Good morning. Futures higher, semis leading premarket: INTC +2.3%, SNDK +1.6%, MU +1.1% — the six-week best day in chips carrying over. No major earnings, but NIKKEI +1.5% as BOJ hiked and yen did the opposite (USDJPY +1.2% TO 158.05). Oil down a third day, WTI BELOW $100. Asia/macro: dovish hike from BOJ means carry unwind vol risk is off the table for now, Japanese capital still flowing out — but 160 is the MOF intervention line. Trump-Xi summit next week, tariffs delayed; credit spreads still near bear territory. Three themes: Memory mainline hits Day 4 — Solidigm reviewing first US NAND fab, CXMT entering NAND, story shifts from price hikes to capacity expansion (bullish MU, SNDK, SKHY). Foundry pricing power is real — TSMC +10% WAFERS, AMD FOLLOWING ~10% IN Q4 (passive), CORNING +15% GLASS; watch NVDA/AVGO if they don't follow. AI capacity spills into secondary data centers and power — Anthropic/OpenAI hunting smaller deals, WULF SIGNS 20-YEAR ANTHROPIC LEASE, structural bid for WULF, GNRC, GEV. Software's AI narrative flips from fear to adoption — SALESFORCE AGENTFORCE 7X ROI, Truist says tech most attractive since ChatGPT launch, but part is short covering (CRM, HUBS, INTU). We'll hit up MU and WULF first, then get to software and the power chain.
Broken story. Not a dip — a reset, and the Street just tore up its models.
Fluence guided FY26 down after the close on 9/16, and it was ugly: revenue to ~$2.4B from $3.0B, adjusted EBITDA to a ~$200M LOSS from a ~$10M loss midpoint. Houston enclosure production delays from the contract manufacturer drove $450M of the revenue cut. That's SIX WEEKS after the last reset on the fiscal Q3 call. Credibility was already thin — now it's gone. Stock closed $9.30, gapped to ~$7.5, down 54-61% YTD depending on the print, sitting on the 52-week low. Consensus ratings went from "buy the storage theme" to "prove it" in 48 hours.
Consolidated picture: the PT cluster collapsed from a $16-24 band to $3-15, with the BULK of the Street at $4-7. Not a single firm raised. Ratings are now dominated by Hold/Neutral/Underperform — the one surviving Buy (Canaccord, PT cut to $15 from $24) is the outlier, not the consensus.
Bear (the consensus, and the winning side right now): This is a cash story, not an earnings story. BNP Paribas says the company needs ~$150M of corporate debt in FY27 — 10% of its ~$1.5B market cap — and doesn't see adj EBITDA breakeven until FY28. BNP's FY27 revenue estimate of ~$2.7B sits 30% BELOW Street. Jefferies, which upgraded in February expecting 45% revenue growth and positive FY26 EBITDA, now models 6% growth and a $200M loss. The firm flat-out says it has "meaningfully less confidence in the fiscal 2027 and beyond fundamental story." RBC thinks recovery takes multiple years and directly questions whether Fluence can still WIN contracts in a competitive market. Management is now "right-sizing" by limiting new orders — that's code for shrinking to survive.
"Another hit to confidence in execution, which was already thin." — Truist Securities
Bull (thin, and getting thinner): The $2.9B FY27 backlog is real and unimpaired. Piper frames management's pivot as executing on existing backlog rather than chasing growth — slower, but it protects cash, and the firm actually LOWERED working capital investment estimates through FY28 on the revised plan (guide is neutral-to-positive operating cash flow next year). Fluence is a pure-play on U.S. BESS buildout, and the Houston issue is a SUPPLIER execution problem — a contract manufacturer's enclosures — not a demand problem. If the CM fixes yields and Fluence delivers the backlog, FY28 is a trough-earnings recovery with a levered equity. Canaccord keeps its Buy, just at a much lower number. Nvidia-adjacent, AI-datacenter-adjacent storage demand hasn't gone anywhere.
The problem with the bull case: it requires you to trust management again, and the bull thesis has now been reset TWICE in six weeks. That's not a setup — that's a pattern.
New and incremental:
It's not revenue. It's liquidity. Every sell-side note that matters right now routes back to the same place: does Fluence have enough cash to work through a multi-quarter Houston fix without a dilutive raise? BNP answered with "$150M of debt in FY27." Truist cut FY26 adj EBITDA to a -$197M loss and only gets to +$17M in FY27. That's a company burning its way toward a financing event.
Negative for the U.S. BESS pure-plays — FLNC just handed the market a reason to question every domestic storage name's manufacturing ramp and margin profile. Watch the CM angle too: if one contract manufacturer's enclosure yields can knock $450M out of a flagship name's revenue, that's a supply-chain yellow flag for anyone leaning on third-party U.S. manufacturing for IRA content. The demand side (AI datacenter, grid storage) is intact — this is a Fluence-specific execution and balance-sheet problem, not a thesis-breaker for storage demand. But there's no read-through BUY hiding in this file today. Not yet.
Bottom line: I'm staying away. The equity is cheap on the wrong numbers, the Street has capitulated, and the next catalyst is a cash-flow print — not an order book. You don't buy a story on the third downgrade cycle until management shows a clean quarter, and we're at least two prints from that. Shorts have the ball.
The Street walked out of UNBOUND and the analyst day still split down the middle — because nobody got the one thing they came for: a new growth framework. Bulls own the margin story, bears own the ARR trajectory, and both sides are working off the same tape. Stock at $229.58, -55% over the past year, -42% YTD. That's a lot of bad news already in the price, which is exactly the bull's point.
Two hard data points came out of Boston. One: FY2030 adjusted operating margin target raised to 30%, a 5-point bump from the prior 25% long-term guide. Two: nothing on growth. Management left the top-line framework untouched, so the range of outcomes around when AI monetization actually shows up in the numbers stays wide open.
Q2 already told you where things stand — 17% cc revenue growth, down a tick from 18% the prior quarter. Management pinned it on pricing/packaging changes plus a softer demand backdrop. The part that spooked people: longer sales cycles and higher buyer scrutiny are "unlikely to ease in the near-term." That's not a one-quarter air pocket, that's a reset.
Bull case (Canaccord, Needham, Raymond James): Valuation already discounts a messy top line, and the margin ramp gives you earnings compounding while the AI commercial model matures. The product work is real — Smart CRM enhancements, agent orchestration, the expanded ChatGPT relationship with a joint AI Growth bundle — and the headless/agent strategy positions HUBS to reaccelerate once customers figure out their own AI path. Canaccord's line is the cleanest version:
"The stock is priced below what an established franchise with significant distribution, an expanding AI opportunity, and several more years of operating leverage should command."
Bear case (BofA, BMO, Wolfe): The growth deceleration isn't macro noise — it's competitive and structural. BMO flags AI and competitive risk plus slowing net new ARR into FY26; Wolfe downgraded on the weaker guide. BofA's Matt Bullock ran the math and kept Underperform, arguing a growth catalyst looks less likely in 2026, not more, and that the company's product/GTM experimentation is a tell that the old playbook stopped working. BofA's $200 PT sits below spot. Stifel's Hold (PT nudged to $225 from $200) is the fence-sitter's version of the same concern.
Read the two camps honestly and the disagreement isn't about whether AI is an opportunity — it's about timing, and whether "later" is 2027 or never. Needham framed it exactly right: the key question is whether the slowdown is macro or customers hunting for their own path forward.
Canaccord has it at 2.5x EV/R and 11x EV/FCF on CY27 estimates, against an 83% gross margin. That's cheap for a franchise with this distribution — if growth stabilizes. It's a value trap if net new ARR keeps sliding.
The margin target is the tell that management knows the top line is stuck. Raising the FY2030 op margin bar by 5 points is what you do when you can't credibly raise the growth bar. Doesn't make it wrong — earnings power is real — but don't confuse a margin story for a reacceleration story. n/t the near-term upside skews to margin, not ARR, and that's a "show me" setup for anyone long the reaccel thesis.
Verdict first: Dreamforce was a narrative win and a numbers dud. Management held F27 guidance AND F30 targets unchanged — no raise, no trim. That's the whole story. Bulls got a refreshed agentic pitch, bears got "mixed pricing" on partner checks, and the Street's price targets still span a comically wide $225 to $315 on the same set of facts. Stock sits ~$243.
RBC walked out "incrementally positive" — demand ticked up in partner checks, pricing came back mixed, AI Force now pitched as the enterprise context layer. Sector Perform, $250 PT. That's a hold with a compliment, not a conviction call.
Citizens went the other way. Market Outperform, $315 PT, and the most useful thing they wrote is the honest part — the drags are still drags:
"Several product areas remain a drag on Salesforce's business including Commerce, Marketing, Analytics (Tableau), and Integration (MuleSoft)."
That's a bull admitting the core is leaking. Worth more than the TAM slide.
Two clusters. Bulls at $275-315 — Citizens $315, Stifel/Cantor/Guggenheim all at $300 (Buy/OW/Buy), Oppenheimer $275. Holdouts at the bottom: RBC $250 Sector Perform, DA Davidson $225 Neutral. Note DA Davidson raised its PT to $225 and still kept Neutral — that's a house saying "fine, but I'm not paying up." Adding to the constructive side: 34 analysts revised earnings upward for the coming period.
Bull: You're paying low-23x earnings and a sub-0.4 PEG for the incumbent context layer of the agentic era, with real pricing optionality — seat-based, pay-go, outcome-based, flex credits, AELAs, premium upgrades. Nine of ten AI natives is a land grab, not a slide.
Bear: Unchanged F27 AND F30 guidance tells you management doesn't see a near-term AI hockey stick. The drag businesses still drag. And "mixed pricing" is code for Agentforce deals not expanding ACV the way the marketing implies.
Whether Agentforce/Data ARR holds the +210% next print — that's the load-bearing number. Second, the pricing mix: if consumption/outcome-based creeps up as a share, the ACV debate resets entirely. Third, does the $300 cluster stick or does RBC/DAD converge up? Not sure we can read too much into a post-conference PT shuffle, but the direction of that migration matters more than the level.
Verdict: management just held a public reset and the Street can't agree on the new math. Stock at $313, DOWN 53% Y/Y, and the investor day wasn't a victory lap — it was a confession that INTU lost 1PT OF IRS E-FILE SHARE in FY26 (DIY, price blamed) and a declaration that FY27 trades ARPU for customers. That's a credibility hit. But at 19x and a 0.9 PEG, the tape is pricing a broken model — and both bulls on the print are leaning on the same word: conservative.
Evercore's framing is the cleanest read of what changed. Intuit is now running a barbell: Credit Karma Tax at $0 federal / $15 state hunting the price-sensitive filer it bled to, traditional TurboTax DIY holding the complex-return customer, and TurboTax Live/Assisted pulling the rest upmarket on AI + human help. The point isn't the tax P&L anymore — tax becomes the acquisition funnel into Credit Karma Money and the rest of the financial products stack. Intuit said the quiet part out loud: they will prioritize customer and e-file growth over revenue per initial tax transaction.
GBS gets the same treatment — QuickBooks Free/Lite for the top of funnel, Money embedded deeper into workflows, mid-market carried by vertical-specific builds and the accountant channel.
Bull: Assisted Tax + Money + Mid-Market are now ~30% of revenue and growing north of 30%, which is a real growth engine sitting under a reseting core. 81% gross margin, FY27 guide deliberately sandbagged on monetization, Assisted growth and new acquisition spend — deliver on customer growth and the multiple expands off a multi-year low. Mizuho:
"Fiscal year 2027 guidance embeds conservatism around customer monetization, Assisted growth and newer acquisition initiatives, creating potential upside if execution remains strong."
Bear: you're intentionally discounting the highest-margin, most seasonal cash flow in the company on a promise of LTV that shows up years later — and this is a management team asking for patience after a 53% draw. Wolfe went to Peerperform on tax revenue and the low-end strategy. Truist cut to $300. Freedom to $370. Piper "raised" to $290 — which is still BELOW spot, a tell on how little credit the Street gives the plan.
Q4 rev $4.354B, +14%, beat by 2.1%. GBS $3.424B, +14%. Credit Karma +16.5%, ahead of consensus. The dispersion is the story: PTs now sit in a $290-430 band, a 40%+ spread on the same set of slides, which means nobody has conviction on the reset math yet.
Three things resolve it: FY27 e-file share, Credit Karma Money attach off the free-tax funnel, and QuickBooks Free/Lite conversion rates. Early traction on the latter two is what Mizuho says gives them comfort. Not sure we can read too much into one quarter of Lite data — but if e-file share stops bleeding in the spring, the multiple re-rates fast.
Verdict: the Analyst Day was a nothingburger. ON laid out a double-digit growth case, the model came in right on UBS's preview, and — this is the tell — margins stayed put despite a data center demand environment that's "improved significantly" since the old model was built. Stock sits at $66.60. Every fresh PT lives above it, but the direction of travel is the story: Stifel Hold to $75 from $90, Cantor Neutral to $90 from $110, Mizuho Outperform to $95 from $110. UBS holds Neutral at $95 and flat-out says it prefers other names in analog. That's a cluster of cuts dressed up as maintenance.
FY26 revenue growth guided to 9%, with a pitch for double-digit annually. Fine. But the Street wanted more — UBS sensed investors walked in expecting numbers above the previewed model — and got an in-line print instead. No margin expansion on better demand is the bear case in one line: ON is participating in the data center recovery, not levering into it. The bull case is 23 upward EPS revisions and a PEG of 0.95 (cheap on growth, if the growth shows up). The steelman for the bulls: sentiment is inflecting, estimates are moving the right way, and $66 against a $75-95 PT band means the risk/reward skews long — you just need the margin line to eventually follow demand.
Worth noting every PT here is a cut or a hold, and none of these analysts moved off Neutral-ish stances following the event. On a light-news tape, that's the signal.
"ON Semiconductor maintained the margin assumptions from its prior model, though the demand environment has improved significantly since that model was established, particularly in data center." — UBS
So: participating in the cycle, not compounding through it. Until margins move, ON stays a show-me story.
Verdict: KeyBanc's pre-analyst-day note is a margin call, not a growth call — and that's the tell. Firm reiterates Overweight, $30 PT (stock at $24.57, +141% YTD, $3.91B cap) into Tuesday's analyst day in New York, and explicitly says it is NOT looking for multi-year revenue targets. Read: the market already pays up for growth; the next leg has to come from operating leverage.
"Not expecting or seeking multi-year revenue growth targets at Tuesday's analyst day... looking for upside to its prior margin forecasts."
KeyBanc models 36.1% incremental margins over the next two years, took FY27/FY28 EPS ests up 4.1%/6.6%. Backdrop is a blowout Q2 — adj EPS $0.15 vs $0.07 est, rev $183.3M vs $173.9M, +23.3% y/y against 17% consensus, fastest growth in ~4 years. Network services +16.5% vs 11.1% expected. Evercore Outperform $32; Raymond James to $29 from $23 on AI/inference traffic tailwinds.
Bull case: AI traffic is real, network services reaccelerating, 61.5% GM plus that incremental margin guide gives EPS genuine torque. Bear case: consumption models cut both ways, +141% YTD prices a lot of good news, and analyst days are classic sell-the-news setups when expectations run this hot. Not sure you need to be a hero into Tuesday.
Freedom Broker initiates Buy with a $10 target against a $5.02 handle ($774M cap) — a straight double on the AI-interconnect angle. The thesis is clean enough: Lightwave slaps its proprietary Perkinamine electro-optic polymers onto standard silicon to do data modulation that plain silicon handles badly, cutting voltage, footprint, and power while hitting the speeds AI interconnects actually need. Tradeable catalyst is the ramp, and management pegs it to H2 2027. That's the whole r/r — everything between here and there is prototyping theater.
Five prospective-customer programs sit in late-stage prototyping (Stage 3: prototypes, foundry runs, testing). Tower is integrating reference designs into its PH18 PDK, with flows pointed at GlobalFoundries and SilTerra — that's the four-foundry partnership progress, and it's the piece that actually de-risks volume. Balance sheet is the quiet bull case: $95.9M in cash and marketable securities, DEBT-FREE, funding operations through at least December 2027. They can afford to wait for the ramp without a dilutive raise.
Bull: the polymer-on-silicon value prop is real and the foundry ecosystem is engaging — Tower/PDK integration plus a GF/SilTerra path means design-ins could convert fast once 2027 hits. Cash runway covers the whole commercialization window, so no cap-structure overhang.
Bear: Q2 2026 revenue was $33K versus $26K a year ago — that's not a business yet, and the net loss widened to $6.6M from $5.0M as spend ramps on manufacturing readiness. You're underwriting a 2027 promise with zero revenue visibility today, and "late-stage prototyping" is a phrase that has lived for years without converting.
Not sure we can read too much into a single initiate with a template-y writeup — but the foundry integration detail is the one thing worth tracking into next year's prints.
Rosenblatt reiterates Buy, $250 PT — and the actual news is the pricing update, not the note. CRWV says it's "continued to contract new compute capacity at higher prices," one day after Nebius hiked across its book, including +16% on FOUR-YEAR-OLD SILICON. That's the whole debate. The bear case on neoclouds was always commoditization plus depreciation eating the P&L; two of the biggest operators pushing price in the same week argues the opposite — capacity is still scarce and contracts are repricing up.
Then the balance sheet. $6.3B raise — $3.0B convert + $3.3B ATM — with management naming "migrate toward investment-grade" as a use of proceeds. Read that as an attempt to term out a $51.6B debt stack sitting at 0.9x debt-to-market-cap. Rosenblatt's framing is that you're getting it cheap: 5x 2027E EV/EBITDA vs ~10x for NBIS and the peer average, against $47B of GPU capex driving $40B annualized revenue on shorter-dated contracts.
That's the r/r in one line — pricing power and multiple compression on one side, dilution plus residual-value risk on the other. Stock's 48% below the $153.20 52-wk high, -6.5% on the week. Rosenblatt at $250, Truist lifted $155 to $165 — a spread that wide tells you nobody has conviction on terminal GPU values yet.
"CoreWeave has continued to contract new compute capacity at higher prices." — company statement
Peripheral but directionally supportive: multi-rack Nvidia Vera Rubin NVL72 deployed on CoreWeave Cloud, new AI Object Storage features (cross-region write accel, Archive tier), and Crusoe locking a ~$13B five-year Jane Street deal. None of it moves the thesis. All of it says demand isn't the constraint.
Truist reiterates Buy, $55 PT, and keeps APG as its TOP PICK — the pitch is that the tape sold APG like a pure-play AI capex proxy when only ~11-12PP OF REVENUE actually touches data centers. Stock is -20% SINCE MAY 1 while the S&P added 5%. Truist's call: the AI rollover risk sits in the price, and you're getting the Safety segment for free.
Break the exposure down and the bear case thins fast. Of that 11-12pp, ~3pp is recurring service work — sticky, higher-margin, indifferent to a hyperscaler pause — and the remaining 8-9pp is construction, i.e. the only piece that actually gasses on an AI capex stall. Safety should print >5% organic in Q3. Wells Fargo just initiated OW with a $53 PT, framing the inspection-first model as a moat around less-competitive, higher-margin work. The quarter backs it up: Q2 adj EPS $0.44 in line, revenue $2.25B vs $2.2B bogey, +13.3% y/y on 10.1% organic. Street targets run $47-55, so Truist sits at the top of the range — r/r here is really about whether you buy the de-rating as overshoot or as the market front-running the capex curve.
The firm sees an attractive entry point following a selloff in data center infrastructure-related equities... a potential rollover in AI capital expenditures would affect the Specialty segment. Truist believes this risk is already reflected in the current stock price.
Worth flagging: one note, one broker, and the whole long case rests on the mix math holding. Not sure we can read too much into a single reiteration, but the setup — beat-and-raise name trading like a cyclical — is the kind of dislocation that gets PMs' attention.
Needham upgrades DT to Buy from Hold, PT $68 — the lone new voice today, but it lands on top of a Street that's already flipped constructive. Stock at $55.83, basically sitting on its 52-week high of $56.27, +39% over six months. The tell: Needham isn't underwriting a new story, it's marking to market a multi-year GTM/product repositioning it had been sitting out — DPS renewal cohort as the net-new ARR acceleration lever, Logs + AI observability as incremental, European data sovereignty as the sleeper catalyst (Austrian heritage, nobody else owns that lane). Arize at $915M is the strategic hinge; MS moved to Overweight $65, UBS stays Buy $65, BMO $62, DA Davidson $65. Consolidation is real — new-money PTs clustered at $62-68 vs. a Street that spent most of last year in the $50s.
The bull case is the ARR inflection: UBS flagged Q1 organic net-new ARR +41% y/y. The honest bear read is in the same note — strip out a large deal pulled forward and that's 22%. That's the whole debate in one number. If DPS renewal cohort conversion is genuinely structural, 20s comps re-rate; if Q1 was pull-forward plus a messy M&A integration, you're paying a premium multiple at the 52-week high for a story that hasn't printed two clean quarters in a row. r/r is fine, not fat.
"The upgrade reflects strengthened conviction in sustained outperformance, with DPS identified as the key ARR acceleration lever." — Needham
Watch the next print for DPS cohort disclosure and any Arize revenue contribution cadence — that's the falsifier, not the headline ARR.
Verdict: a lone bull on a light tape — no new information, but Tigress picked a side on the only debate that matters for GOOGL right now: whether the AI capex cycle is dilutive to ROC or a moat-building spend that reaccelerates economic profit later.
Tigress Financial took its PT to $485 from $415, keeps a Strong Buy. Cited: LTM revenue +20% TO $445.87B, ROE 50%, P/E 17.38, PEG 0.15. That PEG number carries the whole argument — it only looks cheap if you underwrite AI-driven monetization inflecting hard, so treat it as a forward-growth assumption, not a value signal. (A trailing 17x on GOOGL doesn't clear the bar for me either; don't anchor to it.)
Thesis in one line: AI expands Search and improves ad performance, Cloud/Gemini/TPUs open new monetization rails across Search, Cloud and YouTube, and scale plus 50% ROE funds the capex while still returning cash.
"Alphabet's AI investment cycle temporarily compresses return on capital but positions the company for accelerating economic profit growth." — Tigress
Product news flow is cadence, not economics: updated Gemini managed-agent harness (40% FEWER OUTPUT TOKENS on file changes, better task completion), a Gemini app for Windows, family sharing across up to six members on the CC agent, Dreambeans, and passkey portability on Android.
Steelman both ways. Bull: capex is a land grab and Cloud attach converts it into a durable revenue line — Search monetization holds. Bear: depreciation ramps into margins exactly as AI answer formats threaten query economics, and the PT revision is one broker extrapolating a 2Q print into a 2027 earnings story. Single note, no cluster. Nothing here to trade off — file the ROC framing, wait for the capex line.
Verdict: the AI infrastructure trade finally has a second act, and Nokia is it. B.Riley initiated Buy with a $15 target, and the reason isn't radios — it's the plumbing under hyperscaler clusters. OPTICAL NETWORKS +20% Y/Y, IP ROUTING +16% Y/Y, against a 7% LTM revenue base. That's the mix shift that re-rates a "telecom equipment" multiple into a data center multiple.
The bull case in one line (B.Riley's framing, and I buy the logic):
The carrier inventory correction is over, and the €2.8B AI/cloud order intake booked in Q2 gives visibility into 2027. Sum-of-the-parts — legacy assets separate from the data center business, premium hardware multiple on the latter — gets you to $15.
The SOTP method matters. It tells you the sum of the parts is the trade, not the consolidated earnings power. Legacy mobility still drags the print: Q2 came in at $5.51B rev vs $5.59B bogey, only the $0.08 EPS vs $0.06 beat saved it. AI-RAN trials with A1 Group and Chunghwa on NVIDIA's Aerial stack span four regions — optionality, not revenue yet.
Steelman the bear: stock is UP 119% OVER THE PAST YEAR, so the easy money is made; China retreat (closing most mainland sites by year-end) is a real revenue hole; and Rosenblatt's Buy alongside S&P's outlook revision to positive means you're not early or differentiated here anymore.
Own the optical/IP routing line, not the whole story. r/r is fine from $10 with a $15 target, but the next leg needs order intake to convert.
Evercore's annual survey says the iPhone 18 cycle is shaping up better than the base case — and the driver isn't AI, it's old hardware. PT goes to $380 (prior level undisclosed — the same note carries a stale $365 reiteration, take $380 as the fresh print), Outperform stays. Stock at $337, just under the 52-wk high of $344.57, +35% over six months. BofA sits at $370 with the same read, Citi's survey shows upgrade frequency picking up. Nobody's fading this one.
~4,000 consumers. 53% intend to buy an 18 Pro or Pro Max vs a 51% historical average. 32% want the Pro Max specifically vs 29% LY. Duo at 14% vs Air's 9% a year ago — launches mid-October, so Evercore expects that to build in Q4 once people get hands on it. Wearables tacked on: Watch intentions 38% vs 34% (skew to Series 12), AirPods 43% vs 40%.
iPhone purchasing drivers were led by respondents indicating their existing iPhone is old and needs upgrading, at 67% versus 48% last year.
That's the bull case in one line. The wrinkle: average device age came in at 23.5 months vs 26.5 last year. Younger phones, more people calling them ancient. Not sure we can read too much into that — survey framing or mix shift — but it's the number I'd want explained before underwriting a supercycle.
BofA's angle: fatter carrier incentives are absorbing Apple's higher pricing, which keeps unit demand intact at the new ASPs. The xAI/X antitrust suit settled — that headline risk is gone. Kalshi filed to list single-stock perps on AAPL; flow story for later, not a reason to own it today.
Verdict: Barclays joins the silicon photonics chorus. Initiates OW with a $310 PT, calls TSEM the leading SiPh foundry. Stock's at $219 — up 218% YoY, trades 85x — and still sits below the $270-367 PT cluster. This is a narrative stock now, and the narrative is AI connectivity.
Barclays' Simon Coles leans on two decades of ecosystem development and leading share in pluggable products. LightCounting's 24% Ethernet unit CAGR (2026-31) is the demand backdrop; Barclays thinks PIC wafer growth runs faster as die sizes expand. The hard number that matters: ~$300M in customer prepayments supporting at least $1.3B of 2027 revenue — Barclays models $1.6B — with more prepayments expected early 2027. SiPh revenue hits $500M by Q4 2027 under their model, 62% of total revenue by 2028. PT is 30x 2028 P/E. That's a 2028 multiple on a 2027 revenue story — you're paying for the out-year.
Consolidation: Stifel Buy $270, Barclays OW $310, StoneX and Benchmark Buy $335, BofA Buy $367. Same thesis across the board — contractually committed SiPh revenue of $1.3B backed by $290M prepayments. Benchmark flags a wildcard: speculation about FCC restrictions on Chinese-made optical transceivers, which would be a structural tailwind for TSEM if it lands. Bull case: bottleneck leverage in a market that needs capacity, prepayments de-risk the build. Bear case: 85x trailing P/E prices in flawless execution, and every PT in the cluster is 2028-based — any SiPh order pushout and the multiple compresses fast.
"Tower Semiconductor holds leading market share in pluggable products and process performance from two decades of ecosystem development." — Barclays
BMO flips Market Perform to Outperform with a $335 PT, and the setup justifies it: MTSI is DOWN 36% FROM ITS MAY PEAK while estimates kept grinding higher. That's multiple compression, not thesis break. Forward multiple went from HIGH-50s IN JUNE to 29x now, PEG of 0.2, and the fundamentals never cracked — datacenter is the engine, defense stays resilient, and space becomes a material contributor next year. Q3 already in the tape was a beat: EPS $1.40 vs $1.35, rev $342.2M vs $335.3M, UP 35.8% Y/Y, record bookings.
BMO is excited about datacenter as hyperscalers, cloud service providers and AI labs continue to deploy AI compute and networking infrastructure.
Steelman the bear: 29x isn't cheap in absolute terms, and the source's own Fair Value read says the stock may still be rich — after a 103% twelve-month run there's plenty of hot money still long. Also, treat that article with a grain of salt: it references a BMO Market Perform initiation at the same $335 PT, which reads like mangled/stale copy rather than a second opinion. One article, one real signal — but the r/r on a beat-and-raise name that de-rated 1,800bps in four months is what PMs should be circling.
TD Cowen keeps Buy and $400 into the investor forum — but the model, not the print, is doing the work. Stock's at $356 after a +150% YEAR. Cowen rolled out above-consensus FY28 estimates and anchored the target at ~21x FY28 EPS. The pitch: a >30% revenue CAGR through 2029, GM expanding to 50%, OM in the low-30s. Ciena runs 44% GM today with 33% LTM growth — that 600bp of gross margin headroom IS the trade. Supply constraints, not demand, gate the ramp.
TD Cowen highlighted Ciena's financial model projecting a compound annual growth rate exceeding 30% through 2029, with gross margins reaching 50% and operating margins in the low-30% range.
The dispersion tells you everything. Post-FQ3 ($1.67B REV +37% Y/Y, adj EPS $2.11 +215%, beat by $0.38), the targets run from $347 to $615 — a 78% spread on the same tape. Stifel's the high-water mark at $615 Buy on guidance. Rosenblatt trimmed to $525 purely on sector multiple compression, not fundamentals. MS raised to $450 on cloud demand and backlog. UBS cut to $394 on margin timing while lifting FY27/28 EPS. B.Riley sits at $347 Neutral — below spot. Nobody's fighting about the demand curve anymore. The fight is the multiple and how fast 44% GM becomes 50%. B.Riley's PT under the current price is the honest bear mark: great numbers, wrong entry.
Verdict: SNOW is UP 90% IN SIX MONTHS and the sell-side has stopped debating the story — it's just marking up the multiple. StoneX keeps Buy and $400 after a management fireside, but the real signal is the PT cluster that got rebuilt post-print: $425-450 from the bulls, Bernstein at $377 (up from $250) on a 15x target multiple. The debate is no longer "is consumption reaccelerating" — it's "what do you pay for it."
Bernstein SocGen raised its target to $377 from $250, reflecting increased growth expectations and an adjustment of its target multiple to 15x.
That's the tell — the $250-to-$377 move is half numbers, half multiple. When the entire committee re-rates one direction after a 90% run and nobody publishes a bear, the risk is positioning, not thesis. Not sure we can read too much into a single fireside, but the asymmetry here is obvious: SNOW has to keep beating just to hold $330.
MU — Three things stacked at once: 5-7x memory pricing with HBM 2027 contracts up 80-100% (RBC), an ABF substrate deficit at 14%/34%/51% through 2028 (Goldman), and Hot Chips saying accelerator compute compounds ~3x per two years while HBM bandwidth does less than 2x. That widening gap IS the rent. SK hynix NAND at $14.27B, UP 89.5% QoQ, kills the "it's just HBM" framing — the whole complex repriced. Sanand ATMP ($2.75B, hundreds of millions of units in 2027) is long-dated supply, zero relief for the current squeeze.
SNDK — Direct NAND beta and the cleanest LTA expression. Kioxia has ~50% of LTA capacity locked. Oct2 1600c traded 4,200x for ~$41M premium — someone is paying up for continuation. The bear case doesn't exist until CXMT NAND scales, and that's 2027-28.
SKHY — Solidigm's US NAND fab review is a customer-structure hedge, not a cost decision. Whoever signs LTAs with a US-based SK hynix buys supply security — TSMC Arizona logic applied to a commodity cycle product, which tells you the premium clients will pay. Indiana packaging plus possible Intel memory cooperation deepens the footprint. Regulatory multiple, not a margin multiple.
INTC — Trades as a memory geopolitics asset now, not a foundry. Lip-Bu Tan's 5-7x / INTO-2027 call, the Hynix-Ohio JV chatter (real structure $100B+ vs the $7B headline, Mod2 is 345K sq ft and filling it with DRAM tools costs $5-7B), and a client CPU hike alongside TSMC and AMD. The bear case never moved: own roadmap and capex ramp. Bulls need the Hynix JV to actually close.
TSM — The 10% wafer hike is the cleanest pricing power confirmation of the cycle. Longtan back on with three sub-1.4nm fabs, NT$1T+ (~$31.4B), first MP 2030 — the 2023 retreat reversed. CoWoS-L mainstream through 2028 with NVDA high-end GPU shipments +30% YoY in 2026. The print question: does the hike land in gross margin guidance or get absorbed.
AMD — The 10% Q4 increase is defensive, full stop. No buffer against TSM. MI355X closing perf/TCO on GB300 in agentic inference is the only genuinely new bull datapoint and it's unverified. If NVDA eats the wafer hike to defend share, AMD's perf/dollar gap widens and it has to raise price anyway.
NVDA — Jensen's "double next year" plus a 10M accelerator target circulating. Huawei's roadmap (960DT 1Q27, 960PR 3Q27 at 8 PFLOPS FP4, 970 in 2028 at 14) narrows the gap from ~3 years to ~2 — but memory bandwidth, not logic, is the harder bottleneck. Software moat is the real story: Astra ported Blackwell code to Vera Rubin with no rewrite, 3x throughput, then another 2x from optimization in 72 hours. That's switching cost, not a benchmark.
AVGO — Losing Google TPU V9 networking/interconnect to MediaTek is a real share warning. Broadcom does not own TPU attach in perpetuity. Meta MTIA Arke 1H27 / Astrid late-2027 (~$8.5B cost savings on 2027 infra) keeps the franchise intact, and AVGO has the margin cushion to hold price and buy share if TSM costs bite — that's a weapon.
AMAT — $5B/decade India commitment, 140-acre research park, doubled India R&D. But management punted growth guidance on unclear INTC order intent. That's a timing problem, not a demand problem — if it persists, the whole semicap 2027 revenue cadence slips right.
ASML — JPM refreshing into the print, Hsinchu hiring signals Taiwan expansion. Pure beta to leading-edge capex like Longtan. No new tape.
ARM — Haas says supply is the constraint on $2B of AGI CPU demand, not demand itself. Austin bring-up is real silicon. Right framing, wrong year — this is a 2027-28 revenue line.
GFS — Japan-US fab talks (¥2-3T, up to $19.3B) would be roughly half GF's market cap and removes the balance-sheet-constrained capex tail risk entirely. Details thin. Advanced packaging and silicon photonics services expand, but SiGe rated meh vs Tower — mixed execution.
MRVL — GFS SiGe capacity agreement secures specialty supply for pluggable/NPO/CPO. +5.4% post-Fed, Oct2 250c ~$3.85M. Custom ASIC socket churn (see AVGO/TPU V9) is the sector-level caution.
AEHR — Best order datapoint in the group: $22M AI wafer-level follow-on, $41M packaged-AI, plus silicon photonics production orders. Repeat orders validate wafer-level burn-in beyond SiC. Next proof is margin and cash conversion, not bookings. Crowded shorts squeezed.
FORM — Optical/electrical wafer probing and KGD test. Test content per optical link rises as CPO hits HVM. Shorts already squeezed; follow-through is the question.
KEYS — Photonic characterization in the CPO test chain. Leveraged to optical HVM, not pure AI. Content rises before heterogeneous integration bites.
TER — Automated test from wafer to optical engine to CPO. The bottleneck shifts from building photonics to testing every component — that's TER's content-per-link story.
VIAV — CPO module and final test. Smaller position, same follow-through setup as the rest of the chain.
LITE — Lasers and optical components in the CPO test chain. China Q3 previews show 30-75% q/q growth at module makers while optical names pull back. If that's flow and not fundamentals, it's a look.
COHR — The 10-K risk factor on InP capacity ROI and bad 6-inch yields reads like an admission. Margin and capex-allocation yellow flag, plus a competitive opening for other InP sources. This separates COHR from the optical group — the yield issue is theirs.
SMTC — SiGe CTLE amplifiers carrying MediaTek TPU interconnects. Niche, real content, plus LPO/NPO optionality. Not priced as an AI networking winner yet.
CRDO — Sentiment inflecting after a widely read industry report; AHC/AOC growth now broadly understood. Part of the re-rate happened. Momentum intact.
WULF — 20-year Anthropic lease is the most informative datapoint in the power chain. Moves WULF from neocloud spec to AI lab supply anchor, and the duration underwrites inference workload, not one-off training. Risk sits entirely in Anthropic creditworthiness and termination terms — watch the S-1 lease commitments.
GNRC — Initial $2.4B Amazon backup-generation agreement, warrants tied to up to $8B, against $4.2B of total 2025 revenue. Transformative. Amazon just validated power, not chips, as the binding constraint.
AMZN — Paying up to lock backup generation is the tell that the constraint is physical. Trainium 3 rack-scale visible, Annapurna better than believed. Custom silicon dilutes merchant accelerator share over time but buys cost and supply security.
QCOM — AWS warrant deal (9/8) is thin on detail, but the structure is becoming standard. AI capacity optionality, not a business model change.
ORCL — OpenAI usage acceleration is a second-derivative read on the OCI lease book. If OpenAI gains share, ORCL capacity demand rises. Pure derivative, no operating news.
CORZ — $3B ATM on top of $6B, money straight into CoreWeave deployment capacity. Financing arm for the buildout; funded runway intact.
NBIS — Second price hike in three months: GPUs +17-21%, CPUs to +25%, memory ~+41%. H100 $3.85→$4.50, B300 $7.85→$9.50. Clean margin tailwind and contracted capacity on track toward 5GW by year-end. Scarcity rent, not discounts.
IREN — The "power assets are the moat" framing: AI converts power to software, software can't convert back. Neocloud dilution debate still live.
GLW — Display glass substrate +15% Q4. Contract pricing, not AI narrative. EPS leverage if volumes hold, and it stacks on ABF/CCL/wafer inflation. Bear case is consumer electronics demand destruction.
CLMT — Lubricants are the price-inelastic end of the barrel and inputs are a minor cost line. Clean bull read off the diesel record.
VLO, MPC, PSX, DK — Outages at Valero Corpus Christi, Memphis, Ardmore. US refineries at 97% capacity, global runs -6M bpd, diesel at a record $6.27. Crack upside on supply outages — this is product-led, not crude-led, so separate it from the WTI below $100 tape.
META — Muse 730K US downloads since Sept 8 vs Meta AI 707K in its first five days. Consumer agent distribution is shifting and velocity is accelerating (83K on Sept 10). MTIA Arke/Astrid with AVGO is the other leg.
MSFT — Plugin4Shell SHA-pinning flaw on Copilot is both a spend item and a repo-risk headline. Maia 200 (CoWoS-S, 2 SoC/chiplets, 8 HBM, larger dies forcing CoWoS-L) shows custom silicon still behind Google and AWS.
TSLA, UBER — Vegas robotaxi permits. Zoox's Nevada cap expires Sept 25, Waymo already commercial. Competition steps up. Uber's platform angle is the thing to watch, not its permit count.
PLTR — Karp on closed-model token discounts existing to get access to your IP. Open weight keeps proprietary knowledge in-house, cheaper, and it's taking on the classified side. Direct shot at the closed labs.
DELL — Sentiment battleground: bulls at $100+ EPS next year, bears calling it a meme down 50% in two. Results have to clear a very high bar. Micron shipping first India modules for India-built laptops is not the swing factor.
BRK.B — Buffett to chairman, Howard in, Abel stays CEO. $1T conglomerate governance change usually means more portfolio scrutiny. Apple adjustments stop being residual noise.
KKR, GS, SFTBY — KKR says FOMC growth concerns are the lowest since its forecast began. Goldman's Snider says earnings bubble fears are misplaced, and Goldman's own ABF deficit numbers back the memory durability call. SoftBank's credit line +$450M to $6.5B — private AI leverage expanding with 10Y at 5%, no covenant deterioration yet.
TWST, TEM, SDGR, TXG — AI x Science basket +5.2% vs SPY +0.9%. Basket momentum is real. No ticker-specific fundamentals in the feed, so don't size it like there are.
No fresh signal in today's feeds: SITM, ADSK, DY, GEV, SLNH, PEGA, WK, MGNI, CLS, FTNT, CIFR, APLD, JKHY, FPS, KOPN. Nothing to trade off. GEV, APLD, CIFR and SLNH sit downstream of the WULF/GNRC lease prints if you want a theme proxy, but there's no ticker-specific datapoint in the RSS or Twitter sources.
Hearing CXMT is standing up a NAND R&D line pointed straight at YMTC. Commercial scale is 18-24 months out at best, so it doesn't change the 2027 shortage math — it plants a flag for 2028 supply. Anyone underwriting NAND at current pricing into perpetuity needs to mark that.
Word is the Intel-Hynix Ohio structure is a $100B+ JV, or a Hynix-owned fab on leased land, with INTC getting HBM base-die supply preference. That's a much bigger number than the $7B headline. If it closes, INTC re-rates on geopolitics; if it doesn't, foundry execution stays the whole story.
Channel checks suggest Broadcom didn't just lose the Google TPU V9 interconnect socket — MediaTek is now the default second source across the next two TPU generations. Unconfirmed, but the custom-ASIC bulls have been marking it. Socket churn in custom silicon is real and it cuts both ways.
Okamoto (6125.T) wafer-polisher orders UP 370% YoY. The author owns the stock and calls 50%+ upside — discount the target, but the order datapoint is the best 300mm capex tell in the feed.
Hearing onsemi is putting through price hikes. Second-tier analog following foundry is one more leg of the pass-through chain.
Word is AMAT's growth guidance punt is specifically INTC order timing. Demand is there; visibility isn't. If Intel's equipment intent stays fuzzy into the next print, semicap 2027 cadence slips right, not down.
Channel checks on parts: laser tool lead times out to 40 months, deposition at 10 months vs 4 pre-COVID. Supply-constrained oligopoly math. Pricing power sits with whoever has parts on the shelf, not whoever has the best roadmap slide.
Speculative read on credit: HY spreads sit at the edge of bear territory while equities price a full AI capex boom. That divergence hasn't converged since 9/11 and it's the biggest blind spot in this rally. Watch HYG/OAS, not the index.
Hearing the 100K-lot gold call from 9/16 is still on. Chasing and hedging are both bid. That gap usually resolves with a vol event, and nobody is positioned for it.
Word is MOF starts verbal intervention into USDJPY 160. If it triggers, carry unwind vol comes back and it's a pure denominator problem for global tech. AI fundamentals don't enter the equation.
Speculative: Anthropic's IPO window is October. The S-1 lease commitment disclosure is the number that matters. If small 20-30MW deals dominate the lease book, the IPO prices at a discount to the narrative.
Hearing Anthropic and OpenAI are hunting 20-30MW deals in the UK and Nordics directly, not through hyperscalers. Labs aren't waiting in line anymore. Structural for distributed power and secondary data center into 2026-27.
Channel checks suggest token prices are at new record lows while infra spend accelerates. The deflation wedge between volume growth and revenue is intact. That's the bear tell on AI monetization and nobody has solved it.
Word is software is now funding short and the bounce is real money rotating within AI, not new allocation. First production agentic workflow that actually replaces a seat license is the binary — Salesforce's Southwest 7x ROI is marketing until someone shows the seat count coming down.
Hearing Winbond bought Infineon's NOR business for $1.12B cash. Semiconductor M&A spreading from logic into memory. Balance sheets are voting on the shortage lasting.
Speculative on TSM: the 10% hike is the cleanest pricing power confirmation of the cycle. The debate is whether it lands in next quarter's gross margin guide or gets absorbed into customer negotiation. The print resolves it.
Word is warrant structures are becoming the standard hyperscaler lock-up tool — QCOM-AWS, GNRC-Amazon. Lab-level capacity chasing is now forcing second-tier players into the same structure. Watch who signs next.