Good morning. Tape’s grinding up into NVDA, everything else is just noise until Wednesday’s print. Futures quietly green, semis leading — MU +2.5% pre-market after Hot Chips, TSM +1.5% on the AUO panel-plant CPO buy. We’re getting the setup we wanted: fundamentals accelerating INTO the print while sentiment de-rated — NVDA at 17x CY27 vs SPX 19x, stock down 4% over two weeks. HBM IS NOT A COMMODITY — Hot Chips settled that debate with cHBM, zHBM, and 3D DRAM roadmaps across all three vendors. HBM4 consumes ~3x the silicon of DDR for equivalent capacity, so conventional DRAM stays tight even without demand acceleration — that’s a MU/Hynix/Samsung bid plus a direct SUMCO/Shin-Etsu read-through (FIRST WAFER PRICE HIKE IN 3+ YEARS, +10%). Innolight/Eoptolink: the FCC covered-list fear trade is dying — ITI formally opposing, and 2027 1.6T guidance converted to hard orders THIS MONTH. TSM’s AUO panel-plant purchase confirms a dedicated CPO/FOPLP ecosystem build in central Taiwan — extends the moat into optics integration, ripples to BESI/AMAT/TER. NVDA’s $73B ecosystem-financing map (COHR/LITE $2B each) tells you where supply is scarce and where NVDA will own the chain. Asia: Hang Seng +0.8%, Nikkei +0.9%, TSMC +2% in Taipei — the China policy overhang on memory is fading into the background. We'll hit up NVDA and MU first on the callback, then get to the semicap/optics complex — BESI, AMAT, COHR, INNOLIGHT — and frame Wednesday around the circularity question.
Bernstein's the high-conviction name here: Outperform and $800 PT with the stock 31% BELOW its $791 high — that's the whole thesis in one number. Core is compounding (revenue +28%, ARPU $125, 82% gross margins) and the market is treating AI capex like a black hole when the monetization is already trickling in via business messaging, paid model APIs, and subscriptions. The narrative: every incremental dollar gets shoved back into the AI machine, and Meta Connect next month is the next catalyst for product updates and traction data.
"Every incremental dollar generated gets reinvested into strategic AI efforts across model training, talent, and infrastructure build out."
At ~15x forward earnings, Bernstein argues the market has already discounted the heavy-spend uncertainty. Spend stays elevated through at least 2027, with a potential throttle on 2028+ investment depending on how consumer AI demand shakes out. One wrinkle worth flagging: Meta is paying Microsoft hundreds of millions annually for Azure AI models — renting compute while building out its own stack. That's a tell on how real the infra crunch is. Overhangs (UK cinema bans on the glasses, the youth addiction trial) are noise until they aren't — none of it in the numbers today.
Davidson reiterates Buy/$45 into the Aug 25 FQ2 print. Thesis is straightforward: Enterprise Advanced upgrades plus seat growth as AI adoption moves from pilot to production. Stock already +44% over six months and sitting at the 52-week high — market's ahead of the story, so the quarter has to deliver.
UBS gives you the bear case — Neutral, PT to $29 from $28 — but even the skeptic cites acceleration to low-teens growth and stronger-than-expected billings. Billings inflecting is a leading indicator; that's the part that matters more than the EPS beat.
Clean setup from FQ1: EPS $0.37 vs $0.36e, revenue $306M vs $296.5e.
"Box's position in AI-driven enterprise workflows provides secular demand tailwinds."
~80% gross margins gives you the margin durability while AI attach builds. Solid r/r into the print if you trust the upgrade cycle. Light coverage this morning — one contributor on the tape — so don't over-index on any single datapoint.
Clear Street's the lone voice today, but it's the right one: Buy reiterated, $161 PT, shares at $133. The catalyst: THE $6B GRAIN SPECTRUM SALE to SpaceX/AST — that kills the bear narrative that Grain would compete head-on in utility private LTE. At under $0.53/MHz-POP, the market prices this 900 MHz spectrum like a broken asset, not a scarce one.
"This development breaks the bearish narrative that Grain would compete in utility-focused, private LTE networks."
The bid validates spectrum value and removes the competitive overhang in one print. Target dispersion ($75–$450) shows how wide the re-rating band is. Financial Health reads only "FAIR," but liquidity is strong — the optionality comes without balance-sheet risk.
Baird cuts BABA PT to $160 from $164, keeps Outperform. That's not a capitulation — that's a margin call on the margin story. The thesis: AI investments are chewing into near-term profitability while Alibaba defends share in a soft e-commerce tape. Cloud growth remains strong on AI workloads, but it was IN LINE with expectations — no upside surprise to offset the P&L bleed.
The stock sits at ~20x P/E with revenue +4.4% LTM. Not expensive for an AI platform leader, but the market needs to see the margin trajectory bend before paying up. The counter: Benchmark and Mizuho are still at $220 and $195, respectively, after Q1 FY27 EBITDA beat consensus by 7%. So you've got a $160-to-$220 target spread — that's a wide range telling you the debate is live.
"The stock is more suitable for longer-term-oriented investors given current market conditions."
Fair. This is not a momentum name right now — it's a patience name. Quick Commerce unit economics improving is the quiet positive, but the AI capex drag is the loud negative. For PMs with a 12-month horizon, BABA is a coin flip on whether Cloud monetization outpaces the e-commerce drag. For longer holders, the r/r still skews favorable at these levels.
Mizuho holding the line into the Sept 2 print: Outperform, $530 PT. Thesis is simple — Broadcom owns the full stack (ASIC, SerDes, advanced packaging) that Google needs for TPU across MULTIPLE generations. That's a 76% GM and 32% rev growth LTM doing the talking. Street's at Strong Buy consensus, so no contrarian angle here — just confirmation before the event.
"Broadcom has the full technology capability to deliver to Google's TPU deployment plans across multiple TPU/XPU generations."
The bear case isn't demand, it's diversification. Mizuho flags Google shopping for a lower-cost TPU partner — new, untested names carry risk. That's the narrative overhang: not whether AI spend happens, but whether AVGO's slice of the Google pie shrinks over time. (Separately, BofA downgraded AVGO bonds to Marketweight on the XPV/Blackstone/Apollo financing structure — credit sniffing at the leverage, not the semis.)
Rest of the Street still crowded long: BMO initiated OW, TD Cowen reiterated Buy with $500 PT and a $100B+ AI semi revenue bogey by FY27. Wolfe sees strong near-term financing demand, long-term risk. Morgan Stanley says summer 2026 is when hyperscaler capex gets real.
Setup: constructive into the print, but watch the Google sourcing chatter and the financing structure. Sept 2 answers a lot.
Verdict: Everyone's long into the August 26 print — and the setup is fine. Benchmark raises July estimates to $92B rev / $2.10 EPS and nudges October to $103.957B / $2.37, basically pinning consensus. The stock isn't the question. The guide is.
The bull case writes itself: AI capex signals from peers and customers are broad, China is a conservative add, and at 33x earnings with a 0.3 PEG, the multiple is NOT the problem. NVDA at $5.2T market cap with this growth profile is still the cleanest mega-cap compounding story in the market. Benchmark's $335 PT sits near the top of the cluster — BMO $340, Stifel $282, Oppenheimer $265 — but the PT dispersion matters less than the direction: nobody's getting off the train.
The real debate isn't whether NVDA beats July. It's the October guide, the Vera Rubin standup cadence, and the gross-margin bridge. That's where the stock gets re-rated — or not. The July print itself is a formality.
"The October guidance, the first Vera Rubin system standups, and the gross-margin bridge should matter more than a modest July variance."
Softer tape risk into the print? Sure — buy-side positioning is crowded, and a "beat but nothing new" reaction is always possible. But with capex cycles still accelerating and NVDA trading at a PEG that screams cheap, the asymmetry is still to the upside. The trade is owning it into the print and letting the guide do the work.
FIX IS THE infrastructure bottleneck trade, and Davidson's initiation just confirms the setup. Firm starts Buy at $2,100 (24x 2027 EBITDA), but the real story is Q2: EPS $12.53 vs $10.42 bogey, revenue $3.26B — FIRST QUARTER EVER ABOVE $3B. This isn't a construction company anymore; it's a toll booth on AI capex.
The tape is crowded but momentum is undeniable. UBS raised PT to $2,225 after ORDERS PRINTED $4.9B vs ~$3.7B EXPECTED. Management pushed modular square footage target to 5M by mid-2027. Tech end market grew 105% YoY, and the company guides organic growth mid-to-high 30s for 2026. Every analyst reads the same tea leaves — peer-leading margins, differentiated modular offering, and a footprint perfectly positioned for hyperscaler spend.
"Peer-leading electrical and mechanical margins highlight strong execution and the ability to leverage significant technology customer spending increases within an advantageous geographic footprint." — D.A. Davidson
Bear case: stock up 142% over the past year and InvestingPro flags it as overvalued. At 24x EBITDA you're paying for perfection. But with orders running ~30% above estimates and backlog compounding, the bigger risk is multiple compression, not an earnings miss. The bottleneck trade stays intact.
KeyBanc bumps ZS to $210 from $185 (Overweight), and the street's closing ranks — Stifel and TD Cowen both at $200, Cantor at $225. This isn't a fundamentals inflection; it's a valuation call meeting an improving spend backdrop. The stock trades 6.6x revenues versus 10.7x for cyber peers growing >10%. That discount compresses if the October analyst day lands well.
Checks are steady QoQ, but quality is mixed. Platform vendor feedback, competition, and Red Canary chatter all came back lukewarm. Bigger watch item: partners flagging deterioration in ZS's go-to-market alignment with CrowdStrike sellers in the field. No one knows what's causing it yet — worth monitoring into the print.
"Zscaler's discounted valuation versus peers with more than 10% cyber growth at 6.6 times revenues versus 10.7 times revenues."
Bull case is clean: SASE leadership, 77% gross margins, 25% revenue growth, and segmentation tailwind post-Mythos where ZS's cloud security is potentially uniquely differentiated. FY27 already soft-guided, so downside expectations are anchored. Options market shows conviction — 39,437 contracts traded with calls running 7:1 over puts. Short-term squeeze potential into earnings, but the real catalyst is the messaging reset in October. Good r/r at this valuation, just don't confuse a multiple re-rate with an operating acceleration.
Cantor stays Overweight and $220 into the print, but their own channel checks undercut the bull case. Demand is flat. Not deteriorating — just not accelerating. WDAY at $197 needs a story, and "consistent with recent quarters" doesn't move PMs.
Q2 FY27 tracking in line with recent trends. No acceleration. Partners' CY26 expectations unchanged at midyear — slightly more optimism for H2 than negativity, which is the most tepid endorsement imaginable. AI adoption remains slow, with enterprises prioritizing data infrastructure modernization over Workday's AI hook. That's the crux: WDAY's growth algorithm depends on AI cross-sell, and the channel says customers aren't spending there yet.
Quality metrics hold up — 76% GROSS MARGINS, 13% LTM REVENUE GROWTH, two-thirds of the Fortune 500 on platform. But the rate of change is sideways and the multiple rewards acceleration. Cantor's note basically reads "good company, no catalyst."
Street's a mess ahead of the print. PTs cluster $215–275 with ratings from Hold to Buy. Deutsche Bank just downgraded after the run. Guggenheim's the lone $275 bull. Barclays thinks the print won't move the stock. Wells Fargo says PE deal chatter matters more than quarterly results. That's the tell — when private equity rumors are the near-term driver, fundamentals are in the penalty box.
"Demand has not accelerated or shown meaningful improvement."
That line from Cantor is the whole ballgame. Steady demand keeps a floor under the stock, but it doesn't get you upside. Watch the Q2 guide. If WDAY bends the AI curve or guides up, the $275 crowd wins. Otherwise this is a $197 stock with a $220 dream.
Needham keeps Buy and $49 PT into Monday's F4Q print — shares at $37.77, roughly 30% upside. That's the entire analyst voice this morning, one rating, clean. Nothing contested.
The setup actually has some texture. First report since Kevin Buchel took over as CEO from founder Dick Soloway on July 8 (Soloway stays on as Executive Chairman — so this is a transition, not a coup). Buchel's already hired a Chief Revenue Officer, and he's flagged M&A consideration. Balance sheet is clean: MORE CASH THAN DEBT against a $1.35B market cap. That gives them dry powder for the deal talk to become real.
The bullish hook, though, is channel evidence. Needham's read from distribution and industry players exiting June suggests the demand backdrop is healthy — enough to carry a positive narrative into the call even without huge guidance drama:
"Read-throughs from channel players and other industry participants exiting the June quarter suggest a strong demand environment."
The firm doesn't expect strategy shifts under the new CEO — steady as she goes, just with more senior hires and maybe an acquisition. Bear case is thin by default here (light coverage means no one's publishing the other side). The live risk is the call itself: whether Buchel's tone lands, and whether "considering M&A" becomes a bigger story than the quarter. Watch for any signal that deal talk is further along than the market assumes — that's the torque in this name short-term.
Verdict: operating business is fine, the headline miss is bookkeeping. Bernstein raises to $160 (from $150), Goldman to $168 (Conviction Buy), Macquarie trims to $143 — all Outperform-ish, all reading the same tape: margins, not top line.
Revenue beats ($30.11B vs. $29.48B). Operating profit beats by double digits. Then the noise: adjusted EPS $12.02 vs. $15.54 est. — a 22.65% miss. That miss lives BELOW the operating line: RMB 3.0B of investment losses on stakes in Alibaba and PDD. Not a franchise issue. Bernstein says those marks should have improved since quarter-end.
Fantasy Westward Journey and Where Winds Meet carried the quarter. Higher PC mix + lower channel fees = records. Games gross margin hit 76%.
"The firm highlighted the robust performance of existing game titles and noted a record-high games gross profit margin of 76% in the quarter."
Goldman's Lincoln Kong leans in harder: game sales +10%, operating profit +29% YoY. NetEase is a margin compounder, not a story stock.
$192M repurchased in Q2, up from Q1. Bernstein calls that constructive historically. Balance sheet backs it. At $128.67 with PTs clustering $143-$168, the r/r stays skewed long.
KeyBanc takes PT to $180 from $175, keeps Overweight — and the broader sell-side is stacking behind the same thesis. Identity is the chokepoint for AI agent security. Wells Fargo upgrades to OW ($180), Citizens to Market Outperform, Cantor to $170, Guggenheim to $162. The collective view: Okta owns the namespace.
Don't fool yourself on the near-term numbers though. KeyBanc's customer checks say AI agent identity architecture is still being drafted — not deployed. That means AI agents don't show up in cRPO yet. Market sits at 13.5% YoY cRPO growth for FQ2; KeyBanc expects FQ3 guide above 11%. Stifel models ~2% revenue upside. This is a multiple expansion trade on narrative, not an inflection in the operating model. TTM revenue growth ~12%, gross margins 77% — the base business is solid, just not accelerating.
"We remain highly constructive on Okta's position to be a leader in securing AI agents; however, high expectations, along with our customer conversations, suggest it remains quite early in formulating and implementing an identity architecture for AI agents."
Stifel keeps Buy and $141 PT — the entire thesis runs through data centers. XYL trades $113.44, so that's ~24% upside to the target. The market still isn't pricing the growth optionality.
Company guides 300% DATA CENTER REVENUE GROWTH in 2026. Segment reaches ~2% of total sales by year-end — small base, huge rate of change. They sell pumps, valves, and tanks into the buildout. Adding assembly capacity without new buildings (capex-light). No bespoke work, which would crush margins. Data center margins ACCRETIVE to the whole book.
Stifel points at influent water treatment and recycling/reuse as the real long-term prize. Cooling is just the entry point.
"Xylem believes it has an unparalleled offering in water recycle and reuse, a view management did not dispute."
Mixed print. Q2 adjusted EPS $1.46 beat the $1.34 bogey. Revenue $2.3B missed the $2.35B estimate — sales still +1% YoY. The full-year profit raise matters more than the revenue miss. New CFO Andrea van der Berg starts in September. Dividend streak at 15 years gives the floor.
At $26.5B market cap and showing up on undervalued screens, the bull case is straightforward: 300% segment growth, accretive margins, limited capex. Bear case: 2% of sales is a rounding error and the water-cooling narrative has plenty of competition. Stifel thinks the Street underestimates the recycle/reuse optionality. At $113 with a $141 PT, they get paid to be right.
Benchmark says look through the ugly headline print — commercialization is the story. Reiterates Buy and $9 PT (110%+ UPSIDE from $4.28) after hosting CEO Scott Swann, CFO Cody Barnes, and GC David Ray for an investor session Thursday.
Q2 revenue +2% YoY to $5.1M — but that's noise from an outsized ROC Watch deployment rolling off. GROSS MARGINS NEARLY 80%. Net loss $0.8M ($0.04/sh) vs +$0.6M income a year ago.
Benchmark's message: DON'T ANCHOR ON THE 2%. The quarter shows the commercialization strategy taking hold — government procurement recovering, ROC ABIS commercializing, ROC Evidence monetizing, pilots converting into longer-duration deployments.
"The quarter provides evidence that the commercialization strategy is taking hold."
Not sure one quarter makes an inflection, but the trajectory reads better than the revenue line implies.
Execs extended lock-up agreements to February 23, 2027 — covering ~54% of total shares outstanding and 66% of the non-public float. Skin in the game, and it removes a near-term supply overhang. Doesn't change the fundamental story, but it's a positive tell from insiders.
Light coverage = light take. $4.28 stock, 80% GM, $9 PT. Not a crowded book name yet. The real tell is pilot-to-deployment conversion cadence — watch that, not the YoY revenue number.
APP'S SITTING ON THE 52-WEEK LOW AND PIPER JUST JABBED THE PRICE TARGET AGAIN. $325 from $385, Neutral maintained. Stock's at $308.77 — a stone's throw from $303.17 — DOWN 54% YTD. Piper's cutting the multiple to ~18x from 20x. That's not a trim, that's a reassessment of the growth algorithm.
Piper chopped topline AND lowered 2H26 margin expectations after management walked them through the e-commerce vs gaming revenue split. The Q2 print already missed revenue guidance by 30bps at the midpoint and EBITDA by a full 100bps. The market's asking whether the e-commerce story can re-accelerate fast enough to offset a maturing gaming base.
"The business remains high quality — an expectations reset could create a more positive near-term setup."
Nobody's hiding. BofA went Neutral (PT $400 from $430) on Consumer segment sustainability. Benchmark still Buy but PT $440 from $500. BTIG Buy but PT ripped to $408 from $574. Phillip Securities the last real bull standing at $610 — even they're trimming on AI model training costs. The cluster speaks: quality yes, near-term growth trajectory, not so much.
This is a show-me story now. 88% gross margins and 61% LTM revenue growth keep the bull case alive — if the reset resets expectations low enough, the next print could clear easily. But 18x with margins rolling over into 2H26 isn't a gift until e-commerce re-accelerates. Watch the August 27 expert call on mobile gaming health for a catalyst.
KeyBanc's checks are the strongest in their coverage universe — ALL BUT ONE PARTNER ABOVE PLAN, one at plan — and they only bump the PT to $240 from $234. That's the signal. The rate of change in channel momentum matters more than the PT math here. CRWD is the leading beneficiary of incremental cyber spend today, and moreso as budgets open up in 2027.
Stock sits at $190.34, down 12% on the week into the Aug 26 print. That's the setup — expectations partially reset after an 84% trailing year. The rest of the tape is mixed but not bearish: Stifel Buy $230, Cantor $250 (post-split), Scotiabank trimmed to $227, Guggenheim Neutral. The CTO leaving to start an AI fund is noise, but PMs will ask.
The Anthropic anecdote is the best part of the note:
"We believe this supports the case that security is mission-critical and the independent vendors differentiate with extensive telemetry and domain expertise, though does marginally raise questions on the long-term moats for security vendors."
Read: an AI lab with real engineering talent built its own EDR and still runs CrowdStrike for the telemetry. That's the moat argument validated in one data point — and the "marginally raises questions" caveat is the honest hedge on whether the big AI platform guys eventually compress the space. For now, mission-critical wins.
Mizuho's holding $320 while ORCL sits at $142 — a 2.3x gap that screams either the street's badly missing the federal health story or the PT hasn't repriced the broken OCI narrative. The VA contract expansion is real, but it's capacity, not booked revenue.
THE VA RAISED THE EHR MODERNIZATION CEILING ~$17B TO ~$27B, plus three optional one-year extensions through May 2031. Only 14 of 164 VA medical centers are live — that's the bull case in one line. Mizuho sees SaaS acceleration layering on top of OCI growth, and the numbers back the math: LTM revenue of $67.36B at +17%, with the street modeling 34% growth for FY27.
"The expanded VA opportunity could help accelerate Oracle SaaS revenue growth over the next several years, complementing growth already underway in Oracle Cloud Infrastructure."
Caveats matter. The $17B is ceiling capacity, not RPO. Task orders hit as deployments progress, which means this is a multi-year drip, not a step-function. The bear camp's active too — UBS trimmed to $245 on OCI concerns, and the NM gas pipeline feeding a data center slipped to Feb 2027. Power constraints stay the OCI bottleneck.
R/r's asymmetric here, but the tape's screaming that AI infra confidence is broken. Federal health is a slow-burn positive — not a catalyst for a stock 59% off its high.
Rosenblatt's not blinking. Reiterated Buy and $30 PT (26x 2028E EBITDA) after management sit-downs at their Tech Summit — this into a 43% TWO-MONTH DRAWDOWN that leaves WULF at $16.45, 45% BELOW THE $29.84 HIGH. Peers only fell 30% over the same stretch. That spread is the fear trade, not the fundamental one.
The thesis is simple: the 20-YEAR ANTHROPIC LEASE (which more than doubles contracted HPC revenue) is the whole ballgame, and the market is fixated on the financing gap and tenant concentration rather than the asset quality. Rosenblatt frames the overhang as solvable — "clear, near-term resolution paths" — not open-ended risk.
TeraWulf "continues to execute from a position of strength on power, capital structure, and tenant quality."
Caveats worth knowing: Q2 was genuinely horrid on the bottom line — EPS -$1.94 vs -$0.25 consensus, revenue $44.77M missing the $46.52M bogey — and S&P cut the outlook to STABLE FROM POSITIVE on financing-mix uncertainty for the Hawesville campus. So the market isn't wrong that there's a funding overhang with cash burn accelerating and short-term obligations exceeding liquid assets. But at 26x 2028E EBITDA with a $19B contracted tenant in hand, the r/r skews up if WULF closes the financing chapter. That's a big if, but it's a binary, not a drift.
Truist reiterates Buy, PT $440 ($348.14 spot) — but the signal is the channel check, not the PT. The firm talked to a 5M-citizen state customer deploying MORE Tyler AI features. Government was always the last-off-the-block buyer. If Tyler's clearing that bar, the AI cross-sell thesis is live.
Q2 was clean: EPS $3.08 vs $3.07 est, revenue $645.1M (+8.2% YoY), record SaaS bookings and record FCF. Bookings +13% to $715M — that's the 3p that matters in a recurring model. Post-earnings PT resets cluster $435-500 (Baird, Piper, DA Davidson, Citizens). All point to AI traction during the quarter. 19 analysts also revised FY numbers upward.
Bull: defensible gov base + AI attach = re-rating. Bear: $14.3B cap on 8% growth already prices in the AI dream, and one customer conversation isn't a trend. Both fair. But with the stock up strong over the last month and estimates grinding higher, momentum and fundamentals point the same direction.
Rosenblatt stays Buy ($51 PT) — the clean investigative slate is the real catalyst here, not the guidance bump. The board probe found NO EVIDENCE senior management knew about export-control diversion, and no unreliable financials. That kills the "accounting blow-up" tail risk that was compressing the multiple (P/E 11.3, PEG 0.11 — absurd if you trust the numbers).
The growth story remains intact: $60B in new orders from 20+ customers, and consensus FY27 gross margin estimates have REVVED UP 140BPS TO 10% since last week's print. Rosenblatt sits at 11% GM. Hodlers get upside from both earnings power and multiple re-rating.
"Found no evidence that current senior management knew of the alleged diversion scheme."
U.S. and Taiwan government probes still ongoing — that's the residual overhang, and why the stock hasn't already rerated. But with EPS estimates up $1.10 (FY27) and $1.60 (FY28) post-print, the street is slowly accepting the AI server story again. Bernstein ($42), Needham ($46), and Mizuho ($35) all tightened targets too — the divergence between Rosenblatt and the bears is GM conviction, not order visibility.
TSM — The AUO panel plant buy is the real tell: NT$30B conversion turns central Taiwan into a CPO ecosystem around A14. Bottleneck moves from CoWoS to system-level integration, and TSMC vertical-integrates ahead of demand. Foundry cycle broadens too — INDUSTRY RUN-RATE FROM ~$220B IN 2Q26 TO $245B+ IN 3Q26, utilization 85%→90%. Memory at 40-50% of AI server semiconductor cost is the next constraint, not an expense line.
MU — Supply fulfillment runs 50-66% of customer demand. That's structural shortage, not an inventory story. Sixteen strategic agreements cover ~20% of DRAM output and 1/3 of NAND, backed by $100B+ minimum revenue and $22B deposits. HBM4 36GB 12-high is in volume production for Vera Rubin; NAND revenue +99% QoQ. The base-die fight with Samsung is live — whoever wins base-die economics captures the custom HBM margin pool.
SK HYNIX (000660.KS) — HBM4 12-high in production, 16-high in qualification: >2TB/s, up to 48GB. Hybrid bonding arrives at 20-high, and SK hynix owns the earliest advanced-packaging roadmap. THE ~40T KRW BUYBACK LIFTS SHAREHOLDER RETURN ABOVE 50% — memory supercycle plus capital return is a re-rating combo. SoC design is the binding constraint; exited CXL, hiring to close the logic gap.
SAMSUNG (005930.KS) — 2026 PROFIT ALONE EQUALS 1.1-1.2X THE CUMULATIVE PROFIT OF THE PRIOR 20 YEARS. Quarterly profit expected to beat NVDA. HBM4 on D1c DRAM plus 4nm logic base die, with controller offload and zHBM potentially reclaiming 5-10% of xPU silicon. The 90-110T KRW return package is below market expectations — that explains the mixed tape. Publicly trashing Micron's base-die approach is the tell: custom HBM is now a competitive weapon.
KIOXIA (285A.T) — Top 5 NAND revenue +77% QoQ, Kioxia part of that print. Dividend initiation coming this fiscal year. The molybdenum transition in 3D NAND improves leakage and RC. NAND cycle inflecting.
ASML — The Match Act would ban DUV sales to China, including currently permitted legacy DUV. That turns a manageable geopolitical overhang into a direct revenue hit on the China install base and service stream. Largest binary in the whole semicap complex.
AMAT — Taiwan OSAT capex of NT$460B is direct order flow for deposition and etch. TSMC packaging conversions add another demand layer. AMAT is levered to advanced packaging materials and interconnect, not just logic. Broadening semicap cycle.
TER — ASE and Taiwan OSATs are spending NT$460B on test capacity. AI packaging adds more test content per device. Prime tester beneficiary; watch order conversion through 2027.
BESI — Hybrid bonding lands at 20-high HBM. SK hynix cites ~24% greater core-die thickness and sub-18µm pitch. Purest play on the 2027-28 bonder ramp. This is a catalyst, not next quarter.
ADVANTEST (ATEYY) — HBM4 qualification and 16-high stacks drive test complexity. Taiwan OSAT capex adds handler and test demand. Memory test share leader gets the nod.
CAMT — CoPoS inspection demand runs ~2x CoWoS. V5 is on TSMC's initial vendor list, but Camtek is the broader inspection play. AI packaging inspections become the bottleneck.
NVMI — Advanced packaging yield management expands the served market. AUO conversion plus OSAT capex. High-margin inspection niche.
LRCX — Altus Halo is the only confirmed high-volume molybdenum ALD platform. Micron has moly in mass production NAND; Korea and Singapore fabs are early adopters. >50% resistance improvement vs tungsten. Multi-year layer-share ramp, not a one-quarter kicker.
ENTG — This is where Mo-in-NAND value lands. High-purity MoO2Cl2 and solid-source delivery for logic, DRAM, and 3D NAND. Direct materials beneficiary if wordline adoption broadens.
SUMCO (3436.T) — FIRST WAFER PRICE HIKE IN 3+ YEARS: +10% ACROSS ALL SIZES. End-demand finally soaked up supply. Leading indicator that materials capacity is tight. Margins inflect.
SHIN-ETSU (4063.T) — Part of the same across-the-board +10% hike covering 6/8/12-inch. Largest producer confirming materials tightness. Flows straight into wafer revenue.
GLOBALWAFERS (6488.TWO) — Direct beneficiary of the +10% hike. AI demand absorbed materials supply. Margin acceleration now in the numbers.
MURATA (6981.T) — MLCC tightness is literally a Hot Chips joke — Murata balloon in the goodie bag. Passives tightening into AI builds. Price-increase precursor.
ASE (3711.TW) — NT$335B of the NT$460B Taiwan OSAT capex wave. Problem is execution speed, not demand: 13 new projects, 8 renovations. That kind of constraint creates pricing power. Advanced test and assembly is the new bottleneck.
V5 TECHNOLOGIES (7822.TW) — On TSMC's initial CoPoS vendor list. 2025 revenue +188%, EPS NT$14.04. CoPoS inspection ~2x CoWoS. Qualification Q4 2026, volume 2027. Pure-play packaging inspection with improving r/r.
LARGAN (3008.TW) — Into NVDA's optical supply chain for Rubin Ultra with FAs, MLAs, and integrated FAUs. Mass production 2H27. The big number is Spectrum-X: ~20,000 units this year requiring 600-700k FAUs. New CPO content vector.
COHR — NVDA TOOK A $2B EQUITY STAKE, filed and completed. Balance-sheet endorsement and supply-security signal. NVDA sees optics as scarcity. Capital plus a demand backstop.
LITE — Same $2B NVDA stake, filed and completed. Directly tied to NVDA's CPO and optical roadmaps. Order flow protected by the balance-sheet link.
GLW — NVDA paid $500M for warrants, up to $3.2B total if exercised. Broadest passive optical play. Warrant structure gives NVDA upside without immediate equity dilution.
INNOLIGHT (300308.SZ) — FCC covered-list risk fading fast: ITI formally opposes including optical transceivers, with Amazon/Google/Microsoft/NVDA as members. Multiple key customers CONVERTED 2027 GUIDANCE INTO ACTUAL ORDERS THIS MONTH. 1.6T demand broadening beyond CSPs to new cloud and model companies. NPO delivery as early as 2027, with Innolight positioned as first supplier.
EOPTOLINK (300502.SZ) — Same FCC de-risking applies. Chinese sell-side pounding the table. One of two leaders in the 1.6T broadening. Policy overhang fading for the whole complex.
MSFT — First to deploy Vera Rubin at scale, announced with NVDA before earnings. Raised memory spend by $20B. Buying the compute and memory for the next platform. That cements the cycle transition.
GOOGL — Frozen V2 TPU targets 6-10x token throughput by hardwiring Gemini inference into silicon. Direct answer to NVDA pricing power. Campus land underwritten at 1GW Rubin-scale density, but 71% local opposition is a real siting headwind.
AMZN — AWS GROWTH ACCELERATED TO 36.7%, fastest in 18 quarters, beating the 31% street. But FCF turned negative in Q2. Hyperscaler capex is the demand story and the cash flow risk.
CRWV — Hudson River Trading signed a multi-year, multi-billion dollar deal for Vera Rubin access. Neocloud demand pulling frontier accelerators. Validates CoreWeave as scale AI cloud, not a GPU reseller.
AAPL — Pulled the 256GB config on M3 Ultra Mac Studio, leaving 96GB as max. That's a direct DRAM shortage symptom. AI memory crunch hitting consumer SKUs. Apple can't get enough high-density DRAM.
INTC — SK hynix showed EMIB in HBM packaging slides. Intel's packaging architecture inside AI accelerators. Speculation on a return to DRAM manufacturing or CPU+DRAM integration. Packaging roadmap angle even as CPU share erodes.
QCOM — Modular acquisition is a developer-toolchain AI-software play. Modest financially, but gives Qualcomm AI stack plumbing. Battle shifting to toolchains and on-device performance.
AMD — Taalas embeds model weights into silicon. Radical, but ties the chip to specific models. One attendee says AMD needs 16-high HBM4 to compete. Product cycle risk.
T — Cut AI coding costs 56% by routing to cheaper models with only 2% quality decline. Enterprise model routing. Bearish for frontier model pricing, bullish for inference volume.
ROP — CEO says not every AI task needs frontier models; caching and older generations cut cost. Token-cost deflation narrative. Warning for frontier model pricing power.
FCX — Molybdenum in NAND is technically validated but financially immaterial. Even a 20x upside case is <0.7% of global moly demand. Miners don't re-rate from this. Value accrues to precursors and equipment, not copper miners.
SCCO — Q2 moly prices +43% Y/Y but production -11% on ore grades. Pricing story, not volume. NAND wordline adoption adds immaterial demand. No structural rerating.
UBER — Fell 8% on Waymo fears, but bookings still +20%+. Trades at a steep discount to DoorDash. Market pricing robotaxi disruption ahead of actual revenue impact.
WMT — Same-store sales missed for the first time in five years. Apparel grew seven straight quarters with a new fashion brand launch. Consumer health is mixed. Read-through to discretionary spend.
DIS — Record revenue on Toy Story 5 and Spider-Man: Brand New Day. New CEO cut spousal health coverage for 200k+ employees. Content machine works; cost culture changing.
TSLA — Optimus bull case is self-driving world model plus manufacturing supply chain. Missing piece: actuator and joint supply chain for hands. China's humanoid showing at the World Humanoid Robot Games is the threat. Physical AI is a hardware supply-chain contest.