BEAR CASE: GROSS MARGIN IS A TELL

The bears aren't arguing demand — that fight is over. They're arguing quality of the NEXT $200B. Gross margin guide reset to 72-73% (vs. recent ~74-75%) on memory cost pass-through. That's HBM cost inflation flowing through. The bull says pass-through; the bear says pricing power erosion.

Freedom Broker made the sharpest bear-adjacent point: the balance sheet now carries UNPRECEDENTED scale of supply commitments, equity stakes, and customer guarantees. That's a quality-of-earnings issue. Cash conversion weakened as payment terms extended. When you're pre-buying $279B of memory and accepting contingent liabilities, you're subsidizing your own supply chain. Goldman remains NEUTRAL at $300 — the only notable non-buy — which tells you the "priced for perfection" crowd still exists.

Then there's the allocation math. SPCX (the new $SPCX entrant) is implying ~6GW ADDED in CY27. If hyperscalers can't grow their Nvidia allocation, that's a concentration problem in the customer base. UBS flags this explicitly: heavy allocation situation, little room for hyperscaler growth. And with 70% growth guided, you need EVERY segment firing.

THE READ-THROUGH

The China AI angle is live. Goldman noted MiniMax +283% YoY revenue in H1. China AI demand is surging and that's an incremental positive for the whole chain. Also tracking: Coherent Corp says extraordinary optical component demand, supply short for 12-18 months.

The AI trade — which had been de-risking into this print — will get a bid. Truist explicitly calls for suppliers with high AI concentration to recover. Watch TTMI (Truist's easy derivative Buy) and the optical names. The 70% guide also puts a floor under memory: Nvidia just pre-committed $279B, of which memory is the bulk. HBM pricing pressure is NVDA's margin headwind and SK Hynix/Micron's tailwind.

Bottom line: The stock is still cheap against the guide, and the guide is a capacity constraint, not a demand survey. The real risk is execution — can they actually ship ~20GW in CY27 and convert that backlog into revenue without the margin structure fraying further? For a PM: this is the highest-conviction growth name in the index, and today's print gives you a reason to hold through the volatility. The marginal seller is someone who doesn't believe the guide. That's a bad bet to make against the street.

---

CRWD

Verdict: Proof point. NNARR +51% Y/Y — the biggest beat in years — and the agentic AI security trade just got its confirmation print. Stock sits at ~$219, 3.8% off the high, and NOT cheap at ~30x CY27E revenue. But momentum like this doesn't care about your multiple. Until it does.

CrowdStrike's FQ2 (July) delivered $333M net new ARR vs. a $284M FactSet bogey — $46M of air, the LARGEST ARR BEAT IN YEARS. Growth reaccelerated to 51% from 32% in FQ1, strongest since 2021 even stripping out July 19 effects. Total ARR $5.84B, +25.4%, accelerating for the 4th straight quarter. Revenue $1.47B, +25.8%, accelerating for the 5th. Rule of 51.6 with a 25.8% FCF margin. This is a platform company hitting on every cylinder.

STREET VIEW

Price target range is wide: $119 (Bernstein) to $260 (RBC). Strip the outlier and the street clusters $230-260, with the bulk at $250. A dozen-plus firms moved after the print — RBC $260, Needham/TD Cowen/Cantor/Scotiabank/Benchmark at $250, DA Davidson $245, Piper/Stifel/Jefferies $240. Baird is the constructive skeptic: Neutral, $230, but "incrementally constructive heading to Fal.Con."

Bernstein is the lone holdout — raised to $119 but that's still ~46% BELOW the tape. They're not arguing with the quarter; they're arguing with the durability. Their checks say demand hasn't fully inflected yet.

The collective bull thesis isn't "good quarter." It's a regime change. Piper calls it a "step-function change" in the demand environment. RBC says AI adoption creates "sustainable tailwinds that are increasing and accelerating the opportunity."

"CRWD delivered blowout FQ2 results w/ NNARR of $332.8M, well ahead of investor bogeys & accelerating to +51% Y/Y vs. +32% Y/Y NNARR growth in FQ1." — DA Davidson

BULL VS BEAR

Bull: Four consecutive quarters of ARR acceleration, five of revenue acceleration — that's a demand curve shift, not a one-print wonder. Agentic AI expanded the attack surface, and CRWD is the consolidator hoovering budget across endpoint, cloud, identity, SIEM. Falcon Flex at $2.29B (up ~100% YoY) is the multi-product flywheel. AIDR contributing early. Management raised FY27 NNARR growth guidance by 630bps to 34%; the +$63M ARR raise came ON TOP of a substantial FQ1 raise, and the implied 2H NNARR step-up (~$16M) says they see sustained acceleration, not a peak quarter.

Bear: Name the multiple — ~30x CY27E revenue for a stock already up 79% in a year. Even the bulls admit "room for misexecution is virtually nonexistent." That's the trap in perfection: any stumble gets priced like a catastrophe. Bernstein's checks say demand has NOT fully inflected yet and they question durability into 2027. If the post-Mythos modernization wave is pull-forward, the comps get brutal next year. At 30x, deceleration gets punished violently.

"The core endpoint offering is experiencing a renaissance as organizations modernize security to keep up with agentic artificial intelligence." — Needham

WHAT'S NEW

The magnitude, not just the direction. A $46M ARR beat at this scale is structural — the street's model is behind, not wrong for a quarter. And management's second consecutive raise (now $6.608B FY27 ARR midpoint, $63M above consensus) shows confidence is real.

The breadth is the sleeper. This isn't just SIEM/cloud carrying the load anymore. Endpoint is in a "renaissance," Falcon Complete and Exposure Management are contributing, AIDR has early traction. Post-Mythos modernization urgency is hitting the WHOLE platform, not one pillar. That's a consolidation story, and Falcon Flex is the mechanism.

One thing to flag: Baird — the holdout — getting "incrementally constructive" tells you the narrative is winning. When the skeptic starts creeping toward the door, that's momentum.

READ-THROUGH

This is the confirmation print for the AI security narrative. Bernstein frames the key question: is AI-related demand inflecting broadly across the sector, or concentrated among platform vendors? The off-quarter prints from AI security peers answer it. If CRWD is pulling ALL the budget, PANW and the point tools feel the squeeze. If it's a rising tide, the whole security complex rerates — and agentic AI becomes the strongest secular TMT story of 2026.

Positioning: CRWD is the momentum name in security, full stop. 30x on an accelerating consolidator with record pipeline is the price of admission in this tape. The risk: if the Mythos modernization wave is pull-forward, 2027 comps get ugly. Watch 2H NNARR prints like a hawk. The first "guidance in line" quarter is your de-risk signal.

---

VEEV

VEEV GAVE THE BULLS THEIR RECEIPT. Revenue, EBITDA, EPS all beat. Guide raised by MORE than the beat. Vault CRM scoreboard now reads 12 of top 20 pharma committed, two left, management publicly confident. And Falcon — the AI product the long thesis hinges on — drawing "stronger than expected" early interest. Stock is $294, ~5% off the $310.50 high. The question isn't whether the quarter was good. It was. It's whether you chase at 30x EV/FCF into a guided decel.

THE QUARTER AT A GLANCE

• Revenue $928M vs $905M consensus — beat by ~2.5%
• Adjusted EPS $2.35 vs $2.22 est
• Adjusted EBITDA $423M vs $403M est
• Normalized billings +19% YoY ($752M reported)
• Subscription revenue +16%, accelerating from 15% in Q1
• Services +24% — Vault CRM migrations + AI project work
• R&D segment +19% vs 17% consensus; Commercial +13%
• Non-GAAP op margin 44.8%; gross margin 75%

The billings and services numbers are the tell. Implementation revenue today = subscription stream tomorrow. That's why the street is raising, not trimming.

STREET VIEW

Seven PT raises in 24 hours. Range: $283 (TD Cowen) to $325 (RBC). Consensus lands right around $305 — call it fair value vs the current $294 tape. Buy/Overweight across the board, zero stragglers.

Collective thesis: Vault CRM is the cleanest share-gain story in software, R&D is inflecting above street expectations, and AI is additive here — not disruptive. Stifel models FY28 subscription growth at 14% vs street 12.5%. That's the kind of conviction that matters.

"Large pharmaceutical companies are actively consolidating single-point solutions, with Veeva positioned as a vendor with a comprehensive product portfolio and coherent AI strategy." — Stifel

"Veeva Systems remains a high-quality software company with relative insulation from AI disruption and is positioned to sustain Rule of 50 status." — Piper Sandler

BULL VS BEAR

Bull: The migration wave is countable and compounding. 12 of top 20 pharma committed to Vault CRM — three new wins this quarter, including two top-20 (Eli Lilly, Biogen) plus Regeneron just outside the top 20. 180+ customers live. Management raised guide MORE than the beat, flagged win-backs from Salesforce as an increased expectation, and a top-20 pharma just deployed Agentic Call Report to all US field reps. That's production AI, not a pilot. The 2030 targets now look achievable, not aspirational.

Bear: You're paying ~30x EV/FCF (with SBC) for a guide that implies Q4 subscription decel to ~12.5%. That's not Rule of 50 — that's a fade. The two remaining top-20 decisions are binary: one slip and "12 of 20" becomes a delayed story. Falcon monetization? "Too early to gauge" (their words, via Stifel). And Aspen CRM is priced at HALF of Salesforce — nice flanking move into horizontal CRM, but it's a price-disruption play, not the high-margin Vault math. The set-up is good. The setup is also crowded.

WHAT'S NEW / INCREMENTAL

• Two remaining top-20 vault CRM decisions — management is publicly confident on both. New swagger.
• Falcon early demand "stronger than expected"; FIVE early adopters go live this fiscal year.
• Win-back from Salesforce expectations raised by management.
• 180+ Vault CRM go-lives (prior disclosed counts were lower).
• Aspen CRM unveiled — horizontal product at half the cost of the competitor.
• Guidance raise exceeded the quarter's beat — the tell that management believes the momentum.

What's NOT new: Veeva's R&D strength, Crossix stability, and the CRM migration backlog were already the long story. The quarter just gave the market the numbers to trust it.

READ-THROUGH

VEEV is the "AI disruption is overblown for vertical SaaS" trade — and the print is direct evidence. Life sciences demands data integrity, regulatory wrappers, and workflow depth. The LLM natives don't have those. Read that across other workflow-deep vertical software (TYL, GWRE, PCOR): the winners have proprietary data and installed workflow, not the flashiest model. The caution flag is the valuation and the crowded positioning — if Q4 subscription lands at 12.5%, the multiple resets before the story catches back up. Watch the two remaining top-20 decisions like a hawk. That's the next catalyst.

---

NTNX

Best kind of problem for a PM: fundamentals are accelerating, story is getting better, but the stock's already ripped 71% in six months — so the r/r is now a debate, not a no-brainer. NET NEW ARR PRINTED $114M QoQ vs a ~$79M four-quarter average — a 44% step-up — and it happened DESPITE hardware constraints pushing deals to future start dates. The ARR acceleration is real, the external storage TAM is just getting started, and the street spent the morning raising PTs into strength.

Street view: PTs now span $65–$90. Bulls cluster $75–$90 — RBC $90, Needham $85, BofA $78, Piper $75 — with Wells Fargo the lone hold at $65, now below spot at $70.34. Consensus average lands ~$79. The debate has shifted from "is the story intact" to "how much of FY27 is already in the price."

THE QUARTER AT A GLANCE

Q4 FY26 (July quarter): revenue $757.1M vs $738.3M consensus, +16% YoY. Non-GAAP EPS $0.60 vs $0.49 est — a 22% beat. ARR hit $2.549B, +16% YoY, with sequential growth ACCELERATING. FY26 revenue $2.75B, +13%, at an 87% gross margin.

"Net new ARR increased $114 million quarter-over-quarter compared to an average of $79 million in the last four quarters, despite ongoing hardware constraints and elevated pricing delaying revenue conversion." — Wells Fargo

The $114M net new ARR is the tell. That step-up came in a quarter where supply constraints pushed revenue conversion out. Renewal pool is decelerating — management owns it — so the FY27 11-13% revenue growth guide puts the burden on net new. That's a healthy growth mix shift if it holds.

THE NEW STUFF

External storage is the FY27 growth driver, and it's inflecting now. NetApp GA before year-end, Lenovo support live, Dell and Everpure already on board. Seven-figure ACV storage deals, strong QoQ bookings growth. Needham called it the largest incremental driver for the year.

"External Storage is the largest incremental growth driver for the upcoming year. NetApp is expected to become generally available before the end of the year." — Needham

Two other incremental signals: (1) large government contract wins flowing into the future-start-date backlog; (2) Nutanix Cloud Clusters picking up share precisely BECAUSE customers can't get on-prem boxes. Hardware scarcity is pushing workloads to the cloud — a thematic tell worth remembering.

FCF: initial margin ~28%, but supply-chain and customer payment concessions hit near-term FCF guidance. Piper sees FCF inflecting higher through the year — a timing and mix dynamic, not a go-to-market breakdown.

BULL VS BEAR

Bull: Rate of change is up and to the right. ARR acceleration, net new ARR +44% vs the trailing average, fresh storage TAM, VMware displacement still running, NC2 pull-forward from hardware chaos. The backlog is growing — future-start-date orders de-risk FY27 conversion instead of signaling weakness. At 87% GM and ~28% FCF margin, this is a high-quality compounder at an inflection. RBC's $90 target implies another 30%.

Bear: The easy money's gone. Roughly 6x forward revenue for a business guiding 11-13% growth. Wells Fargo at $65 is below spot — that's a colleague telling you the price already pays for the good news. Future-start-date orders cut both ways: if hardware stays tight, revenue keeps sliding and the guide becomes the ceiling. Renewals decelerating means net-new carries the load, and the 71% run has made this a crowded momentum long. Any merely in-line quarter is a de-rating event.

READ-THROUGH

• NetApp (NTAP): Partner/competitor tension. Nutanix supporting NTAP storage validates the software-unbundling trend, but also pushes NTNX deeper into storage TAM. Watch the channel reaction once NetApp goes GA.
• Pure Storage (PSTG): Direct headwind if external storage bookings are as real as the ARR data suggests. Software-defined storage is coming for legacy arrays.
• Broadcom (AVGO): VMware displacement trade keeps paying. Every renewal cycle is another battleground — NTNX's ARR data says it's accelerating, not fading.
• Hardware supply chain: NTNX elongating hardware lead times is a read-across to the whole server/storage complex. Demand is pushed out, not cancelled — but near-term recognized revenue gets compressed everywhere.
• AI/Kubernetes: The longer-duration vector. K8s-native workloads at the edge keep the multiple alive even if a quarter disappoints.

Positioning: this went from crowded to MORE crowded after the print. New money should wait for the NetApp GA catalyst or the first FY27 print to confirm backlog conversion — paying 6x forward for 11-13% growth is paying up for flawless execution. That said, I'd be a buyer on any supply-chain scare that knocks this back to the mid-$60s. The ARR step-up is a fundamental change, not a print artifact.

---

META

$18B settlement, and the Street exhales. KeyBanc and UBS both reiterate buys this morning — the message is identical: the regulatory overhang is gone, the fine is a rounding error, and the AI story is back in the driver's seat. The downside was always the headline risk, not the balance sheet. Print the release, buy the stock.

SETTLEMENT MATH

THE HEADLINE IS $18B — $12.7B guaranteed to 51 states/territories over 10 years, plus $5B contingent on YouTube and TikTok signing similar agreements. UBS makes the killer comparison: the market had flirted with a Big Tobacco-style outcome at $200B. This is 9% of that. And KeyBanc slaps the revenue bogey on it: teens are SUB 1% OF REVENUE. Immaterial.

The product changes matter more than the check. Teen and underage account modifications could've been structural — something that breaks targeted ads, the actual cash register. Both firms read the changes as benign. KeyBanc's twist: Meta just wrote an industry-wide framework that YouTube and TikTok now have to answer to. META turns its own regulatory headache into a competitor's liability. That's not a bad trade.

THE TRADE

This was the excuse to stay underweight. It's gone now. Both analysts pivot to the same upside levers: Business Agents monetization, external compute capacity sales, consumer AI optionality. UBS thinks EPS beats consensus on multiple avenues. KeyBanc frames it as "focus back to the AI product cycle."

PT cluster is 715–886, with Rosenblatt the hammer at $886 calling the settlement "a significant win." BofA at $810, KeyBanc $780, Truist trimmed to $763 (still Buy), UBS $715. No sells. The r/r skews hard to the upside from $578.

Bull case: overhang removed, AI cycle accelerating, and the settlement forces competitors to carry the same baggage on their P&L.

Bear case: Needham's Hold — Laura Martin calls it "costly strategy diffusion." Meta's spraying capital across too many ventures. Fair enough, but with 81.75% gross margins and 27.65% revenue growth, they've got the ammo to spray.

The line of the morning, from UBS:

"The fine amount represents a less punitive outcome compared to the often-cited Big Tobacco analogy involving $200 billion in fines."

That's the whole ballgame. The market priced a worst case that didn't materialize. You buy into that setup every time.

---

SNPS

Beat-and-raise, stock fades after hours, and Benchmark is having none of it. Reaffirms Buy and $570 PT after Q3 FY26 — NON-GAAP EPS $3.91 ON REVENUE $2.48B, AHEAD OF $3.67/$2.44B CONSENSUS, with full-year guidance raised too. The kicker: shares traded at 24x forward earnings into the print, THE LOWEST MULTIPLE IN YEARS for a business compounding 39.5% LTM revenue growth with 82.6% gross margins.

Baird chased the same trade — upgraded Neutral→Outperform, PT $560, calling for a return to double-digit organic revenue growth by FY27. The bull case is simple: EDA is an AI beneficiary, not a casualty. Benchmark lifted near-term estimates "materially" off the quarter's momentum.

"AI is serving as a tailwind to EDA and simulation software rather than a headwind."

Bear case isn't the quarter — it's the narrative. The September 30 analyst day is the real catalyst. That's where management has to bury the AI-disruption fear, and Benchmark thinks the market's already overpaying for that risk at these valuations. Aftermarket dip post-print looks like a gift, not a warning.

---

MRVL

THE DIVIDE

Goldman is the parking brake on a runaway car. Reiterates Neutral with a $195 PT — stock trades $241.42, UP 228% IN A YEAR. They see rangebound chop ahead: results in line, guidance only modestly ahead, expectations already feasted on peer prints and hyperscaler spend. InvestingPro flags it as overvalued vs Fair Value. (The only Neutral in a sea of Buy ratings.)

The bull camp is not subtle. Rosenblatt's at $300. Benchmark $275. Susquehanna $265. BMO and Oppenheimer both $250. All anchored on AI networking, optical interconnects, and the expanded Google TPU partnership. All effectively saying the custom compute ramp is the whole ballgame and it's going to print.

"Investor expectations were elevated heading into the quarter based on peers' reports and robust spending at key customers."

The swing factor is 2H custom compute ramp magnitude and whether management raises CY27/28 Datacenter revenue targets on the call. Microsoft ASIC ramp shape also in the crosshairs. GS says upside hinges on those datacenter numbers — bulls assume they move up. Rangebound vs rerating. Earnings settles it.

---

SMTC

Northland finally capitulates — upgrade to Outperform with a $182 PT, the low end of the recent print cluster ($188 Stifel, $230 UBS/Benchmark). This isn't discovery; it's validation. The stock's already up 142% in a year, 56% in six months.

Thesis is straightforward: connectivity is the AI bottleneck on both power and performance, and SMTC's taking share in copper AND optical. Data center revenue sits at 30% of the mix, Northland sees 44% by end-2027, and the cellular module divestiture + LoRa mix pushes gross margin into the mid-60s. Q2 FY27 revenue beat by 4.2%, Q3 guide crushed by 13.6% — beat-and-raise engine is humming.

"Connectivity is currently the AI bottleneck in terms of power and performance."

The PT spread tells the real story: $182 to $230 is a wide variance for a bull case this crowded. InvestingPro flags the stock as overvalued vs. fair value, and at $140.80 Northland's PT implies just ~29% upside. The easy money got made six months ago. We'd need a digestion pullback or a LoRa blowout quarter before paying up here.

---

GOOGL

WOLFE SAYS THE STREET IS STILL UNDERESTIMATING CLOUD

Wolfe is making the 2027 call now, not waiting for the calendar to flip. Shweta Khajuria reiterated Outperform with a $460 PT, lifting 2027 revenue 10% to $595B and EPS 6% to $15.89. The crux of the thesis: her Q3 GCP growth print sits at 125% YoY vs Street consensus of 87%. That's a 38-point gap — and that gap IS the trade.

"The firm stated it does not believe it is too early to make a 2027 stock pick."

At 17.14x forward with that growth trajectory, the market still prices Alphabet like a slow compounder. 65% return over the past year and the multiple hasn't re-rated materially — that's the tell. If GCP prints anywhere near Wolfe's number, consensus revisions cascade and the stock gets a second leg.

Still, steelman the bear case: 125% YoY GCP growth is a bold bogey. If the quarter comes in closer to the 87% Street number, the momentum narrative takes a hit. The product flow helps — Gemini Enterprise flexible billing, 3.5 Transcribe at 4.0% WER, Live productivity features — but that's sticky AI narrative, not the swing factor.

Citizens also weighed in, reiterated Market Outperform with a $515 PT on cloud/AI model growth — so the sell-side bull camp is coherent, not lonely.

Verdict: Wolfe's GCP number is the outlier to respect. Longs get upside case on paper. Shorts are betting on a Q3 miss vs an estimate that's about to get revised up. I know which side has the better r/r.

---

BRZE

TD Cowen sees a cleaner setup into Sept 8. Raising PT to $36 from $30, Buy maintained. Stock's already up 25%+ off the Q1 print and 62% over six months — the bar is higher, but read-throughs from Twilio and Klaviyo suggest momentum is real. AI adoption ramping, enterprise wins in the bag, and new AI product monetization lands next year.

Street's getting crowded. BTIG at $35. Goldman initiated at $34. TD Cowen's the high print at $36 — modest upside from here, so the r/r is really about whether the print clears a now-elevated bar.

"We think there were favorable developments intra-qtr which should help BRZE get back onto a cleaner qtr in terms of rev beat & GM stability vs. Street ests."

One wrinkle worth flagging: DA Davidson notes the top-line raise to 22% YoY (vs 20% consensus) came with Subscription slightly below estimates — Professional Services did the heavy lifting, beating by ~$6M. That's the dirty part of the "clean quarter" narrative. Sept 8 tells us whether subscription re-accelerates or BRZE gets saved by services again.

---

BE

Bernstein holds at Market Perform / $282 PT, but don't read that as bearish — it's "real demand, real technology, wrong price." Stock at $218.21, UP 151% YTD AND 329% OVER THE PAST YEAR. The entire debate compresses into one ratio: Bernstein targets 41x EV/EBITDA on blended 2H27/1H28 Adjusted EBITDA, while Bloom currently trades at 153.88x. That's not a discount, that's a full-priced membership to the islanded-power club.

The bull case is straightforward: data center interconnection queues are a nightmare, moratoriums are spreading, and Bloom's fuel cells plug in TODAY without waiting on the grid. Q2 proved execution — EBITDA $253M vs $152M consensus, revenue $1.07B +166% YoY. But Bernstein frames the investment question as four separate bogeys: total GW of data center demand this decade, how much goes behind-the-meter, Bloom's share capture, and whether the current valuation already prices all of it.

"Data center development moratoriums and delays in obtaining large-load grid connections continue to support demand for islanded power solutions."

The analyst tape post-print is a split decision: Mizuho to Outperform, Clear Street to Buy on the pullback, Truist cutting PT to $218 / Hold on demand worries, Oppenheimer staying Perform. Everyone sees the same thing — the disagreement is what multiple you pay for it. At 153x, you're paying for a perfect 10 on all four questions. Bernstein isn't sure the market's math works. Neither am I.

---

P

BofA steps up hard on Everpure. Upgrade to Buy from Neutral, PT ripped to $150 from $90 — 38% upside from $108.90, and the stock's already returned 79% over the past year. The message: the DirectFlash hyperscaler story is mid-innings, not late.

The thesis clusters around two vectors. First, architecture validation — a SECOND top-five hyperscaler win confirms DirectFlash, with meaningful revenue starting FY2028. Second, NAND price stabilization displaces capacity-based storage tiers, and Everpure's density/TCO advantage takes share in an inflationary environment. Q2 backs it up: REVENUE +38% Y/Y TO $1.186B, OPERATING PROFIT +77% TO $230M, MARGIN 19.4% — ~120BPS AHEAD OF IMPLIED GUIDANCE. FY27 revenue guide raised to $5.0-5.7B. Eight consecutive quarters of 30%+ growth. Revenue acceleration feeds directly into operating leverage.

"Everpure will likely see some displacement of capacity-based storage tiers as NAND prices stabilize, while the company could take market share in an inflationary environment given better total cost of ownership and density provided by its DirectFlash modules."

Watch the September 23 analyst day for hyperscaler ramp detail. Evercore holds Outperform at $130 — less aggressive than BofA, but no bear camp forming. Main risk: the second hyperscaler's "meaningful revenue" doesn't hit until FY2028. That's a long carry, so the stock lives or dies on continued estimate revisions between now and then. Today, the rate of change is still your friend.

---

VTEX

Jefferies cuts PT to $5.10 from $6.50 — keeps Buy, but the thesis just got a lot uglier. The headline number isn't the target, it's the assumption behind it: ZERO GROWTH in core B2C over the next five years. That's not a trim, that's a reset.

CFO Ricardo Sodré says headwinds are macro- and mix-driven and low-20s operating margins are sustainable. Fine. But Itau BBA just downgraded to Market Perform (PT $4.50 from $5.00) on Brazil/Argentina macro and heightened competition. Two firms, same conclusion: the top-line recovery keeps getting pushed out.

Q2 was the usual mixed bag — EPS beat ($0.05 vs $0.04 est) but revenue missed ($64.38M vs $64.53M). Margins are the story: FCF +79%, non-GAAP op income +62% YoY. Profitability is real.

"Jefferies' revised estimates assume no growth in the core B2C business over the next five years."

The bull case lives in B2B, international, AI, and advertising becoming increasingly relevant drivers. The bear case: core stalls, management blames macro, and the market sits at $3.57 waiting for proof. Jefferies sees ~43% upside to the new target — but with a no-growth core assumption, you're paying for optionality, not fundamentals.

---

INTU

THE Q4 BEAT IS NOISE. INTU printed $4.354B (+14%, beat by 2.1%) and beat on adjusted EPS — but FY27 guidance blew up the story and the stock's down 47% YTD at $345.60. Freedom Broker cut its PT to $370 from $430 (Hold) on muted FY27 outlook. Six analysts already revised estimates down for the coming period. That tells you everything about the setup.

The damage is self-inflicted: TurboTax DIY price cuts (a strategic pivot toward lower-end customers — the thing that got Wolfe to downgrade), Mailchimp stagnation, and a new stock-comp accounting methodology muddying the optics. Offsets are real but narrow: steady mid-market cloud demand and Credit Karma, which surprised to the upside at +16.5% YoY. Street reaction is a mess — Truist to $300 (Hold), Piper and Stifel to $290–300 on reset expectations, BMO still Outperform. Nobody's fighting to own it.

Freedom Broker's framing: "the transitional nature of fiscal 2027 and temporary pressure on client monetization growth."

That's the bull case, and it's honest. But transition years get repriced as permanent impairment until proven otherwise. At 21x earnings with 81% gross margins, quality is intact — PMs just don't get paid to catch falling knives on deceleration. Wait for guidance to stabilize.

---

OOMA

THE PRINT

Freedom Broker lifts PT to $24 from $20, Buy stays. Stock already at $23.04 — so the new bogey is ~4% above spot. This is "we're right, here's the updated math," not a new-multiple call.

Q2 FY27 beat across the board: REVENUE $83.2M vs $81.72M, ADJ EPS $0.35 vs $0.33, and RECORD ADJ EBITDA $27.4M. Revenue growth ACCELERATED TO 17% LTM. Business segment doing the heavy lifting — AirDial ripping on the copper POTS replacement cycle, FluentStream and Phone.com (both late-2025 acquisitions) layering in cleanly.

The real story is MyPhone. It REVERSED the long-standing residential subscriber decline. Management embedded residential segment growth in FY27 guidance for the first time. That's structural, not a one-quarter beat.

"Results were driven by accelerating growth in the business segment and the successful product launches across both business and residential divisions."

Mkt cap $619M. Up 75% YTD. Ooma AI monetization is only just hitting estimates — that's the next leg. Not great r/r from $23 with 4% to the PT, but this is a real narrative shift, not a dead-cat bounce.

---

SEDG

JEFFERIES TAKES PT TO $31 FROM $45 (HOLD) — STOCK TRADES $33.37, ABOVE THE NEW TARGET. That's the setup in one line: even the Hold-rated analyst thinks you're paying up, if only marginally.

2026 is the trough year. Jefferies matches consensus for 2026 but goes below Street for 2027-2028 — they don't buy the recovery slope. Resi softness plus slower margin recovery means guidance comes in light. C&I improving, but that's not the earnings driver. GM at 21% is horrid for the sector, and 12 analysts have marked estimates down recently.

The one interesting wrinkle: SST valued at ~$5/sh in SOTP on 2% market share. Don't hold your breath for a big disclosure there.

"An unappealing short position but not a compelling long investment."

That line nails the whole trade. Q2 beat (EPS $0.05 vs -$0.02 est, rev $346.2M vs $342M) and the stock just sits here — because guidance is soft. UBS plays the bull: Buy, $42, on the FCC inverter import ban as a tailwind for SEDG. Mizuho slashed to $38 from $71. The PT dispersion ($31 to $42) is an admission: nobody has an edge on the US resi recovery timeline.

Verdict: pass. No reason to own into a guide-down, no edge shorting at these levels.

---

IRM

TRUIST PUSHES IRM PT TO $155 FROM $140 — KEEPS IT A BUY AND A "BEST IDEA FOR 2026." They see a real shot at 13% revenue growth in 2027, versus the 9% consensus. That gap is the whole trade. The drivers: sustained Asset Lifecycle Management strength and improving Records Storage pricing.

The stock is already up 48.6% YTD — not a secret. But Truist's math says there's still ~25% upside on an 18x EV/2028 EBITDA multiple, in line with peers at 17-19x. They're not asking you to pay a premium. They're asking you to believe the growth is durable. Q2 helps: record $2.03B revenue, +19% y/y, beat, and they raised full-year guidance.

Truist increased its own 2027 revenue growth estimate to 10% year-over-year, incorporating some conservatism while noting that 13% or higher growth remains achievable.

That's the tell — even their "conservative" number is above consensus. The bear case? This is a data center + physical records story that's already been repriced. The easy money is gone. But for a multi-manager with a 12-month lens, the r/r still skews positively if ALM momentum holds. Keep it on the screen.

---

WK

Stifel's PT raise ($70 → $95, Buy maintained) is the headline, but the real signal is buried in the fireside chat: NEW-CUSTOMER ACTIVITY HIT A MULTI-YEAR HIGH IN Q2. That's the rate-of-change data point that matters here. It suggests the growth story is broadening beyond compliance-driven renewals — and Stifel clearly bought it.

The thesis stacks three accelerants: a growing AI tool portfolio driving cross-sell/up-sell, improving Capital Markets demand, and an efficiency push finally showing up in profitability. Revenue ran ~20% LTM to $965.7M at 80% gross margins. Q2 beat clean ($0.77 vs $0.64 est; $255.29M vs $251.15M), above the high end of guidance. Execution is not the question.

One caveat: shares sold off after Q2 despite the beat. The market's forward-looking, and guidance wasn't enough to spark a rally. Stifel expects only "modest" financial updates at the upcoming user conference/analyst day — analyst-speak for "don't expect fireworks."

"New-customer activity reached a multi-year high in the second quarter."

That's the line. If it's real, the Q2 beat was less about the quarter and more about the setup — new logos convert to recurring revenue with a lag, and that bodes well for next year's numbers. Stifel's $95 PT implies ~35% upside and a bet that the new-customer surge begins a re-rating, not a one-quarter blip. Risk: 19% growth isn't cheap if Capital Markets demand stalls again. Bull case wins for now — but the after-hours guidance reaction says the market needs proof, not fireside chat confidence.

---

EXLS

TD Cowen bumps EXLS to $45 from $43, keeps Buy, and slaps a "top services pick" label on it after sitting down with the CFO and IR. The message was consistent: demand is holding, growth is sustainable, and the AI/data narrative is broad-based — not a one-off quarter. This is a call on momentum AND duration, not just valuation.

The fundamental setup is genuinely good. Q2 was a clean beat — $0.59 EPS vs $0.55 est, $594.8M revenue vs $574.03M est — and management raised full-year guidance. Organic growth hit 16%, the strongest in years. Baird already came along with an upgrade to Outperform and a $45 target, so you've got two firms independently landing on the same number. The bull case writes itself: EXLS is scaling AI/IP faster than the IT services complex, and the market is starting to pay attention.

But TD Cowen is honest about the near-term driver, and so should we be:

"Shares may largely move with AI fund flows, but the company's fundamental outperformance versus peers and IT services is becoming more appreciated as it benefits from scaling AI and intellectual property."

That's the real trading reality. At $37.80 and a $5.8B cap, the stock is up 20% in six months — the fundamental floor is solid, but the ceiling is AI-beta and flows. If the AI trade rotates, EXLS moves with it, fundamentals be damned. Right now, that cuts both ways: more flow-driven upside if the tape holds, but no insulation if AI sentiment cracks. The outperformance-versus-peers story is real and getting more recognized — just don't confuse a good company with a stock that trades on its own fundamentals. It doesn't. Not yet.

---

WAY

BMO re-ups Outperform at $30 after the investor day, and the hook is AI — not as a feature, but as THE TAM DOUBLE. Management's LLM-agnostic stack lets them slot the right model to each use case, and the tell is clients coming to Waystar first to talk AI strategy. That's the ballgame in revenue cycle management: incumbency as the wedge.

The numbers back the narrative — 19% REVENUE GROWTH, 69% GROSS MARGINS, and a raised FY guide after Q2 (rev +18% to $320M, adj EBITDA +21.5% to $137M). Management also reaffirmed the long-term growth and margin targets, so no guide-down scare lurking.

"Existing clients are approaching Waystar first to discuss AI strategies."

Street consensus sits at Strong Buy with targets spanning $27-$44 — Mizuho owns the high end at $42 post-investor day, Truist trimmed to $33 but stays Buy. BMO at $30 is the low end, but the AI TAM-double thesis is the reason to own it. 3p: AI monetization is already commercial, so watch the next couple prints for that revenue line inflection — that's the rate-of-change tell.

---

1. Supplementary Coverage

WDAY

Beat-driven normalization trade, not a revisions story. Q2 SUB REV $2.47B (+13% CC), PF OM 31.1% — headline fine — but cRPO missed and FY guide just meets the street at ~11% growth. AI is only 6% of revenue, so the real story is Elliott's $2B stake, 9% share-count reduction in three quarters, and 200bps of margin expansion ahead. Founders' voting control makes M&A the swing factor — selling below $250 looks insulting.

SNOW

150X+ FWD EARNINGS IS MADNESS after a 100% rally off lows. LLMs that handle complex workflows better could arrive in months and re-ignite disruption fears; network effects and domain knowledge are real but already priced. Software went from dirt cheap to uninvestable real quick.

INTC

The foundry thesis is being engineered from Washington. Commerce is embedded at Intel getting regular process and yield updates; the tariff regime ties relief to US manufacturing investment, and TSMC will never hit the 30% US advanced-capacity target this decade. Process momentum is real — 18A YIELDS ABOVE INTERNAL TARGETS, 14A DEFECT DENSITY BEST SINCE 22NM, EMIB passed Google's TPU qualifications (targets: 40% GM / 30% OM in advanced packaging). Server CPU demand exceeds supply through 2027; the IDM structure gives price-cut flexibility fabless rivals lack.

MU

Memory is in an extreme pricing upcycle. NVDA's CFO flagged extreme pricing headed higher next year — HBM contract prices potentially up 70-140% — and NVDA SUPPLY COMMITMENTS JUMPED $119B→$279B, mostly memory. MU at ~5X EARNINGS on the same horizon where software trades at 150x; the peak-memory-cycle call is misframed because HBM is structurally different from commodity DRAM. Watch NVHBM: NVDA wants to standardize HBM while vendors want differentiation — custom architectures decide who captures pricing.

SNDK

NAND is the most commodity-like part of memory, but HBF is designed to escape that. At ~5X FWD EARNINGS vs software at 150x, this is a contrarian long with less HBM leverage but the same scarcity tailwind — NVDA flags memory shortages into early 2028 and NAND de-specing creates stored demand. Precommitments are primarily securing HBM allocation; the spillover tightens NAND. Enterprise SSD pricing is the SNDK-specific variable.

WDC

Same scarcity tailwind as SNDK with more legacy storage leverage. Memory shortages into early 2028, HBM spilling into enterprise SSDs, NAND de-specing creates a stored-demand floor. At ~5X EARNINGS, the whole complex trades absurdly cheap against software at 150x. If you own SNDK, WDC is the alternate tape — less HBM purity, more storage exposure.

TSM

The tariff regime targets TSMC's US manufacturing shortfall. Taiwan still produces 90%+ of the most advanced chips; Arizona at full capacity is only ~30% of US advanced capacity. Comprehensive semiconductor tariffs under consideration — servers, laptops, datacenter gear — with relief tied to US factory investment. Existential policy overhang on pricing and capacity allocation; Morningstar still says undervalued after five years, but policy risk is exactly why the discount exists.

AVGO

The AI credit chain is tightening. AVGO's $100B debt issuance became the poster child for AI financing stress; credit investors are now scrutinizing AI capex and circular financing. Thoma Bravo had to accept ~40 creditor-friendly terms on AI disruption worries — same dynamic hits leveraged tech. AVGO is the equity expression of AI infrastructure debt risk; watch credit spreads, not just the AI backlog.

AMD

ATOM Inference Engine is a textbook resource misallocation — 20+ FULL-TIME ENGINEERS, including principal engineers and AMD Fellows, stuck on a side project with basically no customers; vLLM and SGLang are what Meta and xAI actually use. Managers cite Alibaba/Qwen but it's a side-business deployment, and engineers report being blocked from raising concerns internally. ROCm 10 is a genuine milestone with HyperLoom auto-optimizations, but org dysfunction undermines the CUDA response. Middle-manager fiefdom risk, not a chip problem.

QCOM

HBC could capture up to 20% of the data center accelerator market. So much hate in the tape that when HBC hits, it hits like a sledgehammer. Samsung and SK hynix are backing HBC alongside HBM. Conflating HBC with edge and smartphone DRAM stacking muddies the datacenter message, but sentiment is so negative the surprise potential is large.

AAPL

SEPT 9 EVENT is the biggest iPhone shake-up in years. First foldable joins the iPhone 18 Pro lineup, reportedly powered by the 2NM A20 Pro — Ternus's first major launch as CEO. The 2nm transition is the lead monitor for N2 capacity allocation and pricing. Watch supply constraints, not just consumer demand.

LITE

Went public attacking Coherent's 200G EML yield on 6-inch INP at a Deutsche Bank conference. Speed and power of InP devices matter, and LITE is telling the market COHR's yield is dogshit. That's a share-shift signal in high-speed EMLs — LITE believes it's winning the next-gen optics socket. InP yield is the choke point at 200G.

COHR

LITE publicly attacked your 200G EML yield on 6-inch InP. If the yield claim is true, share loss in high-speed EMLs accelerates. Speed and power performance matter and 200G is the choke point. Coherent needs to defend spec, not just talk backlog — a competitive attack at a conference is a bearish tell.

AMZN

AWS committed to 2M INCREMENTAL NVDA GPUS from FY2Q27-FY2Q29. Largest disclosed single-customer volume commitment of the cycle — Blackwell Ultra through Rubin Ultra plus Vera CPU. Strongest hyperscaler confirmation of NVDA's FY28 guide. AWS is locking supply, not debating demand; read-through positive for AI infrastructure and power.

ASML

EUV becomes the end-of-decade bottleneck. The industry figured out how to turn $1 of fab capex into $100 of token revenue, which implies massive pricing power for ASML. NVDA's FY28 guide of ~70% growth and SEMI DEMAND RUNNING 115-120% ABOVE CAPACITY means WFE capex heads up; the debate is whether EUV output can scale in time. ASML is the least substitutable exposure in the whole AI supply chain.

LRCX

Etcher demand is inflecting. LRCX will sell a huge number of etchers in the next 12 months on memory and advanced-packaging capex; NVDA's supply gap forces capacity adds. Kioxia's ¥1T NAND fab and HBM capacity all require etch steps. The 12-month order book is the setup — watch memory capex pace, not just AI logic.

AMAT

Semicap complex gets an NVDA supply-closure boost. Unconstrained demand ~100% versus supply ~70% forces capacity adds across logic, HBM, and packaging. AMAT is broad WFE exposure with memory and advanced-packaging leverage. Market debates 2027 order durability; the guide pushes at least one more year of visibility — rate of change positive.

KLAC

Process control intensity rises with complex HBM and advanced packaging stacks. NVDA's supply gap forces capacity adds; KLAC is the inspection toll booth. HYPERSCALER CAPEX EXPECTED TO RISE FROM $800-860B IN 2026 TO $1.1-1.3T IN 2027. Rate of change positive and semicap names screen at teen multiples.

AMKR

Intel EMIB packaging is a credible TSMC rival, and AMKR is the OSAT read-through. First Google TPU built with Intel EMIB hit yield, cost, and capacity targets; Intel targets 40% GM and 30% OM in advanced packaging. CoWoS capacity is the 2027 physical pinch point. Packaging, not lithography, is the bottleneck — AMKR is a leveraged packaging play.

IREN

Q4 REVENUE $137.2M VERSUS $132.3M EXPECTED; FY26 REVENUE $707M; AI CLOUD REVENUE UP ~8X TO $128.8M. Contracted ARR at $4B for 2026 capacity, $1B ARR operating today. Stock still fell 3% after hours — the bar is high. Market wants execution, not backlog; legacy energized capacity gives IREN the best cost of capital in the neocloud cohort.

NBIS

Signed at 3-4X its installed-base unit price within 18 months. ARR per GW spans $8.3B to $50B across the neocloud cohort; NBIS runs on customer prepayments. From 26Q4 expect stabilization with modest increases, then rents down 20-30% by generation as supply lands. Pricing power peaking now — own the transition, not the plateau.

CRWV

CRWV converts power delivery risk into refinancing risk. Buys speed with 2X POWER COST and permitting risk on the shortest contracts in the neocloud cohort. AI credit markets are tightening — AVGO's $100B debt, Thoma Bravo giving 40 creditor-friendly terms, NVDA pausing revenue-share deals. This is a credit story wearing a power story mask.

FAST

Fastenal keeps compounding under Dan Florness, who followed a Buffett-endorsed predecessor. Honest, transparent communication and incredible execution. Industrial distribution is tangential to AI capex but a quality tell on the US capex cycle. Quiet compounder, not a momentum story.

CSCO

Cisco gave 90,000 EMPLOYEES personalized AI agents backed by 800+ SUBAGENTS. Cost routing: 50-60% to open-weight models, 20-30% to software automation, small remainder to frontier models. Built on Azure OpenAI, Claude, Gemini, and Cisco's Deep Network Model. Real enterprise agent deployment validating open-weight economics — not a product revenue signal, but a demand tell for enterprise AI infrastructure.

NFLX

Netflix deployed LLM-as-a-judge for recommendation explanations at scale. One model writes because-you-watched lines, another scores every one before surfacing. A 5-WEEK TEST shifted members toward unwatched titles and more often ended browse by pressing play. Workflow template for AI in consumer engagement — improves activation and retention, not necessarily content cost.

SIVEF

Down 16% as Q2 revenue fell 12% y/y and adjusted EBITDA loss widened to SEK -35.5M. Deliberately winding down NRE work to prep customer production ramps; product revenue rose 18% and pipeline is $1.2B. The story is 2027 ramps and the market priced the gap today. Bear case is fair value zero because newcomers underestimate scale and track record — binary execution stock.

MRNA

Priced $2B CONVERTIBLE SENIOR NOTES and shares fell 8%. Capital raise into weakness — an equity-linked overhang for a biotech with lumpy COVID revenue. Market treats it as dilution risk. Risk-off tell for high-beta growth financing, even outside tech.

GEV

Gas turbine slots are sold out through 2031 on a 4-5 year order-to-COD cycle. If GEV takes more price than ENR, gas OE is over-earning and deserves a peak multiple; if not, the bear case collapses. Scarcity is real: 2027 NA AI DC DEMAND ~35GW IT VERSUS 16.5-23.4GW DELIVERABLE. GEV is the liquid way to price that scarcity.

ESTC

Cleanest AI revenue attach in software. 37% of $100K+ customers using AI features versus 21% a year ago — adoption doubling in the installed base. Market rewards names with AI revenue evidence and punishes names without. Fundamental re-rating candidate if attach keeps converting to spend.

RBRK

Raised full-year guide and still fell 8%. Market paid for AI-driven revenue quality, not guide raises; profit guidance weakness is the offset. First post-rally security tape telling you narrative is not enough. Needs margin discipline to hold the AI-security premium.

S

SentinelOne fell on weak profit guidance despite the AI security surge. Market is bifurcating security into AI-revenue winners and AI-narrative losers — S is in the latter for now. Expect multiple compression until profitability trajectory improves. Mirror image of CRWD's best day.

ADSK

Beat but cut FCF guidance midpoint and dropped 5%. Market is ignoring headline beats and valuing free cash flow quality. Warning for software names with AI narrative but no cash conversion. ADSK needs FCF revision upside, not just billings growth.

BABA

Raised HK$8B VIA 710M NEW SHARES AT HK$112.70, AN 8.4% DISCOUNT; stock fell 8.5%. Morgan Stanley says the sell-off was disproportionate — roughly $19B of shareholder repricing versus the $1.03B raise. FY27E capex maintained at RMB200B; COMPUTE PROJECT IRR 22% VERSUS ~10% COST OF CAPITAL. If 80%+ of proceeds go to infrastructure, EPS is accretive from year 2; MS fair value HK$205, stays overweight into September.

TCEHY

AI funding path is self-funded — operating cash flow, bonds, and an RMB87.5B portfolio. No equity raise needed; Morgan Stanley stays overweight. Key risks: AI deployed capital ROIC below 15%, infrastructure above 40% of AI spend, compute pricing falling on supply loosening. Watch whether it follows Alibaba into equity raises.

BIDU

AI funding path is borrowing plus Kunlun chip subsidiary equity — less balance-sheet depth than Alibaba or Tencent. Morgan Stanley stays overweight, but financing constraints could force more external capital. Kunlun equity monetization is the swing factor. Weaker funding position in the Chinese AI capex race.

UCTT

High-growth semicap subsystem play at TEEN 2027E EPS multiples. NVDA's supply gap forces capacity adds across logic, HBM, and packaging; UCTT is a packaging and subsystem beneficiary. Cheap way to own the physical buildout. Risk is customer concentration and lumpy orders.

MKSI

MKS at a TEEN 2027E MULTIPLE with exposure to advanced packaging and etch/deposition subsystems. NVDA's FY28 guide implies WFE demand above capacity. Rate of change positive. Risk is memory capex swing — pure-play on process equipment intensity.

AEIS

AEIS at a TEEN 2027E MULTIPLE with precision power and control exposure to semicap. HBM and packaging capacity adds are the driver; NVDA's supply commitments jumped to $279B, mostly memory. Power systems demand follows. Cheap if memory capex holds — risk is lumpiness in semicap orders.

CLS

Celestica is an ODM with GPU and custom ASIC program exposure at TEEN 2027E EPS. NVDA's FY28 guide and AWS's 2M GPU commitment are direct demand signals; MRVL's custom silicon ramp flows through ODMs. Watch margin conversion — ODM multiples compress on scale. Rate of change positive.

FLEX

EMS play with GPU and custom ASIC exposure at TEEN 2027E EPS. AI server buildout is the tailwind. Revenue growth is not the debate; margin conversion is. FLEX's role in NVDA's supply chain supports backlog — watch mix shift toward higher-value AI systems.

SANM

Sanmina is an ODM/EMS with AI infrastructure exposure at TEEN 2027E EPS. Hyperscaler capex of $1.1-1.3T in 2027 is the demand umbrella. Story needs margin and mix disclosure to re-rate. Value way to play physical AI buildout — risk is customer concentration.

MOD

Modine is a data center industrial with large backlog at TEEN 2027E EPS. Modular construction only reduces COD slip probability 10-20%; power delivery is still binding. 2027 NA AI DC demand ~35GW IT versus 16.5-23.4GW deliverable. Physical AI capex without chip multiple — watch execution on thermal and power content.

FTAI

BTM power play with large backlog at TEEN 2027E EPS. Google and Nvidia alone are adding ~45GW of compute next year; BTM generation will go brrrr. Large gas turbine slots sold out through 2031. Risk is credit access for a leasing-heavy model — scarcity pricing is the tailwind.

BW

Babcock & Wilcox is a BTM power and equipment play at TEEN 2027E EPS. Power is the binding constraint — 2027 NA AI DC demand ~35GW versus 16.5-23.4GW deliverable. Long-cycle equipment pricing power is the story. Execution risk is high — high-beta power shortage trade.

GNRC

Generac is a BTM generator play at TEEN 2027E EPS. AI data center power needs and grid delays push backup generation demand. The 35GW demand gap is the driver. Multiple reasonable if backlog converts — consumer-industrial expression of the power crunch.

ARM

ARM is in the richly valued AI infra bucket. While many AI infra names trade at teen 2027E EPS, ARM trades at a large premium. IP royalty leverage to custom silicon is real, but already in the price. Risk/reward worse than semicap and industrial peers — watch custom silicon design-win flow for continued royalty upside.

CIEN

Ciena is richly valued versus teen-EPS AI infra peers. Optical systems demand from AI datacenter networking is real, but the multiple is extended. NPO/CPO transition creates both upside and technology risk. Market is pricing a seamless Optical 1.6T ramp — wait for a better entry.

---

2. Street Color / Heard (unverified)

Word is NVDA stepped back from revenue-sharing deals with AI companies last week (WSJ). Recent NVDA announcements guaranteeing and backstopping datacenter investment suggest much of future demand is self-funded. That trims circular-financing optics — but cuts a financing channel just as credit tightens.

Hearing OAI alone is plausibly ~$800B+ of NVDA's backlog. 12GW of commitments through 2030; blend $25B/GW for Blackwell up to $40B+ for Rubin. If that's right, OAI concentration is the systemic risk in NVDA's book.

Channel checks suggest Meta is actively working to depress Anthropic revenue pre-IPO. Nat Friedman reportedly told employees that using only Meta or OpenAI coding tools could lower Anthropic's revenue before its offering. Also hearing OAI is trying to pre-empt Anthropic's latest model — lab economics getting stress-tested ahead of IPO windows.

Pod books were short NVDA and short software, funded by other chip longs — and they're blowing up. That's the chatter for today's strange tape. If true, the unwind has further to go.

SemiAnalysis math is eye-watering: $2T+ OF 2028 LAB CAPEX WOULD BE 1.5-2.5X ALL 2027 TOP-FIVE HYPERSCALER CAPEX COMBINED. Anthropic inference revenue at $50B/GW annualized on highest-monetizing capacity, seen at $70-80B/GW blended by YE27. Financing feasibility is the crux, not demand.

CRM's own seller rankings now show Anthropic as the #4 competitive threat — behind HUBS, MSFT, NOW, but passing Zendesk and SAP. AI pricing still unsolved. The Claudeforce narrative is winning; the dumb-pipe crowd is losing.

Word is Elliott made a killing on WDAY on both sides of the ledger. The $2B "love tap" kicked off the re-rating. Share count down 9% in three quarters helps the math.

GTLB — no data yet on ultimate upgrades. Open question on enterprise AI-assisted conversion. If nobody has conversion data, the stock trades on narrative.

Agentic shopping research is a sleeper risk for ad/affiliate commerce models. Even small differences in page viewing order changed AI preferences in unpredictable ways. If agents can't be steered predictably, the affiliate layer gets disrupted before it benefits.

Heard at the Deutsche Bank conference: LITE went public attacking COHR's 200G EML yield on 6-inch InP. That's a share-shift signal. InP yield is the choke point at 200G.

Kioxia's ¥1T expansion confirms supply response is coming — but not until 2028. NAND de-specing creates stored demand in the meantime. The precommitment wave is mostly HBM; the spillover tightens NAND.

Word is Commerce is embedded at Intel getting regular process and yield updates. Lip-Bu met Lutnick in DC before the tariff push. The regime ties tariff relief to US manufacturing investment — policy converting TSMC's shortfall into Intel volume.

Retail advice floating around on the Anthropic IPO: don't be a sucker — you'll get it cheaper after lock-up expirations. That's the mood. Pre-IPO hype running into lock-up supply reality.