checks softening — the r/r is mediocre. Let the analyst day do the heavy lifting. Don't pay up for the September print.
THE VERDICT: Both bulls stay put a week from the Aug 27 print — Barclays Overweight/$300, Stifel Buy/$285. The quarter's not the question. The guide is. And the MaintainX math makes the guide work.
Barclays models 6% FQ2 billings growth — the step-down from Q1 is the invoicing transition ending, not demand rolling over. Web traffic running better-than-seasonal with YoY improvement against a tough comp. DMI and ISM PMIs constructive. RIF overhang cleared. Then there's the tell: executives bought shares on the open market post-Prometheus headlines. A team that doesn't buy stock. Until now. (Read that however you want.)
Stifel worked three platinum-level channel partners: one above, one in line, one below — though the below one still cleared Autodesk's targets. ACC strength is consistent. Procore is the thorn. ONE PARTNER LOST TWO SEVEN-FIGURE DEALS to them in the pipeline. Another sees Procore traction improving in the UK. Mixed, not horrid. Not clean either.
This is where the guide gets interesting. Barclays says FY27 revenue guidance should rise AT LEAST $65M inorganically — half a year of MaintainX. Margins hold at ~39% EBIT, EPS dilution just $0.10–0.15, call it 1%. A $3.6B deal with that little bleed is genuinely clean. The knock: Stifel's partners either didn't know about MaintainX or got minimal info from Autodesk. (Comms gap? Or just the wrong channel for that conversation?) They all think the deal makes sense.
"Autodesk has drivers to sustain at least high-single-digit revenue growth and improving profitability in coming years." — Stifel
Bull: Beat-and-raise base case. Web traffic positive, RIF noise gone, insiders buying, MaintainX juice with negligible dilution. Stifel explicitly expects "typical quarterly upside." At $246.79 ($52.1B market cap) with the $300 bogey out there, the r/r works if FY27 guidance moves up.
Bear: Procore wins the big ones — two seven-figure pipeline losses aren't rounding errors. Checks are mixed, not clean. Guggenheim's "Buy" at $245 is BELOW SPOT — not a vote of conviction from new coverage. Goldman sits Neutral at $260 on the construction/manufacturing macro. The bull case rests on guidance, not on what's already in the quarter.
Barclays $300, UBS $290, Stifel $285 on the bull side. Goldman $260 Neutral and Guggenheim $245 Buy on the other. That's a wide band — new coverage starts bracket the stock instead of clearing it. Earnings a week out. The quarter is what it is. The print is about the guide, MaintainX integration optics, and second-half commentary. (The AWS collab helps the cloud narrative but doesn't move FQ2.)
MDB is the AI database trade everyone's been waiting for — and it's now showing up in the data. Into Wednesday's print, this is a beat-and-raise setup. Don't overthink it.
A wave of sell-side notes hit the tape ahead of the print. PTs now $430-560, with the majority clustered at $475-560. (Goldman's $453 sits below the pack, but they're still Buy.) The common thread: ATLAS CONSUMPTION IS ACCELERATING — guided at 26% growth, tracking 29%+. Sales execution is improving, and AI-native workloads are layering in. Citizens' DD is the strongest independent check: 15 data points, 14 positive, one negative. Second-quarter sales attainment improved.
"npm download growth has accelerated to more than 80%, compared with approximately 55% in the first quarter of fiscal 2027."
That's Goldman's most compelling datapoint. It's a leading indicator for Atlas workloads. Consensus now expects at least 29% Atlas growth in Q2, and the company should raise H2 guidance. The Street models Q3 at 22%; Goldman sees mid-20s. Rate of change is positive.
Bull case: MDB is the later-cycle AI beneficiary — enterprise app workloads are just starting to hit the database layer. The discount is real: 9x FY28 EV/Sales vs SNOW at 15x and DDOG at 16x. If Atlas reaccels to 30%+, that multiple compresses on its own.
Bear case: The SBC-adjusted EV/FCF is 866x. Not a typo. You're paying for growth, but the cash generation isn't there yet. And "later-cycle" cuts both ways — if AI capex slows, MDB is the first ticket sold. Citizens found one negative datapoint, and the stock already ran 34% in a month. The r/r into earnings is no longer asymmetric on the long side.
Verdict stays: I'm a holder, not a chaser. The numbers are likely good, but the stock has front-run the fundamentals. If the print doesn't blow the doors off, the multiple contracts. If it does, these PTs start looking conservative. The trade is to own it, not add it.
Verdict: The -$2.98 EPS print is horrid. The CS-4 reveal says it doesn't matter for the revenue story. Stock's down 16% on the week to $220, and the tape is fixated on the margin crater. But Supernova 2026 just showed the product cycle is accelerating — and three analysts now converge at $300 for opposite reasons. Needham and Rosenblatt anchor there on conviction. Mizuho trims to $300 from $310 on margin pressure. Same target, different strokes. That's the trade.
CS-4 runs three WSE-3 Turbo chips, doubling the WSE-3's PFLOP performance. The system delivers up to 6x the CS-3's performance, 10x throughput per watt, and stretches the tokens-per-second advantage over GPUs to 30x. That last number is the whole pitch: at 30x token throughput, customers can charge premium AI inference pricing tiers and still crush GPU economics. This is an inference story, not a training story.
The "backpack" subsystem is the sleeper detail — decouples power from compute, attaches vertically to the power array, cuts deployment cost and time at scale. That directly answers the "great silicon, impossible to deploy" critique.
Rosenblatt frames the roadmap as alternating levers: architectural gains this cycle, WSE process node transition next. If that cadence holds, management's "double performance every year" line reads as a plan, not a promise.
"The CS-4 is evidence that Cerebras can continue to drive higher AI inference performance which allows customers to offer higher pricing tiers of token generations." — Rosenblatt
Record Q2 revenue: $209.9M, +103% YoY. That's the headline they want. Adjusted EPS at -$2.98 vs -$0.18 consensus is the headline nobody wants. Sept guide at $215M vs $216M consensus — in line, so the top-line trajectory hasn't wobbled. Nine analysts revised estimates upward anyway. (That's a tell: the sell-side is looking through the P&L to the capacity ramp.)
Manufacturing capacity is scaling for a big FY27 revenue step-up. CS-4 ships THIS QUARTER. CS-5 lands in 2027. The product train does not stop.
$220 is the line in the sand. Down 16% in a week with product momentum intact smells like flows, not fundamentals. Hold $220 and the $300 PT cluster is the air pocket. Break it and the next stop is wherever the growth crowd capitulates. The Sept guide was in line — the real tell is the whisper number and any CS-4 customer wins off that endorsement list this quarter. That's the r/r that matters.
BIDU IS A FALLING KNIFE WITH A FANCY AI PITCH. BELIEVE THE TAPE, NOT THE STORY. Stock got hammered — closed $90.87, DOWN NEARLY 13% FROM $104.12 — now 7% off the $84.82 low. MS cut to Underweight, PT $80 (from $130). Barclays followed to Equalweight, PT $96 (from $124). Down 30%+ YTD. The downgrade cycle is still rolling; don't be the hero catching this one.
EIGHT STRAIGHT QUARTERS of YoY ad revenue declines — and the slope is steepening. Barclays has it at ~20% IN RECENT QUARTERS; MS says -18.5% IN Q2. Both refuse to call a near-term turn. That's the crux: high-margin ads are the margin engine, so the P&L compresses faster than the top line. MS models core operating profit FALLING TO 2.3B YUAN FROM 3.8B YUAN IN Q3 — operating margin back to 9%, a historical low.
Q2 print was ugly. NON-GAAP EPS $7.22 VS $9.84 EXPECTED. Revenue $31.33B vs $31.95B. The EPS miss was the bigger miss — operating leverage running in reverse. Plus a sharp drop in other income. No line item saving them.
MS sees AI CLOUD INFRASTRUCTURE +66% YOY — accelerating past Q2's 50% print. Capacity scaling to 2+ GW from <1 GW. New AI marketing tools? FLAT IN Q2. The monetization bridge is still under construction.
They hired Tianxiang Sun to run the AI model unit in July, restructured Ernie. Goal: state-of-the-art in 1-2 quarters. Ambitious. Also dilutive to margins exactly when the ads business can't fund it.
"Limited visibility on turnaround timeline for advertising revenue." — Morgan Stanley
That one sentence is the whole trade. Everything else is noise.
Bear: Eight consecutive quarters of ad declines, accelerating. No timeline for a fix. AI capex hits margins when the highest-margin legacy business is shrinking. PT $80 implies 12% more downside — and NINE ANALYSTS ALREADY CUT FORWARD ESTIMATES. Buybacks capped by AI spend; no artificial floor.
Bull: AI cloud compounding at 66%. KLX IPO delayed, not dead. $1-1.5B DIVIDEND still on track for year-end. At 10x 2027 P/E, a lot of bad news is already in the number. The risk/reward only flips when the ad bleed shows a pulse.
The stock sits 7% above the 52-week low with the sellside PT now BELOW THE CURRENT PRICE. That's a setup for more downside, not a bottom. The only capital return is the dividend, and AI spend limits buybacks. Watch the sequential ad revenue delta — if the YoY decline stops widening, the narrative breaks. Until then, PMs, this is a stand-aside, not a buy-the-dip.
Two Buys, PTs 93% apart. Jefferies at $32.80. BofA just trimmed to $17 from $19 on "valuation." Stock trades at $7.80, down 41.5% in six months, another ~3% off yesterday. Whatever the quarter said, the tape isn't listening.
Q2 revenue $36.2M, +69% YoY, squeaked past consensus by ~$1M. Robotaxi revenue +691% YoY. Fare-charging revenue +849%. Robotruck +40%. Gross margin 17.5%, up 1.4pp YoY and 1.3pp QoQ — BofA flags mix, robotaxi pulling the weight. Non-GAAP loss $44M, flat YoY. The beat was on revenue, not on the P&L.
This is where the two reads diverge. H1 capex $44.6M against a ~$100M full-year guide — H2 gets heavy. Q2 alone was $32.2M as the Gen-7 fleet rollout ramps. Jefferies doesn't care. Their logic:
"Overseas expansion is increasingly shifting toward JDM, allowing Pony AI to scale internationally without taking on most of the vehicle capex itself."
That's the unlock. If the JDM model holds, the capital intensity per incremental dollar of revenue falls off a cliff. BofA's PT cut on "valuation" reads like they're a step behind Jefferies on that — or a step ahead on the reality that $44M quarterly losses against $36M revenue is a hard table to underwrite.
Bull: 849% fare-charging growth is an inflection, not a blip. Margins expanding. Cash > debt, current ratio 16.12. At $7.80 the market's pricing in failure, not a 69% grower with a structurally cheaper path to scale.
Bear: Revenue is smaller than the loss. TTM revenue $110M vs a $3.47B market cap — ~31x. The H2 capex step-up hits right as JDM benefits are still theoretical. The cash cushion buys time, not proof.
Positioning call: Jefferies has been right to stay long, but the spread between $17 and $32.80 says this is a model-sensitivity name, not a consensus one. If robotaxi mix keeps beating, the bears look silly at $7.80. If capex surprises higher, neither PT matters.
TD Cowen isn't moving — Buy reiterated, $500 PT, stock at $379. That's ~35% implied upside, and the thesis rests on one number: >$100B in FY27 AI semiconductor revenue. If all ~10GW of backlog actually deploys, that's the floor, not the ceiling. Fundamentals back it up — 32% revenue growth LTM, 76% gross margin.
The nuance is in the mix. TD Cowen is solidly behind the Google TPU franchise but explicitly wary of everything else. Meta's MTIA timeline softened again last quarter. OpenAI's ASIC is still unproven. And Mediatek taking share of the TPU ramp remains a narrative headwind — even if their checks say the actual revenue impact is manageable.
The Google relationship is the whole ballgame, and TD Cowen thinks Broadcom has earned trust there.
"Broadcom has earned the benefit of the doubt in retaining high market share at Google given the technical and commercial challenges of ramping a new compute platform and its long-term agreement with Google."
Separately, BofA cut the credit rating to Marketweight on XPV/Blackstone/Apollo uncertainty — not an equity call, but worth knowing as a risk flag. Nvidia's $500B compute financing platform with BlackRock/Goldman is the other AI infrastructure story in the tape; if that scales, the whole AI semi complex re-rates.
Evercore ISI pushes Dell PT to $550 (Outperform), and the bull case isn't the AI server order book — it's storage. The market's treating storage as a boring cash cow; Evercore sees it as the EPS accelerator the market hasn't priced.
Storage is only ~10% of FY27E revenue, but prints LOW-20S OPERATING MARGINS versus mid-single digits for AI servers and mid-teens for traditional servers. DOUBLE-DIGIT storage growth + mix shift toward Dell-controlled IP and software-defined storage = ~$2.00+ EPS contribution today, and potentially ANOTHER $2+ of FY28 upside if margins take out high-20s (base) or approach ~32% (bull).
"Dell's storage portfolio remains underappreciated and should enable revenue and profit acceleration."
Valuation still screens cheap: 24x FY28 EPS on a 0.38 PEG, with 38.6% LTM revenue growth and a 251% one-year return. Yes, shares sit above InvestingPro's Fair Value (flagged Most Overvalued), and DRAM/NAND supply constraints plus hyperscaler concentration are live wires. But the rate of change in Dell's margin mix — not the AI backlog — is the 2028 story.
Northland upgrades CIEN to Outperform, PT to $500 from $450 — this is a momentum-dip buy, not a fresh discovery. Stock sits at ~$405, off the $637 high but STILL UP 357% OVER THE PAST YEAR. The upgrade thesis rests on supplier signals: LITE's strong print and CSCO's commentary confirm the scale-optics market is bigger than expected, and LITE's plan to 4X pump laser capacity is the supply-side tell that demand visibility is real.
"Indications of a larger than expected market for scale across solutions evident in recent reports and comments from LITE and CSCO."
Needham's the other bull in the room, reiterating Buy at $600 on the Verizon-Google dark fiber project — $1B+ program, $100M+ CIEN revenue tailwind across FY27-28. Add the $2.875B zero-coupon convertible raise (0.00% coupon, smart money when rates are this high) and you get: pullback + supplier confirmation + funded backlog. Risk: this multiple is extended as hell after a 357% run, so any LITE/CSCO hiccup hits CIEN first.
Stifel lifting OKTA to $160 from $120 (Buy) into the Aug 26 print — stock already up ~58% over the past year, trading at $143.92. They're modeling ~2% revenue upside and 2.5-3% cRPO upside, roughly in line with recent quarters. The focus list: O4AA/A4AA AI traction, Permiso close, and the newer product stack (OIG, OPA, ITSM, ITP, Device Access, FGA) starting to show up in go-to-market.
Identity security has become a top security priority.
That's the macro hook, and it's why the broader analyst complex is stretching targets. Wells Fargo went to $180 (OW) on low-teens growth plus AI model leverage. Citizens to $170 on cross-sell/upsell. KeyBanc to $175 on ID-JAG alignment and the Anthropic partnership. Bernstein's survey has cyber spend outpacing IT budget growth, and Goldman sees AI-driven security budgets inflecting by 2027 on a cloud-cycle timeline. All three upgrades clustered in the last few days — this is a momentum tape, not a value call.
AI agent identity is the bull case's center of gravity, though Stifel explicitly calls it early stage. That's honest. The market is paying for 2028 today. Not sure we need to chase here, but into earnings the bias is clearly up — the risk is less "miss" and more "beat-and-undeliver on the AI narrative." Core IAM is fine, cRPO has been consistent, and the Permiso deal rounds out the identity security story. Position is likely crowded by now; watch the 160 strike for resistance on the print.
Goldman's pressing the bid into earnings, reiterating Buy with a $106 PT as the Street resets for a Subscription ARR beat. GS models >32% growth — Q1 printed 32.5%, and that's ~$18M ABOVE consensus. The stock's already up ~30% in a month (IGV +10%), sitting at an all-time high, ~11x FY28 EV/S. This is a "good number's priced in, better number prints" setup.
The tape's constructive for a reason. Channel partners beat plan and entered 2H with healthy pipelines as the accelerating AI threat landscape pushes businesses to reassess recovery readiness. IDENTITY ATTACH IS THE SLEEPER — >$100M ARR by year-end vs. >$50M in Q1, with a medium-term path past $200M. On-prem data protection (roughly 20% of NNARR) keeps trending lower, so the mix story stays intact. Memory-cost inflation? No material impact on conversion or pipeline formation.
GS sees ARR growth exiting Q4 above 30%, with multiple paths to sustain premium growth vs. Street low-20s estimates beyond FY27. The rest of the sell-side is aligned: Loop (Buy, $100), BMO (Outperform, $98), Baird ($110, Fresh Pick), Cantor (Overweight). No one's short this into the print. Risk: stock's a buck off the 52-week high at $100.55 — the +86% six-month run means positioning is crowded. But momentum's the trade until the print says otherwise.
Cantor is dismissive of the PA governor's new data center executive order — and the read-through is you should be too. Firm reiterated Overweight and $5 PT into the headlines. Stock sits at $3.16, DOWN 10.7% ON THE WEEK but +147% OVER THE PAST YEAR. The political backdrop writes itself: PJM capacity auction prices at RECORD HIGHS, ratepayers on the hook for ~46% of capacity charges, and 20 proposed data centers in a state with 14 existing/in-development sites holding active exemptions. Shapiro had to throw a bone to voters.
Keel's three PA facilities (Panther Creek, Sharon, Scrubgrass) were already deep in the permitting pipeline before the order dropped. Cantor's take: this is forward-looking, not a retroactive gut-punch.
Cantor Fitzgerald noted the executive order is forward-looking and should not present significant obstacles for the company.
The real story remains the transition. Q2 revenue FELL TO $30M FROM $61M YoY as the Bitcoin mining pivot to HPC/AI infrastructure eats the P&L. Loss from continuing ops widened to $64M ($0.11/sh) vs a $13M profit a year ago. But the balance sheet is the bull case: $819M IN CASH AND MARKETABLE ASSETS as of early August. That's the war chest for the AI data center buildout. The PA headline fades; permitting progress and cash deployment are the variables that actually move this one.
THE AUG 27 PRINT IS A SIDESHOW. Barclays keeps OW/$200 into F2Q27 but frames it as "muted top-line beat, solid margin upside, largely reiterated guidance" — code for nothing that forces a repricing. WDAY +48% in 3 months vs IGV +10% means the easy money is made. The stock trades on Silver Lake speculation (Reuters, Aug 13, mgmt silent), and that's the only thing that matters near-term.
Guidance math does the talking. Current guidance implies subscription growth decelerates to ~11-12% in 2H — that's the bogey, and it's not getting better before October. The valuation isn't crazy (13.5x EV/2027E FCF, 4.2x sales, PEG 0.72), but you're paying for a decelerating franchise with a PE overhang. R/r is muddled at best.
Street's all clustered in a $200-220 box: DB cut to Hold ($220 PT), UBS Neutral ($220), Wells Fargo $215, BMO still Outperform. Nobody's paid to get aggressive here. The catalyst queue — October Investor Day and Q3 results — matters more than the print. Don't trade the quarter; trade the Silver Lake headlines.
FREEDOM CAPITAL OPENS HUT WITH BUY, $132 PT — but that's a TWO-YEAR target, not 12-month. Stock sits at $80.78, already +231% over the past year, beta 6.11. This is a cautious buy, not a conviction stamp. Thesis: crypto miner turning AI data center operator, playing the same collocation game as CORZ and APLD.
The catch: Freedom explicitly prefers Applied Digital in this subgroup. HUT gets tagged as riskier, earlier-stage in its buildout. The firm's suggested stop-loss at $87 lands within 10% of the $79 worst-case valuation — that's the analyst telling you the downside is bounded but real. This is a trade with guardrails, not a buy-and-hold endorsement.
"Hut 8 is riskier because the company is at an earlier stage of its journey compared to peers."
Fresh tape doesn't help the bull case: Q2 adjusted rev $105.18M vs $111.29M est (+81% YoY — the pivot is working, directionally), but EPS a -$1.78 loss vs -$0.519 expected. Horrid bottom-line print. The PT is all 2027 story; the fundamentals are still catching up. Steelman: AI colocation demand is real, HUT's power assets are scarce, and 231% tells you the market already sniffed that out. At a 6.11 beta, PMs should size this like the lotto ticket it is — 2-year PT, tight stop, early innings. That's a risk-managed position, not a religion.
Mizuho raises PT to $16 from $13 (Neutral) — the lone bull-ish move, leaning on SASE displacement, cross-sell, and AI security optionality. But the broader tape points the other way: FOUR FIRMS JUST CUT PTs INTO A $13-19 CLUSTER FROM A PRIOR $14-25 RANGE. The collective message: ARR deceleration is the real story, not the AI narrative.
That's what makes the +75% run since mid-June (vs HACK ETF +18%) feel ahead of fundamentals. Mizuho expects a "solid" FQ2 print on September 2 — but solid is priced in. The market needs to see ARR re-accelerate, and Mizuho isn't underwriting that. Balanced r/r at $16; the street's high targets got cut for a reason.
"Remains uncertain whether Netskope can deliver enough ARR upside over the next few quarters to drive the shares significantly higher."
MS cuts PT to $14 (from $16) but KEEPS OVERWEIGHT — the cut is pure debt math, not a business downgrade. Stock sits at $6.58, hugging the 52-week low of $6.17, while the target range spans $7.55-$24.98 with consensus Buy. That gap between tape and model is the whole trade.
MS actually RAISED 2026-28 adj EBITDA estimates 2.4-3.2% on opex control, then shaved the target 13% after revising net debt forecasts. New target implies 11x 2027 EV/EBITDA — cheap IF you trust the order book. The bogey is the balance sheet: ~$5.5B net debt and a 7.52 debt-to-equity ratio. Heavy capex is the market's obsession.
Q2 print was fine: revenue 2.78B yuan (+14.2% YoY, just under the 2.80B bogey), adjusted EBITDA +25.4%, wholesale IDC revenue +30%. Growth is real. This is a balance sheet story, not a demand story.
Tom Tang (MS): "sustained EBITDA growth backed by a clearer order trajectory."
Bull case: EBITDA compounding + visibility = re-rate once the debt scare clears. Bear case: at D/E of 7.52, one bad capex cycle and equity gets crushed. The 2x upside is the market pricing in distress — either it's right and you wait, or it's wrong and this is the entry.
Stifel's staying on the bid. The EPIQ acquisition for $1.1B cash is another step "up the chain," and the market's not paying up: 17.4x synergy-adjusted 2027 EBITDA, roughly where TTMI itself trades, but for a business with mid-30s EBITDA margins. That's MIX-ACCRETIVE to a margin profile Stifel already thinks the street underprices. Net leverage goes to 2.3x at close, but the path back to 1.5-1.7x in 12-18 months is credible given FCF.
"Provides another example of the 'up the chain' directive management framed at May's Investor Day and is mix-accretive to a margin structure the firm already views as understated."
Meanwhile Q2 was a blowout — first time through $1B in quarterly revenue, $0.99 non-GAAP EPS vs $0.92 est. The stock's up 237% in a year, and Stifel's $205 PT still implies ~45% upside from the current $140.98.
Needham to $100 from $75, Buy maintained. The message out of Monday's investor meetings with CEO Collins and CFO Del Preto: the Carbon6 divestiture resets the story to one of VISIBILITY AND PREDICTABILITY. Management is back to calling it a high single-digit grower near term — with genuine levers to get back to double digits.
Needham said it gained confidence in its view that the business will return to historical high near-term visibility and strong predictability following the recent Carbon6 3P divestiture.
The levers: AI Max monetization, improving gross retention, tariff impacts still dissipating. That's the bull case. The honest asterisk — the divestiture lowers organic growth rates going forward. They pulled the Band-Aid off. Fine.
The numbers back the narrative: Q2 REV $197.8M vs $195.45M consensus, recurring revenue and core ops driving the beat, margins and cash flow up. Stock at $77.38, +33% over six months — market already sniffing this out.
The frame: above-trend growth 2020-2024, below-trend through H1 2026, now normalizing back to trend with predictable margin leverage. That's a "clean base" setup. Needham's PT implies ~29% upside. Not a screaming bargain after the run, but this is a quality compounder with a cleaner story and the sell-side is just starting to re-rate it. Solid r/r, not a hero trade.
MS CUTS PT TO $12 FROM $13.10, KEEPS OVERWEIGHT, AND THE STOCK SITS AT $6.00. That's a 100% gap between price and target — either the tape is pricing in death or the street is pricing in a miracle. WeRide's YTD -31% isn't a fundamental collapse; it's a sector-wide de-rating from China robotaxi scrutiny, Middle East geopolitical noise, and fund outflows. Q2 REVENUE +82% YoY TO RMB 232M, LTM GROWTH +102%, gross margin 37.5% (+143% gross profit), and overseas revenue now ~40% of total, up 164% YoY. The growth is real.
The bull case: China resuming robotaxi license approvals suggests the worst is behind. Management's pivot from development-heavy to commercial stage is showing early operating leverage — OPEX only +9.2% YoY. BofA also trimmed (to $10.70 from $11.80, still Buy), so the PT cuts are valuation math, not thesis breaks. The bear case: aggressive AI R&D spend eats cash, analysts don't model profitability this year, and the stock trades at $1.97B market cap — options are getting repriced for a China policy overhang that nobody can model with precision.
"L4 autonomous driving is the goal and L2++ is the means."
That's the smart framing. L2++ becomes the cash engine funding the L4 robotaxi moonshot — and it's already scaling globally. Not sure we can read too much into a single PT cut to $12 when the stock trades 50% below it; the signal is MS staying long while acknowledging the tape doesn't care yet. Risk/reward skews to the upside if license approvals keep resuming and Q3 holds the 80%+ growth clip.
Mizuho's the lone voice but they're not blinking: Outperform, $45 PT, stock at $29.11 — down 52% on the year. Their core claim: 2026 guidance is conservative, especially on revenue. That's a beat-and-raise setup on a name the market already threw out.
UPST did $364.7M revenue vs $347.2M consensus — clean top-line beat — and flipped to GAAP profitability. The miss was EPS: $0.16 vs $0.18 expected. Revenue growing 45% LTM and the stock still trades here? That's positioning, not fundamentals.
Mizuho's diligence addresses the key debates:
The risk nobody flags: revenue beats but earnings lag. Cost of capital still eats the P&L. If funding structure maturation next year lands, this is a $45 stock. If it slips, sub-$20 is live. I'd want to see an originations inflection before paying up — Mizuho's right that the setup is good, but the catalyst timing is everything.
Noble Capital just lit up XRX with an Outperform and a $5 PT — 69% above the $2.95 close. The bull case is pure turnaround math: Lexmark integration, IT/digital services mix shift, and cost discipline betting on a multiyear earnings recovery. Shares already up 54% in six months, so Noble is leaning into momentum, not catching a falling knife.
The neat trick in the PT: $5 implies 7.1x EV/EBITDA on FY28 forecasts — below the 8.5x EV/EBITDA the stock trades today. That's how you get 69% upside without asking for multiple expansion. It's all about EBITDA growing into a lower multiple. Execution-dependent? Obviously. But the Q2 print gave them cover: adjusted EPS of $0.38 vs. consensus -$0.08, revenue just a hair light at $1.92B, and guidance raised. Debt paydown continues post-Lexmark.
Risk/reward is the phrase Noble pushes — and it's fair. But for PMs, the real question is whether revenue declines actually moderate to the point where the recovery shows up in FCF, not just adjusted EPS. Noble's PT is a multiyear target; don't mistake it for a 12-month signal.
"Noble Capital believes these initiatives can moderate revenue declines, improve profitability and cash generation, and ultimately support a multiyear earnings recovery and valuation re-rating."
That's the entire thesis in one line. If execution slips, this is a $3 stock with a yield and a value trap narrative. If it works, $5 is the floor, not the ceiling.
Mizuho holds Outperform/$750 straight into the bellwether trial — an analyst saying the market's got the risk calculus backwards. META trades $543.67, basically pinned to the 52-week low ($520.26), DOWN 27% OVER THE PAST YEAR. Fear's already in the price; the question is whether a verdict breaks the model, not whether it blips the stock.
The June expert event (a litigator who knows the Big Tobacco playbook) saw parallels but real differences. Fines in the tens of billions are plausible — that's not the kill shot.
"A fine in the tens of billions of dollars was possible but emphasized that potential behavioral remedies represented the biggest risk to the company."
Behavioral remedies are the tail. Mizuho's March history check: two negative verdicts cost META just TWO DAYS of negative relative returns; the stock recovered fully within TEN TRADING DAYS. That's the rate-of-change tell — dip buyers treat headline sells as entry points, not regime changes.
R/R is asymmetric from here. No verdict = relief rally. Bad verdict with no remedy = same dip-and-recover tape. Worst case (remedies) is what you'd hedge, but Mizuho isn't building the base case around it, and neither should you.
Buy reiterated, $55 PT — and DA Davidson keeps ratcheting up (went $45 → $55 as the Domo deal came into focus). Stock prints $43.78, so that's ~25% upside on a 0.34 PEG. You're getting a ~40% adjusted operating margin software business at sub-1x growth-adjusted valuation. That's the whole ballgame.
CEO Yogesh Gupta hosted an investor call Monday to sell the $400M Domo asset purchase as a data transformation and governance story. DA Davidson came away incrementally constructive on demand durability — the bet is Domo's tech platform, customer contracts, and IP extend PRGS's data/infrastructure reach into a bigger cross-sell opportunity. CITIZENS STILL HAS DOMO AT UNDERPERFORM WITH A $2.25 PT — the asset's been a problem child, and integration is the swing factor. But at 0.34 PEG, the market's pricing in zero upside. Fine risk/reward into the deal close.
Guggenheim stays on the sideline into the print, but the setup has teeth. The real story isn't the quarter — it's Duo Agent: paid consumption run rate hit ~$20M exiting June (nearly 2% of ARR), up from $15M in April, and NONE of that sits in FY27 guidance. That's the free call option, and the lack of history makes it a coin flip to model.
Base case: Q2 FY27 revenue growth of 19%, a 400bps QoQ decel against last year's brutal 29% comp. Guggenheim expects Q3 guide in line, the Q2 beat largely passed through, and FY27 guidance nudged up ~50bps to 17% growth at the midpoint. Checks show the core business holding up — partner seat growth stable to slightly higher on tool consolidation and self-managed expansion where the governance/security pitch lands. Gross margins at 87% LTM are quietly elite.
The bull case writes itself: +44% over six months, Duo Agent adoption accelerating per Guggenheim's AI coding survey, and usage-based upside that guidance ignores. The bear case: 23x EV/FY28 FCF with monetization still lumpy and trial-to-paid conversions unproven. Steelman on both sides, but the stock has already repriced a lot of good news.
"Guidance does not factor in any Duo Agent Platform contribution in fiscal 2027, which could be a source of upside despite difficulty forecasting the metric given lumpiness and lack of history for trial to paid conversions."
Elsewhere: UBS nudged PT to $35 from $32 on improved demand outlook; Cantor holds Neutral at $35; Bernstein stays Outperform at $60 calling it a tactical buy. New Chief Business & Legal Officer in the form of Thomas Lloyd. Nothing changes the core question — is Duo Agent a $20M curiosity or the next leg?
Cantor's $27 PT is a $1 nudge — the real signal is the $500M, 55MW data center lease DUOT dropped Monday. Stock ripped 38% to $12.58. That still leaves ~115% upside to the new target, but a 38% one-day move on a $6M quarterly revenue base is a narrative print, not a fundamentals print.
LARGEST CONTRACT EVER. Total contracted capacity now 75MW. Cantor expects most of it generating revenue by END OF Q1 2027. Management's 2027 revenue bogey: $160M — with ~$100M annual EBITDA already locked from existing contracts (62% margin on that revenue target? Slick if it holds).
Bull case is simple: power capacity is the scarce asset in AI, and DUOT figured out how to source 75MW and convert it into contracted revenue. Cantor raised FY28 EBITDA estimates off this exact loop — add power, print contracts.
Bear case is equally simple: Q2's headline beat was a mirage. EPS $1.61 vs -$0.02 expected, yes — but driven by a $53.2M one-time gain from the New APR Energy sale. Revenue MISSED: $6.18M vs $10.7M forecast, a 42% shortfall.
"Duos highlighted the quarter as a pivotal moment in its transition towards AI infrastructure and edge data centers."
One-day 38% prints don't happen on fundamentals. They happen on narrative. Cantor's $1 PT bump tells you the analyst wants delivery before giving the story full credit. Delivery date: Q1 2027. Until then, this is a positioning tape — and the positioning just got a lot more crowded.
GS drops a Sell on CRSR this morning — $11 PT, which the press release frames as 100% above the $5.48 tape. That math is bizarre for a Sell (double upside?), so I'd treat the PT as mostly ornamental. The real signal is the thesis: AI is eating the memory supply chain, and consumer PC hardware is the residual claimant. GS sees PC UNIT SHIPMENTS DOWN 15% IN 2026, ASPs UP 18%, and the Gaming Components segment (67% of revenue, 54% of gross profit) missing FactSet by 2/12/15% across 2026-28.
"Rising PC system build costs could crowd out consumer discretionary budgets and stretch replacement cycles."
The clever part: this isn't just a components story. GS explicitly ties the memory crunch to the higher-margin Peripherals segment — when build costs spike, gamers stretch upgrade cycles, which stalls the accessories attach rate too. So both segments catch a headwind simultaneously. Supply constraints run through at least 2027. If you're long CRSR here, you're betting gaming discretionary spend holds up against an input-cost supercycle. That's a bad r/r, and GS is right to flag it.
Stifel fires the contrarian shot — PT to $90 from $65, Buy, and the whole thesis is "cheap enough to own the wait." At 13x CY27 EV/FCF with 76% gross margins, the market has already punished the cloud deceleration. (Stock sits ~$83, so this is an 8% call, not a hero swing.)
The bear case isn't hiding: CLOUD REVENUE FELL QOQ IN Q4 FOR THE FIRST TIME — $217.4M vs ~$220M consensus, growth slowed to 20% YoY. That's the headline. Morgan Stanley threw in the towel with a downgrade to Equalweight on growth-inflection concerns. The restructuring (7% workforce cut, CPO Exner's departure) adds an overhang nobody loves.
But Stifel's counter is positioning, not fundamentals. SaaS comparisons ease through FY27. Management stays agnostic on consumption — contracted SaaS and self-hosted both count. And the rest of the street isn't fleeing: UBS trims to $85, Rosenblatt to $83, Piper holds at $85 — all Buy/Overweight. They're shaving PTs, not abandoning ship.
Steelman for Stifel's $90: you're paying 13x for a 76% GM, 17% grower with comps finally turning tailwind.
Steelman for the bears: the organic SaaS engine stalled once. One QoQ miss into a restructuring is a pattern forming, not a blip.
"There remains a positive bias in the name in coming quarters based on the valuation."
That's the whole trade. Valuation over inflection.
GLJ Research out with the same $24.86 PT and Sell — the headline is noise, but the FSD telemetry they're pushing is the actual bear thesis and it's worth chewing on. TSLA already -24.55% YTD vs S&P +13.14%, and GLJ sees the bleeding continuing into H2 2026 toward sub-$200. The direction of travel matters more than their price target (a permabear caricature at this point).
The numbers underneath are genuinely horrid: Q2 operating margin 1.4%, FCF -$1.1B, energy gross margin nearly halved to 20.4%. But the core argument is the software — crowdsourced FSD v14 on HW4 disengaging EVERY 40 MILES (865 tracked vehicles, just 18 active in the past week). Critical disengagement comparison: ~2,000 miles for FSD vs ~500,000 for the average human. That's not a close race, that's a different sport.
The firm called the Austin Cybercab launch "a promotional event for a two-seat vehicle with no steering wheel."
The bull case isn't dead — TD Cowen reiterating Buy on AV positioning, Einride's 500-Semi deal for Amazon gives the industrial side a pulse. But with the stock at 315x P/E, you're paying for optionality, not fundamentals. GLJ's $24.86 target is a punchline. Their FSD data isn't.
JPMorgan just dropped the hammer — CCOI to Underweight, PT to $9 from $22. Stock trades at $10.13, DOWN 72% OVER THE PAST YEAR, 61% IN SIX MONTHS. The gap to the new target isn't massive, but the message is: no line of sight to the 4.0x leverage target, period.
The thesis is brutal in its simplicity. Legacy Sprint contract runoff keeps crushing revenue, and the Waves business — the supposed growth story — keeps disappointing on install pace and customer acceptance. Data center monetization? JPM says that catalyst is largely behind the company. NET LEVERAGE AT 6.75x, LEVERED FCF NEGATIVE $143M, current ratio 0.63. That's a balance sheet in the danger zone, not a trading vehicle.
One wrinkle in the wreckage: Q2 EPS printed $1.38 vs -$0.97 expected, but revenue missed ($235.6M vs $239.8M). Margins are fine. Top-line is broken. That's the whole story in one quarter.
"The data center monetization catalyst is now largely behind the company."
JPM's $9 target implies 9.9x 2028E EV/EBITDA — at current levels you're paying 8.1x 2027E. Not a value trap. A falling knife with leverage. Wait for actual proof of install acceleration before catching this one.
Citizens keeps Market Outperform and $100 PT — ~34% upside from the $74.66 close, and the stock trades 11.1x the firm's 2027E EBITDA of $14B vs. a 15x target multiple. Global delivery leadership plus a top-tier mobility book; that's the whole thesis. Bears will cite Tesla's CyberCab in Austin, but the Pony.ai expansion (2,000 robotaxis in Europe) is the counter — and Mizuho ($112 PT) plus Piper Sandler ($105 PT, bookings growth >20%) back the same side. Don't even need AV to make the long work; core business at 11x forward EBITDA does the heavy lifting.
"Uber's global delivery platform holds a leadership position across multiple markets."
HXSCL (SK hynix) — KRW 40T BUYBACK/CANCELLATION is management drawing a floor under the three-day chip selloff. Roughly 3.3% of shares cancelled in three months — largest ever for a Korean listing — and it's balance-sheet funded (KRW 69T net cash). The "at least 50% of FCF" return target is a structural capital-allocation shift, not a one-off. Watch FCF/capex splits next two quarters to validate. HBM moving from commodity to custom co-design extends LTA visibility; the bear case is HBM4e 12Hi qualification slippage.
SSNLF (Samsung) — Foundry raised SF4/SF5 prices 10-15% for US/China clients, 8nm up ~10%. New orders only, so revenue hits 2-3 quarters out. This is the FIRST HARD PROOF AI demand is spilling into wafer pricing beyond HBM and packaging. If TSM follows, the whole advanced-logic chain reprices. P5 triple-fab expansion and Google SF4 talks add optionality, but watch the cannibalization: memory wins, the device division pays (rising memory costs are an explicit device margin headwind).
TSM — Samsung's 15% hike is a leading indicator. CoWoS 5.5x-reticle in volume production with yields >98%, scaling beyond 14x reticle by 2029 — packaging is now a competitive weapon, not a bottleneck. Worker fatality at Fab 20 is an ops/execution risk, not supply impact. Intel Malaysia absorbing packaging overflow is the first sign CoWoS isn't the only allocation game.
MU — SIXTEEN strategic customer agreements: 5-year take-or-pay, no-cancel, locking ~40% of revenue, backlog near $100B, >$20B upfront deposits. Memory now trades like a contract business, not a commodity. HBM bit shipments quadruple by 2027 vs 2024, and each HBM bit eats ~2.5 DRAM bits — that loads supply/demand well into 2027. The "engineering out HBM" thesis is getting shredded by practitioners who'd benefit directly from cheaper memory.
SNDK — NAND complex tightening: top-5 supplier revenue +77% QoQ in 2Q26 with sharp ASP increases. The upcycle broadened past HBM. HBF debate matured to memory-hierarchy placement — Google and Meta are in, NVIDIA adoption is the swing factor. Just don't price HBF as a KV-cache vehicle; endurance burns on transient writes. Kioxia XL Flash owns that use case.
CRWV — Repriced +25% ACROSS SKUs in July. A 2020-vintage A100 cluster recontracted out to 2029 at an attractive price — pricing power on old and new gear. Active power 1.5GW, WACD down ~300bp YoY (~$1.1B annualized interest savings), DDTL 5.5 securitization accepting 2-3yr contracts as collateral. Non-GPU ARR >$400M, managed inference from $1M to $100M+ in one quarter. Second growth engine is live.
IREN — The "AI hidden debt" analysis says most of the $3T shadow-capex framing is contract value and pay-as-you-go, not balance-sheet debt. Compute stays short; efficiency gains raise, not lower, the value of each incremental GPU-hour. That implies more profit pass-through to specialized GPU clouds. No company-specific print — read-through only.
MSFT — Pennsylvania's EO raises scarcity value of in-state clean-firm generation. The Crane restart (835MW, 20-year PPA) is post-2025 incremental — stronger claim on incremental clean-firm energy. Positive for CEG, broadly neutral for MSFT; they already have contracted power locked. Azure distribution survives the OpenAI/Anthropic share shift — Bedrock is ~45% of the API ARR, and clouds are demand creators, not disintermediators.
AMZN — Drone delivery expanding to ~500 cities, 1M delivery target, BVLOS approval is a regulatory unlock. Financially immaterial today; long-run logistics cost repricing option. Louisiana data center investment $12B → $18B with up to $400M self-funded water infrastructure — that template mitigates PA-style political risk. Retail media projected at 67.2% of US commercial media by 2028 — underappreciated margin offset to infra spend.
GOOGL — Don't build a merchant-TPU bull case. External TPU sales deserve skepticism; the real benefit is internal cloud TCO. Google joined the HBF consortium — architecture influence, not near-term revenue. Q2 frontier revenue ~$2.5B vs Anthropic's $11.6B — NOT monetizing the model layer despite Gemini distribution. Samsung SF4 talks add supply optionality, nothing more.
INTC — Socionext adopted 18A-P — small volume but the first tangible leading-edge merchant win. The proof point the foundry turnaround narrative needed. EMIB-T yields approaching 90%, roadmap beyond 8x reticle this year, >12x by 2028 — substrates remain the bottleneck. Malaysia absorbing TSMC packaging overflow confirms Intel has capacity optionality.
KEYS — FQ3 blowout: revenue $1.85B +37% y/y, orders $2.09B +56% y/y, Comms Solutions $1.35B +43% with 71% gross margin. Q4 guide $1.93-1.95B vs $1.83B consensus. AI and optical test demand accelerating. Cleanest read on AI capex broadening beyond the compute layer.
ASML — Samsung says Low-NA EUV with multipatterning is the DRAM workhorse; High-NA depth-of-focus is prohibitive. That pushes High-NA DRAM adoption further out. Logic remains the early adopter. High-NA revenue timing from memory just got more uncertain.
STM — THIRD PRICE HIKE of 2026, effective August 23, across multiple product lines. Broad-based pricing power beyond AI silicon. Analog and MCU demand stronger than the market assumes. If STM pushes price, the whole discrete/analog chain has pricing power.
COHR — Broad CPO portfolio on display at Laser & Photonics Taiwan: high-power CW lasers, external laser sources, silicon photonics PICs, fiber attach, PM fibers, thermal management. Portfolio breadth is real. But the optical supply chain is still immature for NPO/CPO attach — revenue ramp will be gradual, not a step function.
LITE — Field checks say NPO/CPO attach is a slow roll; manufacturing maturity gates it, not demand. Tempers the bull-case attach timelines embedded in optics names. Market's paying for optionality, not near-term CPO revenue.
HNHPF (Foxconn) — $358.4M into Vietnam for more CPO capacity after entering mass production in Q3, plus NT$859M on Mexico land for AI servers. Foxconn invests behind optical attach and server buildouts, not just final assembly. Strong read-through for AI server supply chain momentum.
ASX (ASE) — CoWoS overflow into Intel Malaysia and continued packaging demand surge benefit ASE. They say CoWoS and EMIB-T aren't mutually exclusive — substrates plus assembly/test for both standards. Advanced packaging is the bottleneck; ASE is the leveraged capacity play.
MDTKF (MediaTek) — Opened phase one of Tongluo R&D data center for AI, HPC, advanced-node design, and packaging. GUC and Novatek building similar capacity. Taiwan's custom ASIC design infrastructure scaling into 3nm and below. Google TPU potentially involving MediaTek adds a design-services layer beyond smartphones.
MRAAY (Murata) — 47µF MLCC lead times up "a lot." Strategy: raise high-end prices, cut low-end to fend off Chinese/Taiwanese competition. Passive-component tightness is echoing the memory complex. AI server BOM inflation broadening beyond compute.
IBM — Connected two modular cryogenic quantum modules: 4K in under five days, below 15mK shortly after. Up to 12x more wiring space. Targets 1,000+ programmable qubits by 2027, Quantum Starling 2029. Real progress. Programmable ≠ fault-tolerant. Long-dated option, not a 2026 story.
CEG — Pennsylvania's EO raises scarcity value of in-state clean-firm generation. Crane restart (835MW, 20-year Microsoft PPA) is post-2025 capacity with a stronger claim on incremental clean-firm energy than existing nuclear output. Policy shock is a net positive. Nuclear is becoming the AI-political hedge.
KUASF (Kuaishou) — Earnings fell the most in five years. AI spend +33% y/y while monetization lags; creator payouts compound it. First hard proof Chinese AI competition hits application-layer margins, not just hyperscaler capex. The debate — capitalized or P&L flow-through — gets settled by the next two quarters of China platform prints.
TCEHY (Tencent) — ~10k H200s delivered in China over recent weeks. Keeps some training capability alive despite export controls. Import channel is narrow and can tighten quickly — US lawmakers pushing to close the overseas-cloud loophole. Policy-driven datapoint, not a durable supply unlock.
ASIYF (Acer) — Acer says PC shipments may plunge >20% in 2H26 as higher prices bite. Memory/SSD increases plateauing, but consumer demand damage is done. The memory upcycle has a victim: consumer PC OEMs.
RDDT — Reddit's share of ChatGPT Search citations fell 3.8% → 0.5% between early and mid-August — ~86% DROP. If it persists, referral traffic and ad inventory soften. No P&L damage yet; vector is negative. Key question: feature change or permanent search-routing shift?
MRNA — +53% after melanoma vaccine study readout. Major clinical event changing platform optionality. Market pricing a real commercial path in adjuvant/therapeutic cancer vaccines. Watch durability and regulatory timing before underwriting full platform value.
AMD — No incremental company-specific data this window. TPU-merchant skepticism cuts both ways: custom silicon won't fully displace merchant accelerators near-term. AMD stays #2 merchant vendor. Frame is still NVDA vs custom silicon, not AMD vs TPU.
PLTR — OpenAI's sequential growth slower than Palantir's — the source's benchmark for AI software monetization. No PLTR-specific numbers here. Comparative datapoint, not an earnings signal. Don't over-read into a long or short.
NVR — Director bought 63 shares at ~$6,431.51 on August 7 — about $405K. Small relative to market cap. Modest insider confidence vote, not an operational signal. Treat as noise for TMT positioning.
DOCU / ENVX / GRRR / HIVE / INTU / PDFS / PLNW / RCMT / S — No fresh signal flow in the feeds today. Nothing to add without making stuff up. Sitting this one out until prints or channel checks surface.