Good morning.
Futures are bouncing, Europe opened green — but let's be clear-eyed: this looks like mechanical rate relief off Williams' dovishness, not new buy-side conviction. Yields backing off the 30Y's near-25-year highs, oil still sliding on Gulf exports back to 23.3M bpd, gold up — that's a duration bid, not risk-on. The fly in the ointment: the DOLLAR just posted its STRONGEST MONTHLY GAIN SINCE JUNE, and nobody in the AI complex is pricing the second-order liquidity drain yet. Tuesday's "disturbing day" (Goldman's phrase — flat index, internals in a bear market) hasn't been repaired, just papered over.
MU IS THE WHOLE SHOW TONIGHT. The shortage evidence chain got four new links in the last 24 hours — Phison's CEO saying the gap takes "many years" to close, SK Hynix OUTSOURCING DDR4/DDR5 packaging to Winpac again (own lines full), Samsung foundry losses narrowing 42% on HBM4 base-die pull, and Netlist picking this exact moment to sue MU/NVDA/AVGO/GOOGL on HBM patents. But positioning matters as much as fundamentals here: the $370B rebound has already underwritten HBM ASP doubling. Clean beat doesn't cut it — MU needs a RAISE, and the two numbers that matter are HBM4 pricing language and the 2027 bit supply guide (management talking supply ramp is how shortage longs get killed by their own thesis).
Overnight Asia read-throughs were friendly — Taiwan AI-chain names firm, Samsung/Hynix catching bids on the HBM4 foundry story, though the China tape is still two-speed (ByteDance at ~20% of domestic AI DC capacity while the auto chain preps 2H26 inventory cuts). Oil down, yields down, MU printing into the friendliest tape it's seen in 48 hours — if AI can't hold THIS window, that tells you more than the red would.
Theme one: the bubble debate went institutional. MSCI launched an AI supply chain index with "hedge AI bubble" as the literal marketing pitch, and BOE warned of a correction SHARPER THAN JULY — with a rates transmission chain (AI productivity miss → debt sustainability repricing → duration pain) nobody has priced. Key read: hedging tools make short expression cheap, so the next AI correction is dispersion INSIDE the supply chain, not a beta flush. Long memory/short optical type books. Position accordingly.
Theme two: the paper market vs. the physical market. Anthropic's IPO leak — $40B+ annual losses, 47% of revenue routed through Amazon/Google (up from 32%, up from 11%), a $2T ask — dropped the same day NVDA wrote $668M into GMI Cloud and Amazon took equity in Gold Circuit. Circular financing is now the bear case's mechanism, not its vibes. But it cuts both ways: equity-for-capacity-locking is ALSO what real shortage looks like. Both can be true; the arbiter is whether these suppliers' non-sponsor revenue is growing.
Theme three: agent liability went from regulatory to litigable. First rogue-AI lawsuit against OpenAI — causation framework TBD, but "vendor indemnity" just became a standard line item in enterprise agent procurement. That's the CRWD/PANW/S governance trade moving from theme toward orders, and it pressures app-layer deployment pace at the margin.
Theme four: existing DC assets keep getting scarcer. Raskin's probe into secret hyperscaler-DC deals, community rev-share demands, Russia targeting Kyiv data centers, water stress — the friction raises marginal project costs (post-Oracle force majeure, credit is already repricing) while EQIX/DLR-type incumbent scarcity premium strengthens.
We'll hit up MU, NVDA and the memory complex first, then get to security (CRWD/PANW/S) and the DC names...
CEO bombshell Monday, analyst day Tuesday, and the Street basically shrugged and said "we still like it." That's the setup. MDB -21% on the week after CJ Desai walked out the door less than a year into the job to go run enterprise platforms at Meta (reporting directly to Zuck, so at least it's not a competitor poach). Dev Ittycheria — the guy who ran the show from 2014 to last November — is back as interim. Stock at ~$328, -16% YTD vs Russell 3000 +11%, but still +38% over six months. Context matters: this is a pullback in an uptrend, not a broken chart.
Nobody flinched. SIX firms reiterated Buys/Outperforms through the chaos — PTs cluster at $460-465 (Mizuho, DA Davidson, Stifel), with JMP at $519, Evercore at $525, and Guggenheim the high man at $560. That's 40-70% upside from here against a stock trading at 8X EV/CY27 SALES with 25% LTM revenue growth. The collective thesis: Desai built real enterprise relationships and bigger/longer commitments, but the secular story (data growth, modernization) doesn't live or die with one exec.
The analyst day itself gave them ammo. New AI ARR data points and — the big one — Atlas mid-term guide RAISED TO MID-20%s. Evercore's read captures where the smart money sits:
"The risk/reward remains attractive for investors taking a six-month view given steady revenue growth and margin expansion... any material re-rating is likely on hold until a permanent CEO is appointed."
That last line is the whole trade.
Bull: You're buying a 25% grower at 8x forward sales with a freshly raised Atlas bogey, the original founder-CEO steadying the ship, and a stock that's already puked 21% on a headline that doesn't touch the product or the customer base. Ittycheria back in the seat removes execution risk during the search — he built this thing.
Bear: Two CEOs in under a year is a pattern, not an event. Desai was the enterprise-relationship guy and he chose to leave — days before investor day, "effective immediately." And per Evercore, there's still NO idiosyncratic AI narrative here — it's a data modernization story wearing AI adjacency. Re-rating stays gated until a permanent hire lands, and every quarter of limbo is a quarter the multiple stays compressed.
My read: This is a positioning trade, not a fundamentals trade. The 6-month chart (+38%) says the fundamental story was working before Monday's headline. If you believe Atlas mid-20s is real (and they just raised it with a straight face while their CEO walked), $460-525 PTs against $328 is a fat r/r. But size it knowing you're capped until the CEO announcement — this drifts, it doesn't rip, until then. Not sure we can read too much into the "Most Undervalued list" stuff, but the setup is clean: wash out the weak hands on the leadership scare, own the raise.
The tape has already voted — now the analysts are chasing it. NTAP trades at $209, up ~98% YTD, sitting essentially ON its 52-week high ($210.04) and above nearly every PT on the street. BofA is the outlier, bumping to $220 from $200, but even they stay Neutral. Wells holds $190 Equal Weight. Evercore holds $210 In Line. GS stays Buy at $210. Nobody wants to pound the table at these levels, and after a near-double, can't really blame them.
All three desks hit the same event in Vegas (Kurian presenting, CFO Jabre on Q&A) and walked away with the same story: Novus is the headline — a software-defined architecture on ONTAP that separates metadata from the data path, targets 100+ TB/s aggregate throughput, and goes after AI factories, neoclouds, and enterprise. Management put a $40B TAM on it in three years vs. $20B today. AIDE (the AI data engine) expands across ONTAP, StorageGRID, and even non-NetApp storage — and the metadata layer is FREE for the first six months. Adoption before monetization. No financial update given at the event; long-term framework promised "at a later time."
The color that matters for positioning: customers are moving from experimentation to production, deployments are scaling from single-digit PBs to 100+ PB, and sales cycles are compressing from MONTHS TO WEEKS. That's the rate-of-change data point. BofA moved its multiple from 19x to 21x CY27 EPS specifically on this:
"Citing incremental TAM from AI that can support higher growth."
Management also reiterated opex growth won't exceed half of revenue growth and said the new offerings won't be margin dilutive (software attach). That's the discipline story that got the stock here in the first place — Q1 printed EPS $2.58 vs. $2.11-2.12 bogey, revenue $2.03B +29.9% y/y, gross margin 70.6% at the top of guide. 17 analysts have revised estimates up. The momentum is real.
Bull: NTAP is morphing from legacy storage into an AI data infrastructure platform with an expanding TAM, compressing sales cycles, and production-scale deployments — plus the Oracle Cloud add to the ecosystem. If the $40B Novus TAM is even half right, the multiple re-rates again.
Bear: The stock already prices the story — it's trading above GS, Evercore, and Wells' targets, and the $40B TAM is a management number three years out. The free six-month AIDE layer delays the monetization proof, and Novus walks into a knife fight: Dell's Project Lightning, Pure //EXA, DDN, Vast, Weka all contesting the same pNFS/AI data layer.
Good company, crowded trade. The fundamental setup (production inflection, margin discipline, estimate revisions all pointing one way) argues you hold it if you own it. But initiating at $209 with the street's PT cluster at $190-220 is paying for the re-rate that's already happened. R/r favors waiting for either a pullback or an actual financial framework for Novus — management teased one "at a later time," and that's the next real catalyst. Not sure we can read much into a PT bump to $220 when the analyst still won't go Buy.
The hurricane-play obituary keeps getting written and the stock keeps not caring. GNRC +56% YTD at ~$212, and the Amazon award ($2.4B initial orders, up to $8B total over time, second hyperscaler on the roster) has forced the Street to re-underwrite the C&I story in real time. Yet the stock still trades at a 28% discount to where Jefferies thinks it should. That's the whole debate: is this a data center power name at a discount, or an execution story the market is right to haircut?
Mostly one-way traffic. UBS at $340 is the high-water mark, Needham $283, Stifel nudged to $290 from $285 this morning on a higher 2027 EBITDA bogey ($1,174M vs $1,150M prior — higher C&I revenue post-Amazon). Barclays sits Equalweight at $278 (the only real skeptic on the sheet, and even their number is ~30% above the tape). Jefferies TRIMMED to $279 from $286 but stayed Buy — a trim on execution risk, not thesis.
Stifel's math: 15-16x revised 2027 EBITDA, with a kicker that timing of initial Amazon shipments could push numbers higher. Their broader point is the one that matters for the multiple:
"The rapid commercial and industrial ramp should erode the 'hurricane play' perception some investors hold on Generac and underpin multiple expansion."
That's the re-rating argument in one line. Residential outage demand is lumpy and weather-dependent; hyperscaler backlog is contracted, visible, and gets a different multiple.
Backlog math is getting serious. $2.4B Amazon initial orders stack on top of ~$1.6B data center backlog exiting Q2, and management plans to TRIPLE data center capacity to $3.75B/yr by 2H27. Jefferies adds a share-gain angle: rivals face more capital-heavy capacity builds with less theme leverage, and extending delivery times industry-wide are pushing orders GNRC's way.
One wrinkle worth flagging: the Amazon deal includes a seven-year warrant for up to ~1.69M shares at $200.93 — roughly 2.6% dilution if fully exercised. AMZN gets upside participation; GNRC gets a locked-in anchor customer. Fair trade, but it caps a bit of the torque.
Bull: Contracted hyperscaler backlog + capacity tripling + competitors stumbling on delivery = multi-year C&I revenue visibility the market hasn't priced. At ~$212 vs a $278-340 PT cluster, even the bear-adjacent Barclays number implies big upside. Re-rating from "hurricane trade" to "AI power infrastructure" is the multiple unlock.
Bear: The 28% discount to Jefferies' estimates exists because the market doesn't trust the ramp — tripling capacity in ~18 months is exactly the kind of plan that slips. Add genuine sensitivity to December bulk power executive order clarification, and there's a real binary-ish regulatory overhang near-term. Beta of 1.9 means you feel every wobble.
We're inclined to side with the bulls but respect the execution discount. The r/r at $212 vs a $283ish consensus-ish cluster is attractive IF the Amazon shipments land on schedule — that's the catalyst path to watch into 2027. Position sizing should respect the December EO wildcard; that's the kind of headline risk that gaps this name 10% either way. Not a table-pounder today, but the narrative shift from weather lottery to contracted AI backlog is real and still early.
The AI compute story is now the stock. Starship reaching orbit matters, but Needham's $54B ANNUALIZED RUN RATE on announced AI deals is the number that re-rates this name — if it shows up.
Two threads this morning. First, the space milestone: Flight 14 put the Starship upper stage INTO ORBIT for the first time (third V3 flight), deployed 26 operational Starlink V3 sats, booster splashed down clean. Not flawless — THREE ENGINE FAILURES (two of 33 on the booster, one on the ship, which briefly put the orbit goal at risk) — but the FAA shrugged and said no investigation needed. Second, and more important for the multiple: Needham talked to IR and got timing clarity on the AI compute contracts. Majority ramp in 4Q26, one starting in December.
Street is uniformly long. PTs cluster $200-250 — Bernstein $248 (Outperform), Mizuho $200, BofA $235, UBS $210, TD Cowen initiated at $200, Clear Street $217, Needham $250. Full range runs $140-450. Stock at ~$149, so you're looking at 35-70% upside to the base case cluster.
This is the crux. From Needham:
"The deals are projected to lead to a combined annualized revenue run rate of approximately $54 billion for only those announced AI compute deals. This supports management's commentary for targeting total company annual recurring revenue of approximately $100 billion by the end of 2026."
Context: LTM revenue is $23B. Consensus has FY26 at +144% growth. So the bogey is effectively a 4x on the revenue base in a year, and it's back-half loaded — 4Q26 heavy.
Worth flagging: Needham quietly CUT its 2H26 and CY27 AI revenue estimates in the same note to reflect "a more gradual ramp" and "lumpy renewals," and still kept Buy/$250. That's the tell. Even the bulls don't fully underwrite management's cadence. UBS, meanwhile, has 3Q revs at $13.8B (+7% vs Street $12.9B) and adj. EBITDA of $7.5B (+9% vs consensus) — so near-term prints look fine; the debate is slope, not direction.
Bull: V3 Starship to orbit unlocks the Starlink V3 cadence and the orbital data center roadmap, while $54B of contracted AI compute transforms SPCX from launch-and-satellite story into an infrastructure AI play with $100B ARR visibility. Milestone + revenue inflection landing in the same quarter.
Bear: The $100B is management commentary, not guidance, and the deals ramp lumpy and back-loaded — Needham already walked its own numbers toward consensus. Meanwhile the rocket that will fly this cadence just lost three engines on flight 14 and got a pass because the failures were tolerable. Execution risk on both legs of the thesis.
We're constructive but size it as an execution story, not a de-risked one. The 4Q26 compute ramp is the next real catalyst — prints between now and then are noise vs. that bogey. Watch for any renewal slippage chatter; that's where the bear case bites first.
StoneX stays Buy, and the thesis is the one that matters for MSFT right now: in the agentic enterprise, the plumbing beats the models. The firm lumps MSFT with MDB, SNOW, PLTR, and ORCL as names converging on the same architectural conclusion — AI EXPANDS the value of memory, retrieval, governance, and orchestration layers rather than compressing it. MSFT's seat at the table is the connective tissue between apps, workflows, agents, and enterprise systems. That's the moat argument, and it's a good one.
"The infrastructure supporting agentic AI and enterprise SaaS applications may ultimately become more valuable than the frontier models themselves."
Numbers check: $3.78T market cap, 28.4x P/E, revenue +17.8%, stock +38% over six months. 17 analysts have revised earnings UP into the next period — broad estimate momentum, not one loud bull. Stifel moved to Buy on the AI growth outlook and Citizens reiterated Outperform post-Copilot refresh (Home, Code, Autopilot, Today), so the Street is coalescing around the same story from different angles.
Noise on the tape: Xbox cutting 268 heads as studios consolidate under Activision (minor, cost discipline reads fine), and Nadella at the Xi state dinner (watch China headlines, no read-through yet). Nothing here changes the positioning case. Consensus long, estimates rising, multiple not cheap but not stupid. Own it.
Buy-side debate here isn't fundamentals — it's whether AI kills CAD. Baird reiterated OUTPERFORM / $325 (stock at $203, so ~60% upside to that bogey), leaning on the World Labs news — AMD's acquisition of World Labs is a win for ADSK, which invested back in February and keeps the collaboration intact. Arcadis expanding its AI work on the Autodesk platform adds a real-world adoption data point.
The broader desk view splits on one question: does AI disrupt the core CAD franchise or feed it? UBS ($325) and Guggenheim ($283) say moat intact — UBS did the channel work at industry events and found nothing. The strongest line:
UBS reiterated a Buy rating with a $325 price target, citing limited evidence of AI disruption to Autodesk's core computer-aided design systems after conducting extensive discussions at industry events.
Piper is the cautious voice — cut PT to $300 from $338 specifically on "anticipated debates around AI disruption" post-OpenAI's Astra release. GS stays Neutral at $265. So the cluster is $265-325 with the bears arguing narrative risk, not numbers.
R/r reads attractive if you believe the moat holds — PEG of 0.44 per the source, buys clustering $283+. But the OpenAI/Astra overhang means this trades on AI headlines, not prints, for now. Not sure the World Labs ROI moves the needle on the core debate — it's optionality, not thesis.
The street doesn't believe the math on Anthropic, and honestly the math doesn't help itself. DA Davidson sat down with IR and the Treasurer to walk through returns on the $11.6B, seven-year deal — Akamai's largest ever, with potential to scale to ~$20B — and came away reiterating Buy with a $150 PT (already cut from $185 on capital concerns). The bogey: ~15% IRR, NPV accretive, versus a current ROIC of just 4% over the LTM. On paper that's a massive upgrade in capital efficiency. The catch is the reinvestment risk — cash flows in 2028 and beyond need to keep earning that 15%, and that depends entirely on how the compute demand environment evolves.
Stock at $107.52 tells you what investors think. DA Davidson flags it directly:
"Investor conversations indicate disappointment on the economics of the deal."
The rest of the street is split between validation and skepticism. Raymond James (Outperform, $170) sees the confusion as the opportunity — market doesn't get the pricing/returns yet. RBC (Sector Perform, $135) credits the CIS expansion. TD Cowen nudged to $149 but stays at Hold, calling the deal validation of AI infra demand without endorsing the returns. Bulls get EPS lift by 2029 from CIS commitments; bears get a company levering up on compute at unproven returns. R/r leans on whether that 15% IRR survives contact with 2028 — not sure we can underwrite that from here.
StoneX staying at Hold even as META keeps shipping — Muse for Small Business dropped with a free tier plus hooks into Shopify, QuickBooks, Stripe and Slack, timed almost head-to-head with OpenAI's paid "Dots" launch at DevDay. StoneX's read: OpenAI will eventually cave with a free Dot tier of its own, and the real issue for META isn't the product, it's the math — Muse is infra-heavy, and that stacks on top of an ROIC debate already hanging over the name (23% today, against 81.75% gross margins and a $1.82T market cap).
The bullish counter in the same tape: BofA reiterated Buy on the new Meta Enterprise Platform unit, and Truist is flagging Muse's travel-booking chops as a genuine threat to OTAs — consumer distribution plus data is a hard combo to defend against. CJ Desai hire from MongoDB getting a positive stock reaction too. Net: product velocity is real, but the "how much capex for how much return" question is the bogey that keeps a desk like StoneX on the sidelines. Watch the free-vs-paid agent pricing war with OpenAI — that's where margin narrative gets set.
Rosenblatt raising PT to $360 (from an undisclosed prior) with AMZN at $246.67, and the thinking matters more than the number. The firm's read: the market is pricing agentic commerce as a BINARY THREAT to Amazon's ad economics — the idea that AI shopping assistants compress the discovery funnel and displace the retail media model entirely. They call that narrative false.
The pushback worth steelmanning on the bear side: retail media is the margin engine behind the whole retail P&L, and if agents own the top of funnel, Amazon loses the toll booth. Rosenblatt's counter is basically "they've navigated secular consumer shifts before" — which is true, but a bit vibes-based as defenses go. Still, ~46% implied upside on the new target says they put real estimate revisions behind it, not just a rerating.
"Concern is emerging that Amazon's retail media model is at risk of complete displacement... The firm called this narrative false."
Net: this is the right fight to be having on AMZN. If agentic disruption fear is what's capping the multiple here, that's a sentiment overhang, not a numbers problem — and those resolve faster.
Elsewhere in the tape: $3B into India quick-commerce by 2030, a $100M+ robotics plant in Indiana, and a UK class action (the 2018 Apple sales agreement one) allowed to proceed. Nothing thesis-changing, but the capex cadence keeps confirming they're spending through the AI cycle, not around it.
Workday cuts 2.5% of staff — mostly product and tech — plus select office space, and takes $65-80M in FY27 charges ($55-70M landing in Q3, split between severance, SBC, and lease impairments). Cantor holds Overweight, $205 PT, updating GAAP estimates only — non-GAAP untouched, and guidance reaffirmed ex the GAAP margin hit.
This is the margin-expansion playbook, not a demand signal. Management says it's still hiring in strategic areas through FY27, and 30 analysts have taken estimates UP into the print. Q2 backed the story: 31.1% adjusted op margin vs 30% consensus, driven by cost discipline and internal AI automation (doing the AI thing, not just selling it).
The bear case isn't the RIF — it's the guide. KeyBanc flagged a downtick in Q3 cRPO growth and soft initial FY28 subscription growth, which is why the stock trades on "beat but mixed" vibes despite the margin delivery. PT cluster sits $190-215 (DA Davidson, Freedom, KeyBanc, Cantor), which tells you the street sees upside but nobody's pounding the table.
Net: cost actions support the margin narrative, but until cRPO reaccelerates this is a show-me story. Modest upside, no urgency.
Oppenheimer upgrades to Outperform, PT $150 — and the pitch here is pure valuation + optionality, not a growth inflection call. Stock at $107 trades at 1.4X 2027 REVENUE and ~8X EPS, a 50%+ discount to its own historical median multiples, with a 10% FCF yield. That's the whole argument: you don't need the AI story to work perfectly at this price.
The SOTP kicker is what makes this interesting. Oppenheimer thinks Actimize sells for ~2x the consolidated multiple, Cognigy has already doubled in value to $2B+ since they bought it — which implies the legacy CX business is sitting at ~2X EBITDA. Meanwhile the actual numbers aren't broken: Q2 beat ($2.70 adj EPS vs $2.64 bogey, rev $782M), cloud +12.6%, and AI ARR +52% Y/Y TO $362M. RBC also lifted to $145 off CFO conversations around personal AI agents and completed strategic renewals. Guidance reiterated.
Bear case is obvious: stock fell on the Q2 beat, which tells you the market doesn't trust the growth durability — AI ARR ramping while total revenue grows low-double-digits means the core is decelerating underneath. Classic value-trap setup in legacy CX names. But at 8x earnings with a likely Actimize catalyst and a re-rating path to $150 on just 12x 2027 EPS, r/r leans long. Not a hero position, but the cheapest AI-monetization story in software right now.
SNOW UP 119% OVER SIX MONTHS, +50.6% YTD at $330 — and StoneX's Yi Fu Lee stays at Buy with a $400 PT after management talks centered on AI monetization. The stock's already run a lot. The question is whether the agentic-AI infrastructure narrative justifies the multiple, and the analyst case is that SNOW is the governance/orchestration layer of the enterprise AI stack — model-agnostic, sitting on top of whatever LLM the customer picks.
That's the emerging consensus framing across the data-infra complex: MDB owns memory/retrieval, SNOW owns governance/orchestration, MSFT wires the workflows together. Everyone converging on the same architectural conclusion — the infra under agentic AI gets MORE valuable, not less, even as model costs compress.
Fundamentals backing it: revenue +32% LTM to $5.43B, product revenue +37% y/y in Q2 with AI driving roughly half the growth acceleration. 39 analysts have taken estimates UP into the print. Fresh initiations keep coming — PhillipCapital at $423, Freedom Broker to $450.
"AI is expanding the importance of data, memory, governance, retrieval, and orchestration layers rather than compressing their value. The infrastructure supporting agentic AI may ultimately become more valuable than the frontier models themselves."
Goldman's flag: rising bond yields pressure long-duration software, and SNOW at $115B market cap after a 119% run is as long-duration as it gets. R/r at $330 vs a $400-450 PT cluster is decent but not fat. This is a positioning story now — everyone's long the same thesis. Works until it doesn't.
Truist keeps its Buy on, and the read-through runs through CoreWeave. The neocloud says it's comfortable with parent-level financing through 2026, has rising rates baked into forecasts, and structures contracts so backlog exceeds debt obligations with margin cushion ($51.6B total debt, 10.27 D/E — elevated, but they say the math works).
The NVDA tell: CoreWeave's 25% price increases across SKUs could add ~9pts of contribution margin because the hikes outpace NVDA's planned GPU price increases. That's pricing power on both sides of the transaction — NVDA raising prices, customers raising faster. Demand signal stays intact.
"CoreWeave has contemplated rising rates into its forecasts and structures contracts such that backlog exceeds debt obligations with additional margin cushion."
Not a new catalyst, but it quiets the "CoreWeave credit blowup takes NVDA orders with it" bear case for now. 115% revenue growth at the biggest neocloud customer is the kind of number that keeps the GPU shortage narrative alive. Not sure we can read much more than that off one Truist note, but the direction of travel is friendly.
Needham initiates BUY, $16 PT — that's ~160% upside from $6.14, with the stock sitting A DIME ABOVE its 52-week low after bleeding 18% in a week. Timing isn't accidental. They're buying the drawdown, not the momentum.
The thesis: Pasqal is the most commercially advanced neutral-atom platform out there — revenue TODAY from analog systems, with digital/fault-tolerant on the same architecture later. That's the key differentiator vs. the quantum crowd burning cash with nothing to sell pre-fault-tolerance. Needham flags "attractive neutral-atom scaling economics" and says liquidity covers execution (always the first question on these names).
The fundamentals back the commercial story more than the P&L: H1 revenue EUR 4.9M, +14% — tiny — but booked and awarded business of EUR 70.4M is the number that matters. That's the pipeline. Losses widened on SBC and listing costs, which is noise, not signal.
"One of the most commercially advanced neutral-atom quantum platforms" — revenue before full fault tolerance.
Bull: real deployments, real backlog, scaling economics better than trapped-ion/superconducting peers. Bear: $1.4B market cap on sub-EUR 5M half-year revenue is a lot of faith, and one initiation doesn't make a floor. R/R is interesting down here IF you believe the EUR 70M converts — but this is a position-sizing conversation, not a conviction one. Not sure we read too much into a single-shop initiation, but the setup (washed out, at the lows, first bull stepping in) is at least worth a look.
William Blair kicked off coverage with an OUTPERFORM and the pitch is simple: CRWV at 3.1x EV/SALES and 5.2x EV/EBITDA (CY27) trades at roughly HALF the peer median of 6.1x / 14.2x — for a company growing revenue 115% LTM. That gap is the whole thesis. The bears aren't wrong about the balance sheet (D/E of 10.27 is real), but the Street is increasingly arguing the market is pricing the financing risk and ignoring the earnings power once contracted capacity hits mature utilization.
"The market underestimates CoreWeave's ability to finance its growth efficiently."
That financing question is the stock. Management raised a $3.7B CONVERTIBLE plus a 35M share ATM offering and still tells the story of being funded through 2026 — dilutive, yes, but proactive rather than desperate. Meanwhile demand signals keep cooperating: 3Q capacity contracts reportedly clearing at ~$40M PER MEGAWATT, which supports JPM's recent upgrade on premium short-term pricing and Cantor's $176 PT. The bull case writes itself. The bear case is one AI capex pause away from a leverage problem. At half-peer multiple on a $46.9B cap, the market is pricing meaningful skepticism — but the rate of change in sentiment (multiple initiations/upgrades stacking up) says positioning is still early, not crowded. Watch the contract pricing prints; that's the bogey that matters.
GEV at $962 sits ABOVE Mizuho's $949 PT — and that pretty much tells you the whole story. The Neutral reiteration isn't a call on fundamentals, it's a call on valuation. The fundamentals look fine: Mizuho models book-to-bill at ~2.6x in Q3 and 2.8x in Q4 on 11GW and 16GW of new commitments, which puts 2H comfortably ahead of the conservative 20GW guide. Backlog ~$189B, tracking to $200B+ by early 2027. 2026 revenue at the high end ($46B+ vs $45.5-46.5B guide), EBITDA margin ~14% vs 12-14% guide.
Watch the mix though. Electrification orders dip q/q on lumpy data center timing (the $1.5B India HVDC project lands in Q4, not Q3), and wind stays soft without a utility procurement pickup. Mizuho trimmed 2026 EBITDA to $6.5B from $6.7B and 2027 to $9.3B, while nudging 2028 UP to $13B — they're pushing the earnings power out, not killing it.
The Street range here is wild: Baird at $1,400, Bernstein $1,298, Jefferies $1,185 on the bull side (services + electrification backlog), GLJ initiating at SELL with a $470 PT on cycle concerns. That's a 3x spread between top and bottom target — nobody agrees on what multiple this thing deserves at peak gas turbine demand. R/r at $962 with the bull cluster $200-440 above and GLJ $490 below basically says you're underwriting the AI power narrative holding through 2027. Mizuho's neutral stance with a PT below market is the honest read: story intact, price full.
Cantor stays constructive off Forge, and the setup here is interesting — stock at $23.15 is DOWN 38% from its 52-week high and off 10%+ in just the past week, yet the PT deck sits $30-50. Cantor reiterated OW at $38 after attending the annual Forge conference, calling the post-quarter negativity a "short-term view" of demand. That's a big gap between sentiment and fundamentals, and the conference read-throughs support the bulls.
The real news: Decisioning Studio Go (the OfferFit tech) is expanding beyond the premium Pro tier — and it consumes Action Credits rather than requiring new contact sign-ups, which lowers the friction to attach. Partners at the event flagged strong pipeline for the current quarter and beyond. Marketing teams buying engagement automation instead of building in-house remains the secular tailwind.
Context matters: the quarter itself was fine — 25% growth across US and international, revenue beat by ~3%, FY26 guide raised $15M at midpoint. Goldman, Stifel, BTIG, and Raymond James all kept Buy/Outperform ratings post-print, with GS pointing to AI product adoption and enterprise vendor consolidation. So this looks like a positioning/expectations washout, not a fundamentals break.
"The negative sentiment following the fiscal quarter is a short-term view of the demand environment."
R/r skews attractive at $23 if you believe the 27% LTM revenue growth holds, but we're working off one fresh note here — the rest of the Street love is stale, post-print commentary. Would want to see the stock stabilize before calling the bottom. (Worth watching whether Go converts to Action Credit consumption upside in Q3.)
Deutsche Bank kicks off coverage with a Buy and $200 PT, and the thesis is the one thing FORM bulls have been waiting for: a second act beyond HBM. The call hinges on FORM establishing itself as the SECOND SOURCE of probe cards for NVDA GPUs at TSMC — a TAM DB sizes at ~$500M, equal to FORM's entire 2026E logic probe card sales. First sales 2H26, path to ~20% share by 2028 (>$200M in sales). That's the incremental revenue stream the stock has been missing, and it explains why shares are +274% over the past year to ~$136.
"FormFactor has established itself as the second source of probe cards for Nvidia's GPUs at TSMC, opening up a total addressable market equal in dollar value to FormFactor's total logic probe card sales."
The rest of the model stacks up on existing strengths: HBM TAM more than doubling by 2028 on the vertical MEMS transition (DB still has FORM at $400M HBM sales by '28 even AFTER ceding ~40% share to Technoprobe), plus custom ASIC shifting from pseudo-MEMS to vertical MEMS — modeled at $200M by 2028, ~10x vs 2025. Photonics is the wildcard; FORM lags in optoelectronic probe cards and DB is merely "hopeful" on recent acquisitions.
Fundamentals are cooperating — Q2 beat with $0.82 vs $0.61 bogey on record $258M revenue, third straight quarterly high, and Cantor sticking with OW/$175 while dismissing despecing fears. Caveat: after a 274% run, you're paying up for execution on the NVDA second-source ramp that doesn't start until 2H26. The $200 PT implies the market pays for 2028 share today.
Deutsche Bank moved NFLX to Buy from Hold — even while CUTTING the PT to $95 from $100 and trimming op income/FCF estimates post-print. The call is pure multiple math: stock trades at 18X DB's 2027 EPS estimate versus ~40X forward at the June 2025 peak, and Kraft sees room to re-rate into the low-to-mid 20s on 23% earnings growth in 2027. PEG of 0.6 does some of the talking.
International time spent has increased year-over-year in each of the past four six-month periods... Netflix now produces more than 60% of its content outside the United States.
The bull steelman is that the market undervalues the international engine — 60%+ of content produced ex-US, a production moat rivals can't replicate, plus an AI leverage argument on content, UX, and ads (tech org vs. legacy entertainment streamers). Bear side is live though: Wells Fargo went to Underweight on viewership declines and a soft 2H26 content slate, and 39% of Americans cancelled at least one streaming sub in the past six months. Not sure we can read too much into one upgrade with a lower PT, but the valuation reset from 40x to 18x changes the r/r math — this is a positioning call as much as a fundamentals call.
FCEL is trading as a BE proxy today — that's the whole story. BE +10.9% on RBC reiterating Outperform/$335 PT (Fremont capacity expansion, Oracle reaffirming the 2.4 GW Project Jupiter), and FCEL caught the sector bid alongside it. Nothing company-specific in the tape. If you own FCEL, you're renting BE's demand narrative (AI datacenter power, behind-the-meter) at a fraction of the quality — BE at $87B market cap with S&P 500 inclusion, FCEL without the contracts or the margin story to match. Fine as a beta punt on the fuel-cell complex, but don't confuse the move with a signal on FCEL itself. If BE fades on the pipeline delay noise (pushed to early 2027), FCEL fades harder — that's just how high-beta sympathy works.
Buy the dip — that's DA Davidson's call, reiterating BUY with a $330 PT against a $184 print. MOD has bled ~11% over the past week (still +24% on the year), but the firm came off a virtual call with CEO Brinker and CFO Lucareli pounding the table on data center thermal demand. Key point: demand is broadening BEYOND NORTH AMERICA — global customer discussions now — and management is locking down critical component supply via multi-year LTAs to get ahead of the chain issues that bit them in FQ1 (revenue +28% to $874M but a slight miss vs. $891M bogey on production constraints; EPS $1.53 beat $1.38).
"an impressive and durable level of inbound data center thermal demand"
Yes, the 68x P/E is rich, and the Gentherm merger / Modexus rebrand / Performance Tech spin adds execution noise (KeyBanc already running pro forma models post-spin). But Hunterbrook's thesis that the DC cooling business is bigger than disclosed gives this pullback a real narrative floor. R/r skews constructive if supply chain actually gets fixed.
Texas Capital plants a flag on the neocloud pivot. Buy, $15 PT — ~95% upside from $7.72, though the stock's been bleeding (down 11.5% on the week, so someone out there disagrees). The thesis: RUM isn't a video platform anymore, it's an AI infra/services story, and the $13.7B SIX-YEAR GPU SERVICES DEAL (reportedly Anthropic, per press reports) validates the compute platform and fixes the revenue visibility problem that has dogged this name.
"Strong execution, Tether's strategic support, and the potential value of Rumble's content ecosystem for agentic AI differentiate the platform from its neocloud peers."
The asset math is the bull case: ~70MW available capacity today, development pipeline over 1GW. Bear side is real though — Q2 printed a $0.28 loss vs. $0.09 expected (revenue did beat at $40.4M, +61% y/y, record), acquisition costs are heavy, and tranche 3 of the megadeal is contingent on the customer approving delivery dates. Consensus has them flipping to $0.11 EPS this year — if that bogey holds, the AI infra re-rate narrative has legs. Light coverage, one desk, treat it as a data point not a consensus.
The e-commerce inflection isn't coming in 2026 — Wells Fargo's pixel tracking just took the legs out of the near-term bull case. Firm reiterates EQUAL WEIGHT, $325 PT vs. $305.66 spot, which is sitting NEAR THE 52-WEEK LOW of $297.50 and DOWN 57% over the past year (this is a broken chart, not a dip).
The data is the story. APP added ~220 e-comm pixels/week from June-August against an install base of just 6,000 sites — versus SNAP at 118K and PINS at 489K. Then pixel adds "inflected" to 750 and 1,600/week over the last two weeks. Sounds great. Except:
The majority of the inflection in pixels added over the last two weeks was driven by APAC sites with little to no traffic... The percentage of sites adding the pixel with no traffic began to climb in June, reaching 85% in the last two weeks.
So the acceleration is junk adds, not real advertisers. Worse, WF flags that APP pivoted away from buying e-comm customers via Meta/Google direct response in 1H26 and is rebuilding go-to-market through channel partners — a strategy they call "early," with no customer growth inflection likely BEFORE 2027.
The rest of the street has been quietly marking to market: PT cuts across the board — MS to $450 (Buy), Evercore to $510 (OP), Needham to $475, Piper to $325 (Neutral), Benchmark to $440 (Buy) — mostly on overcooked e-comm advertiser spend assumptions and a maturing games business. Note the cluster: even the bulls now sit $440-510 against a stock at $305.
R/R here: bears have the tape and the third-party data; bulls have valuation and the games cash engine. Not a short at the lows, but the e-comm second-act narrative needs a real pixel cohort — with traffic — before you pay up again. Watch holiday pixel quality, not quantity.
TD Cowen raised its PT to $550 from $500 — but kept the HOLD, which tells you most of what you need to know about where the easy money sits here. The $95B AI server backlog (mostly noncancelable, per Sankar after hosting IR head Paul Frantz last week) is the real number. That's visibility most hardware names would kill for, and it explains the 49% revenue growth over the LTM.
The interesting wrinkle: Sankar flags 2H26 orders as the tell for 2027 AI upside, and says higher CPU rack mix helps margins off that ~20% gross margin base. Memory constraints in 2027 are the bear case — and it's a real one given where DRAM pricing is heading across the complex.
"Dell holds $95 billion in AI server backlog, mostly noncancelable, with leverage to major neoclouds."
The quiet negative buried in the note: CSG (PCs) faces near AND mid-term headwinds as IT budgets rotate from client to infrastructure. That's the classic DELL trade — you're long the AI server story and short the legacy drag, which is why the stock keeps getting Hold ratings from people raising their targets. Street is constructive elsewhere too (Truist to $505, RBC initiating Outperform, Susquehanna positive on FY27 server growth), so positioning is crowded-long on the AI narrative. Watch the 2H26 order prints — that's the bogey now.
The r/r thesis is dead — that's the call. Barclays cut MBLY to Equalweight from Overweight and slashed PT to $9 from $14, admitting the upgrade earlier this year was a bet on advanced-content wins that never showed up. Stock DOWN 28% YTD vs SPX +12%, and now the CEO transition hangs over a name already at a strategic crossroads.
The frustration: near-term numbers are actually fine. Q2 REVENUE $508M vs $475M consensus, FY26 guided to ~$2.0B (a hair above the street), EyeQ units nudged up to 39M from 38M. NINETEEN analysts have revised earnings UP. None of it matters when the multiple can't re-rate without bigger proofpoints.
"Larger proofpoints for Mobileye's advanced content and leverage to the autonomous vehicle theme have not materialized."
The bear case per Barclays: no big advanced wins, China export pressure going international, and real questions around the VW opportunity. The bull case (Piper at $12 on robotaxi, Tigress at $18) needs exactly what isn't happening. Street PTs now scattered $8 to $18 — that's a name nobody can underwrite until the new CEO lands. We'd stay on the sidelines; cheap for a reason until the AV narrative has a body attached to it.
MU — Tonight's the print and the bar is the problem. $370B rebound already done, street models HBM ASP DOUBLING, so a clean beat does nothing — needs a raise on HBM4 pricing and 2027 bit supply guide. Shortage evidence chain keeps building (Phison "many years," SK Hynix outsourcing packaging, Samsung HBM4 base-die on 4nm) but positioning flipped long and r/r skews DOWN on anything less than a raise. Netlist suit pre-print is a nuisance overhang, not thesis.
GOOGL — Two stories. Anthropic routing 47% of sales through Amazon/Google is the bear case quantified — investor, supplier, channel, competitor all at once, and the S-1 revenue-quality footnotes are where this gets adjudicated. Offsetting: v9 TPU going 100% EMIB-T on Intel packaging with >5M unit life-cycle vs bear >2M says Google's own silicon ambitions are bigger than street models.
AVGO — Clean beneficiary of the Anthropic $518B buildout and still in every ASIC conversation (v10 Merope "should be AVGO" per one feed, Icefish to MediaTek). Pipeline hot. Netlist suit is noise.
ORCL — RPO went $450B → $660B in a quarter and the doomers are still writing obituaries. >$200B net new RPO is capacity reservation at locked prices — that's demand, not vapor. Project Jupiter reaffirmed 2.4 GW on schedule (gas pipeline slips to early '27 on permitting, whatever). Watch the next AI DC financing print: if a deal lands <10% after SoftBank's 9.75%, the credit execution-risk repricing was overdone.
CRWD / PANW / S — The OpenAI rogue-agent suit is the starting gun on the agent liability regime. Enterprise procurement adds vendor indemnity as a standard line item, agent governance spend follows. Theme → orders conversion is the thing to track; right now it's a narrative long, not a numbers long.
AAPL — Smart home launch Oct 13 is a beta story (first new category under Ternus). The real signal is A20 Pro dropping PoP for WMCM — memory and logic dice co-designed, TSMC packaging incremental, mobile DRAM starting to customize like HBM did. Watch if A21 follows. Offset: restructuring/job cuts and an on-device Siri strategy that looks wrong vs cloud agents. Modem moat stays QCOM's ($1B for Intel's modem, 6+ years on C1, still paying $2.5B+ royalties — brutal).
TSM — Compounding advanced packaging demand from every direction: A20 Pro WMCM, A14 flows certified, COUPE CPO flow, N2P tape-outs across PCIe7/HBM4/LPDDR6. MediaTek supply FUD rejected by the chain. One genuine offset: Google v9 going 100% Intel EMIB-T challenges CoWoS at the margin — single-supplier dependence gets nibbled, not broken.
EQIX / DLR — DC political friction expanding on four fronts at once (Raskin probe, community revenue share, war targeting, water). Secret agreements era ending. That raises marginal project cost but hands a scarcity premium to existing assets with power and permits. Own the incumbents.
ACN — Bloomberg's Julie Sweet AI-pivot feature is a good ticker for what's already priced. Low-cost-labor consulting is the highest-certainty AI casualty in the application layer. Next quarter's bookings split (AI-related vs traditional) is the leading read for the whole IT services complex.
ASE — $68M Singapore plant for AI chip testing. Small number, real signal: OSAT capex physically moving Taiwan → SEA. Geopolitical diversification premium becoming actual dollars.
GCE — Amazon taking a stake (NT$1.59B, ~0.36%, 20% discount) after the Alchip precedent. Aug revenue +76.4% YoY, supplies all four hyperscalers. Two readings: capacity lock (scarcity premium, likely how it trades near-term) or circular financing (discount). Watch non-Amazon customer revenue mix — that's the tell.
NLST — Patent troll with a winning track record vs memory makers, swinging at the tightest point of the HBM cycle. Overhang for MU into the print, noise for NVDA/AVGO/GOOGL. Pure monetization play.
MSCI — Launched an AI supply chain index explicitly marketed to "hedge AI bubble." This institutionalizes the hedge — you can now run short memory / long optical instead of naked short SOX. Next AI correction likely plays as intra-chain dispersion, not broad selloff. The tool's existence amplifies it.
INTC — The Google v9 EMIB-T win is the real thing: 100% share, yield ~50% (better than feared), >5M unit life-cycle, packaging revenue +$2-3B peaking $1-1.5B/yr '28-'29. Plus 18A capacity tracking >60K WSPM next year vs UBS modeling a third less. Foundry turnaround is getting underestimated by consensus. Two separate proofs in one week.
MRVL — Lost a DC sales VP to Intel. Minor. Celestial keeps it in Amazon's strategic orbit.
QCOM — Modem/IP moat confirmed by Apple's own wallet: $1B paid for Intel's modem, 6+ years on C1, still cutting $2.5B+ royalty checks. The annuity survives every in-house attempt.
CBRS — Lost GPT6.1 Sol Ultrafast to NVDA on capacity/timing, not architecture — but wafer-scale yield/RAS questions stay unanswered and the lockup expiry tomorrow is a near-term overhang. Institutional breadth real (Altimeter 16.2% of book, 7/12 elite funds initiated Q2). GB300 disagg optionality is the bull case; nothing to do here into the unlock.
AMD — Agentic CPU thesis quietly favoring x86 headnodes over ARM (NVDA Vera) — AMD "destroys" in RL/agent sandboxes per the feed. Market just starting to price this angle. Keep on radar.
CRDO — Pivoting uLED → uVCSEL validates the incumbent optics roadmap and undercuts the uLED challengers. Quietly bullish for the franchise.
OKLO — Lost the PJM queue fight on a flawed application. Nuclear/SMR-for-datacenter timelines carry real regulatory execution risk — this is the receipt.
BE — Leasing another 158K sq ft at Fremont (vs 164K existing) is a demand signal, not noise. RBC Outperform, $335 PT. Oracle Jupiter reaffirmation helps the fuel-cell DC narrative.
HOOD — Summit 2026 numbers are wild if real: 150K+ agentic trading accounts, agents hitting Robinhood tools ~30M times/day, plus 24/7 trading, perps, 4x intraday margin. Product velocity is the story; monetization of agent flow is the question.
DASH — The agentic-threat read is the interesting one: ~half of customers arrive not knowing what to order, and early agent traffic cut basket size by two-thirds. DASH refusing to outsource discovery to agents is the right strategic call but a real margin fight. Citizens' $250 PT looks generous against that headwind.
GRAB / GOTO — Grab at ~$12.6B is ~6x GOTO's $2.1B. Relative value observation only, no catalyst.
BA / RTX / LHX / HII — Defense prints: F/A-XX >$20B to BA, $20.7B AMRAAM to RTX, $6B THAAD propulsion to LHX, $5.1B carrier overhaul to HII. Budget flow intact.
TSLA — The disclosure that matters: SpaceX semiconductor fab project, $30B credit line, 2026 capex >$25B vs $8.53B in 2025. Equipment stocks already moving on it. Musk-complex capex is becoming a second AI buildout.
SNPS / CDNS — SNPS got the full TSMC lock-in treatment: A14 flows certified, 3DIC Compiler co-optimization, joint COUPE CPO flow, N2P tape-outs. CDNS demonstrating HBM/UCIe/CPO at OIP. EDA duopoly moat widening with the packaging complexity curve.
ASML — Margin gap vs TSMC "never bigger" into the Oct 14 print. Event risk both ways; no edge from these feeds.
AMAT — Joining Kioxia's EPIC Center. If the trade is Western equipment over China equipment, AMAT is the first buy.
AXTI — InP capacity 50K → 60K wafers/month by year-end, 90K in 1Q27, BOOKED THROUGH END-2027. No price war, demand migrating to 4-inch. Two-thirds China domestic revenue, zero US shipments — geopolitically insulated in a weird way. One of the cleaner capacity-shortage stories in the small-cap stack.
GLW — Only Corning and Intel can mass-produce 3D optical waveguides, and NVDA is pushing glass substrate pre-2028. Duopoly on the substrate pivot.
005930.KS / 000660.KS — Samsung: HBM taking ~30% of DRAM wafer capacity next year (from ~20%) and foundry loss narrowing 42% on HBM4 base-die — HBM now feeds foundry utilization. SK Hynix: resuming DDR4/DDR5 packaging outsourcing (own lines full) plus a $29B buyback. Both confirm shortage depth, both feed the MU print setup.
8299.TW (Phison) — CEO says shortage takes "many years" to close. Closest to spot of anyone talking. Multi-year shortage call from the controller layer.
2454.TW (MediaTek) — TP raised to NT$6,650. v9 life-cycle >5M units vs bear >2M, 100% share, v10 Icefish a major SKU. Stock fell 7% on v8t delay noise; the feed source takes the other side hard. AVGO +2.6% vs MTK +234.6% is the ASIC dispersion stat of the week.
600183.SS (Shengyi) — Leading NVL576 switch tray material evaluations, with NVDA accelerating sourcing from Chinese hydrocarbon CCL suppliers. HC content expands; PTFE ramp timing is the risk. Direct Rubin Ultra BOM read-through.
SFTBY — CDS through post-Iran-war wides. Biggest OpenAI backer staring at $30B raise to fund $1.5T commitments. Funding-hole risk is real and the credit market is saying so first.
9501.T (TEPCO) — Kashiwazaki-Kariwa Unit 6 commercial April 2026, 8.2 GW. Expect SoftBank/Oracle/OpenAI DC siting nearby. Japanese AI datacenter buildout is gated by nuclear restarts — this is the gate opening.
SNDK — Mixed tape: NAND spot soft (512Gb TLC -2.45% w/w) against enterprise SSD bit demand +80% YoY and a fresh $14B buyback. Net read: still tight where it matters.
COHR / AAOI — Vocal (and profane) bear cases on non-UHP laser suppliers. Sentiment signal, thin evidence. Don't size on it, note the positioning.
LITE / LPKF / BOE / 300394.SZ / WX / WBT / MS / GS / UBS / SIVEF — One-liners: LITE is in every compute-complex long list but ECOC resolved nothing on optics architecture. LPKF's Samsung Electro-Mechanics order (5-8 units) got a weak stock reaction despite the NVDA glass news — odd. BOE pilot at 500-600 panels/month scaling to 15K by 2028, NVDA engaging — China glass substrate progressing. 300394.SZ closest in China on 3D optical waveguides, fresh "8+2" order. WX +70% since the sanctions-mispricing call — easy money made. MS abandoning the weak-dollar thesis (DXY ~102 end-'26) matters more for EM/carry than for AI directly. GS flagging oil exports back to 23.3M bpd — rate relief + oil down is the friendly tape MU prints into. UBS's Intel 18A numbers look a third too low — consensus foundry underestimation is a theme. SIVEF ADR is an access story, not a revenue story.
No-signal names — AGX, AI, AVPT, FIVN, IONQ, OKTA, PAYC, PCTY, PLTR, PRIM, PYPL, UNIT, VRT, WBD, ZI: nothing in the feeds today. No edge, no comment.