Good morning. Tape's all about the AI server buildout — indices are noise next to the DELL and MDB prints. DELL blew the doors off: $47B REVENUE, +58%, AI ORDERS $60.9B and BACKLOG $95B, FY27 guide up $25B to $192B. THE DEBATE IS DURABILITY, NOT DEMAND — bears see a cycle crest, bulls see a pipeline multiples of backlog. CRDO did the beat-and-raise again ($525-535M guide vs. $516.9M est.) and still faded 6.4% post-market. Guidance beats are table stakes now; the multiple does the talking. MDB raised FY27, printed 30% growth, got hit 11% AH. Expectations, not fundamentals. SNOW's Cortex Code consumption ~3x May-July vs. Jan-Apr — the reacceleration story holds into the print. Asia's quiet; China open-weight token prices sub-$1 while DeepSeek hikes 3-12x without losing volume. THE BIG THEME: scale-in optics. Memory tier runs ~10x the scale-up bandwidth per GPU, and capital already validates — $2B each NVDA into LITE/COHR, $3.25B MRVL into Celestial AI. Power/cooling at 27.5% of AI capex stays the underpriced input. We'll hit up DELL, CRDO, and MDB first, then get to optics and semis.
Stifel's the first real post-Investor Day bull: Buy, $60 PT (~37% upside). The pitch — this turnaround is cheap. At ~3x EV/2nd-yr SALES, GF trades at the LOWEST multiple in Stifel's coverage group, and the market's pricing a declining trailing-edge foundry, not a silicon photonics story in the making.
Management's May Investor Day framework is the hook: 10-12% revenue CAGR, 40% GM exiting 2028 (vs ~27-30% now), 25% OM, ~$4 EPS by then. Photonics is the accelerator — revenue MORE THAN DOUBLING this year off a $400-500M base, $1B run rate exiting 2028, and capacity that scales to $4B. Uncle Sam's chipping in: $300M DoC LOI for US photonics/packaging on top of a $375M quantum award (2% government equity ownership — small headline, real validation). New management team, up to 50% FCF return framework. That's a different capital story than the old GF.
Q2 printed clean: $0.46 vs $0.43 est, $1.79B rev, and comm infra + data center FULL-YEAR GROWTH RAISED to 50-60%. GM at 29.9%, tracking to ~30% for the year.
UBS is the counterweight — Neutral, PT CUT to $55 from $61. September guide came in ~2% light at the midpoint, and mobile's rolling over faster: now a low-double-digit decline vs the prior high-single-digit view. So the bull case is really about photonics and datacenter outrunning mobile's slide. So far, the tape's buying it — +28% YTD. The r/r at 3x sales with a credible GM ramp is honestly not terrible.
Stifel's the new bull on the tape: initiates BUY, $515 PT, and frames TSM as a long-term industrial-cyclical holding, not a momentum name. That framing matters — this is a compounder you buy on pain, not chase on strength. The right register for a stock nearly everyone already owns.
The raised long-term model does the talking: ~25% USD REVENUE CAGR THROUGH 2029, +31% LTM revenue as of Q2 2026, AND MANAGEMENT LIFTED THE THROUGH-CYCLE GM FLOOR TO 56% FROM 53% (printing 64% today). Stifel pencils in 2-3pp of margin drag from the 2nm ramp and overseas fab buildout, widening to 3-4pp later — offset by mix from leading-edge and advanced packaging. That's the industrial-cyclical story: margin compression as a feature of growth, not a bug.
"The firm recommends adding shares during industrial-cyclical drawdowns and focusing on trough-on-trough earnings growth over time."
One positioning flag for PMs: nearly all analysts rate it a buy. High conviction, CROWDED. Bernstein just pushed PT to $554 from $430 on a CPU growth angle (high-$30B CPU revenue by 2027), Moody's went positive on the outlook. Yet BofA's flow data shows $44.2B of semis sold by active long-only funds on AI sustainability fears. So the narrative is fine but the marginal seller isn't done. That's the tension here.
Valuation at 8-11x forward EV/sales is a premium to the foundry group but a discount to AI fabless names — the gap is the Taiwan-strait discount. U.S. fab milestones are the catalyst to compress it. Samsung jacking advanced contract prices up to 15% on capacity tightness is the demand-side tell that matters more than any PT. Buy the drawdowns.
RJ keeps Strong Buy but cuts PT to $51 from $61 — and the stock sits at $31.12. That gap says the firm still sees a double from here. The cut says the path is messier than the AI narrative implies.
Q2 bookings were the strongest management has "ever seen." Raised FY26 guidance. AI infra demand is real. The problem is the EXISTING BASE repricing lower — RJ models NEGATIVE MRR GROWTH THROUGH 2H26 AND ALL OF 2027 as the pricing reset washes through. Transitional years, plain and simple.
2026 and 2027 are transitional years before pricing pressures begin to ease in 2028 and incremental capacity becomes the primary growth driver.
The bridge: FY26 normalized EBITDA base of RMB 5.47B (ex one-time contract settlement), then just 6.5-7% growth in 2027, then ~20% in 2028 as pricing pressure eases mid-year. EBITDA per MW steps down from $2.2M to $2.1M — that's the reset in the math. Capex keeps climbing: every incremental 1GW costs RMB 15-20B.
P/E of 12.59 looks like a value trap if the reset drags, a gift if the 2028 inflection holds. This is a 2028 story trading in 2026. Watch the MRR trajectory each quarter — that's the tell.
Scotiabank's the loudest bull — PT to $430 from $320, Sector Outperform, on "improving momentum in CISO checks" after 15 investor meetings this month. The stock's up 157% in six months, trading at 206x EV/EBITDA and 332x P/E. That's double the five-year average multiple. This is not a value debate; it's a momentum-and-print debate.
Scotiabank frames the setup cleanly:
Palo Alto must report NGS ARR of $8.96 billion in Q4 and guide to $11.13 billion for fiscal 2027 at the midpoint for shares to remain flat.
That's the bogey. Beat it and the tape works. Miss or guide soft and a 206x multiple compresses violently. CrowdStrike's strong print last week raises the comp bar — PMs will hold PANW to that standard tonight.
Street's mostly onside: Evercore at $415, JPMorgan raised to $384 on platform strength and FCF, BTIG at $380, D.A. Davidson at $345. The collective thesis — improving partner sentiment, fatter demand pipelines, AI security driving larger deal activity. No one's bearish, they're just arguing about how much premium to pay.
The bear case is lazy but real: valuations are euphoric and the options market implies an ~8.6% move tonight. History says PANW beats the implied move half the time. Scotiabank calls the near-term r/r "slight positive," not a conviction buy. That's the tell — everyone loves the story, but even the bulls are hedging.
Verdict: THE STREET IS STILL RESETTING THE E-COMMERCE NARRATIVE. Evercore chopped PT to $510 from $630, cut Q4 and FY27 estimates, and admitted its historical e-commerce advertiser spend numbers through Q2 were TOO HIGH. Stock at $312 — hugging the $297.50 52-week low, DOWN 54% YTD. Outperform rating intact, but the credibility of the growth algorithm is the battleground now.
August mobile game advertiser checks aren't the problem: 3 of 5 contacts positive on early Q3 model enhancements, 2 citing material lifts, 2 neutral. Core gaming holds up. The issue is what comes next — Evercore assumes ONLY self-learning gains and seasonality for FY27 under Axon 2, with Axon 3 potentially 12 months out. And they flag a real mechanism risk: increased brand utilization of mobile in-app inventory could raise clearing prices, muting take rate accretion.
Five firms cut this cycle — BofA to Neutral ($400 from $430), BTIG $408, Benchmark $440, Needham $475 (kept Buy), Piper $325 (Neutral). PT range tightened from $385-630 to $325-510. All converging on the same theme: gaming matures, e-commerce ramp is tougher to model, and incremental model lifts get harder to trust.
"Model lifts may become harder to achieve under the current Axon 2 model architecture."
Bull case isn't dead: 88% gross margins, PEG 0.28, Axon 3 catalyst inside 12 months. At $312 the stock prices in a lot of bad news — but with estimates still sliding, catching this knife is a game for patient PMs. Not a momentum trade.
Stifel opens the book with a Buy and a $270 PT, but the real message is simpler: Tower's bull case just stopped being a story and started being a schedule. $1.3B of contractually committed silicon photonics revenue, backed by ~$290M of customer prepayments — that's backlog, not a pitch deck.
THE THESIS: Management raised the 2028 target model to $3.6B revenue, and nearly all of the $800M increase is SiPho. The model spits out 45.3% GM / 38.3% OM — structural margin expansion, not a quarterly beat. Current revenue sits at $1.71B, so we're talking about a doubling on the back of one product cycle.
Stifel cited the company's contracted backlog and visibility in silicon photonics as drivers for structural margin improvement.
THE PACE: Management targets a $1B SiPho run-rate exiting Q4 2026, up from $680M+ in Q2. Stifel sees that reaching $2.5B+ by Q4 2028. Japan is the post-2028 growth engine — METI Track 1 production-ready in Q4 2027, Track 2 quadrupling Japanese 300mm output.
THE CROWD: BofA is the standout bull at $367 (Buy) on AI data center exposure. Benchmark stays at $335, and they're the ones flagging the tape — recent pressure traces back to FCC restriction speculation hitting China-based module customers. Worth watching if that headline gains political traction.
THE CATCH: The stock is up 249% in a year. At $199, the easy money is made — you're paying for flawless execution on a contractually de-risked plan. Q2 printed a beat (EPS $2.68 vs $2.21 est; revenue $1.4B vs $1.32B est, though the quarterly figure was $460M, +24% YoY — the $1.4B looks like a TTM mix-up in the source). The risk asymmetry now: SiPho slips a quarter, or the China political angle gets ugly, and the 249% run corrects fast. Steelman for the bulls: prepayments de-risk the downside, and $290M of customer money is a strong commitment signal in a capex-heavy industry.
All three desks on the same side — GS reiterates Buy/$18, BofA at $21.40, Jefferies at $17.70. Stock trades at $10.24. That's 75-109% upside to the PT cluster, with street targets spanning $16.01-$26.37.
The thesis is AI cloud. Revenue +82% YoY, NOW 43% OF TOTAL (per GS). Company puts the broader AI mix at 56% of Q2 revenue. The margin kicker: MaaS carries the highest gross margin inside AI cloud. Mix shift is the whole ballgame.
Q2 delivered. Revenue RMB3.07B, +31% YoY, beat by ~RMB40M. FIRST POSITIVE GAAP OPERATING MARGIN EVER. Management guides stable GM/EBITDA into 2H26, then positive adjusted net profit in 2027. That path is real, not a promise.
Capex is the watch item: Rmb15B budgeted for 2026, split between direct procurement and computing power rental. AI IaaS payback at 3.5 years by revenue — shorter by cash flow thanks to prepayments. The rental component softens the headline number.
"Model-as-a-Service carries the highest gross profit margin within the AI cloud segment."
That's the line that compounds. If MaaS scales, blended margins climb even under the capex load. The gap between $10.24 and the $18 PT says the market wants proof on 2027 before paying full freight. Fair enough at these levels.
Bernstein back on the bullhorn: Outperform, $800 PT, and a bold call — META IS ON TRACK TO SURPASS GOOGLE SEARCH IN AD REVENUE THIS YEAR. Ex-Maps/Gmail, the firm says Meta may already be there. The headline is sharp, but the substance is why AI helps Meta more than it helps Search.
The logic clicks. Search reads explicit intent; Meta has to predict what you'll engage with. AI-driven recommendations, targeting, and measurement attack that prediction problem directly. HUGE LEVER FOR A $1.45T BUSINESS PRINTING 27.7% REVENUE GROWTH LTM. The wrinkle: digital ad fundamentals have "arguably never been stronger," yet ad stocks keep bleeding. Investors are punishing Meta and Google on AI capex, return uncertainty, and growth durability — not on ad demand.
"Meta must predict what content and ads users are most likely to engage with, unlike Search where users explicitly signal intent."
Rest of the tape: the Street reads the $18B teen safety settlement as a clearing event, not a catastrophe — UBS at $715 Buy, KeyBanc at $780 Overweight. BofA's $810 Buy rides the Hatch AI agent news (Instagram/WhatsApp integration). Needham still can't get there, citing costly strategy diffusion. Stock sits 28% below its 52-week high. Risk/reward skews favorable if AI-driven ad targeting compounds — and Bernstein makes a compelling case it does.
Long setup into a binary print. Stock sits at $331 — UP 97% IN SIX MONTHS, basically pinned at the 52-week high — and the street is still adding fuel ahead of tomorrow's FQ2 report (Sept 2). Rosenblatt to $345 from $285, Buy, on consistently positive channel checks and AI product pull. That's the whole game right now: migration-driven base revenue plus AI attach on top.
The bar is decel, not acceleration. Rosenblatt models ~31% organic product revenue growth vs 34% printed in FQ1. They see a marginal beat. CoCo at 7,100 customers, CoWork at 5,000 — early, but the data-native coding agent story is gaining real traction with the enterprise crowd. Cortex AI Gateway adds another consumption vector.
The broader cluster is far more aggressive than Rosenblatt: Cantor at $405, TD Cowen $370, Benchmark $360, Guggenheim stuck at Neutral. Bull thesis is simple — data warehouse migration is a multi-year tailwind and AI workloads are just starting to monetize on Snowflake's consumption model. Bear case is equally simple: this is a 97% six-month move into an event with OPTIONS IMPLYING A 12% POST-EARNINGS SWING. That's not a risk/reward you need to force.
"Consistently positive channel checks in recent months and ongoing healthy enterprise cloud migration efforts."
The AI narrative is real but the setup is priced. If they guide up, this thing gaps through $350. If they don't, that 97% move gives back a chunk. Coin flip on guidance — slight edge to the upside given the check tone.
DEAL'S DEAD, BACK TO THE GRIND — THAT'S THE BULL CASE, AND IT WORKS. BofA resumed coverage at Buy with a $42 PT (~18% upside) after Element Solutions and Solstice walked away from the merger, no fees, no drama. The reframe is simple: stop staring at the M&A arb, start staring at the P&L.
And the P&L is good. Q2 EBITDA PRINTED $184M, +35% Y/Y AND ~10% ABOVE CONSENSUS. Electronics carried the quarter — 20% organic sales growth, 47% adjusted EBITDA growth. Total net sales +56% (15% organic). This isn't a one-off beat; BofA sees high-teens EPS growth for the next two years on AI infrastructure buildout plus Kuprion scaling.
"With the deal off, focus returns to the company's fundamentals." — BofA's Matthew DeYoe
The street's already crowding in. Freedom Broker upgraded to Buy with a $48 PT (post-termination), BMO sticks at Outperform/$50. So the PT range is $42-$50, with the resumption guy at the low end — meaning the fundamental path is clearer than the price target dispersion suggests. Not sure we read much into the $42 vs $50 gap; it's just conviction on timing, not trajectory.
Risk to the trade: the stock ran on deal hopes, and now the "AI electronics" narrative is getting crowded across the space. But with 6 analysts revising estimates up and EBITDA compounding in the mid-30s, you're paying for growth that's actually showing up in the print. Fine r/r here.
BofA TO $600 FROM $500 — but the PT is just the hook. The real signal is the rate of change in Azure. FQ4 growth ACCELERATED to 43% from 39% in FQ3, with a 45% guide for FQ1. That's the inflection PMs chase, and RPO +84% YoY gives it credibility — backlog growing faster than revenue means the forward view isn't a hope, it's a compound.
The Copilot print is the quiet alpha in this one. PAID M365 COPILOT SEATS >30M, NET ADDS DOUBLING Q/Q. Monetization is finally showing up in the seat data, not just the keynote slides. BofA slapping 28x CY27 on it (from 24x) is math, not multiple worship — stock trades at 28.5x with a 0.89 PEG, so you're not paying for the acceleration, you're getting paid to own it.
"We remain constructive on Microsoft's long-term positioning across AI infrastructure, models and applications."
Meta spending hundreds of millions annually on Azure deserves a nod — the frenemy dynamic keeps writing checks to Nadella. Nvidia picking up Hugging Face is interesting M&A gossip but doesn't move MSFT today. If anything, it confirms everyone's chasing the same AI distribution lane — and MSFT is the one already collecting tolls in it. 17 analysts revising estimates up says the street smells it too.
Cantor bumps PT to $25 from $23, stays Overweight into FQ2 — and the read-through is that the stock can work even after ripping ~16% off pre-Q1 levels (vs +4% SPX). Partner checks show 76% of respondents at/above plan, up from 70% last quarter, though ahead-of-plan ticked down to 38% from 40%. Directionally fine, not blowout.
The bull case isn't the quarter. It's the optionality. At ~9x FY27E sales, Cantor argues the market isn't paying for agentic AI or non-human identity growth — and management embedded minimal AI contribution in guidance.
"The valuation does not reflect growth opportunities in agentic AI and non-human identities. Yet."
That "yet" is doing heavy lifting. The beat bogey is $8-10M organic net new ARR, plus a guidance raise and commentary on AI ARR inflection. Entro Security acquisition closed — gives them standalone non-human identity products and a real foothold in a category that's about to explode with agent workflows.
Consensus is quietly constructive: Truist Buy on agentic identity/real-time governance, RBC Outperform ($19 PT, stale), Mizuho the lone Neutral anchored to the $2.1B ARR by 2029 target. Risk/reward skews positive here — you're getting the AI narrative free at 9x sales with partner checks confirming execution.
Verdict: Truist takes CRWD to $300 from $245, Buy. Stock already +24.6% in a week to $231 — market front-running Fal.Con. Record FQ2 and another FY27 guide-up fuel the move, but the trade is now about durability, not the print.
The quarter was clean: ARR BEAT BY $46M, NET NEW ARR +51% YOY, revenue +26%, EPS +33.5%. Bulls are piling on — PTs span $240 (Piper) to $425 (Argus) with Benchmark and DA Davidson in the $245-250 zone. All telling the same story: AIDR monetization, Falcon Flex expansion, and the Identity/Cloud/SIEM stack extend the net-new ARR runway.
Lone bear is Bernstein at $119 Market Perform. Not sure we can dismiss that outright — but against this tape, it's a contrarian seat.
Truist's Fal.Con checklist:
"CRWD heads into Fal.Con following a record quarter and another meaningful increase to FY27 expectations. We will primarily be looking for evidence that the post-Mythos demand acceleration can support a higher long-term growth framework, with particular focus on AIDR monetization, future net new ARR potential, product-level growth drivers and Falcon Flex expansion dynamics."
The tell: if AI expands both the size and duration of the opportunity, this rally extends. If it's a one-time refill after Mythos, you're chasing a 24% move that already happened.
DAVIDSON REITERATES UNDERPERFORM INTO THE PRINT — $7 PT vs $10.42 tape, so they're calling for a third of the equity to vanish. Tom Siebel's first quarter back as CEO, and the bar is low: LTM REVENUE DOWN 35.67%, Street still modeling ~10% sales decline this year.
The firm sees potential upside to total revenue from professional services and expects stable subscription trends this quarter. But that's a quarter trade, not a durable one. The whole thesis in one line:
"The firm said it continues to struggle to underwrite durable subscription revenue growth over the longer term."
Peers aren't fighting it: UBS at $12 Neutral, KeyBanc Underweight $6. And the bull case from the last print — FY27 guide at $225M midpoint, blowing past the $200-210M whisper — runs into a horrid bookings number (consolidated sales -53% YoY). Freedom Broker blamed execution, not demand. Siebel's return is the bull's answer. The bear's answer: bookings don't lie, and one CEO coming back doesn't make subscription revenue durable.
PT to $303 from $300, Neutral maintained, stock at $316.85. Read that again — the target sits $14 BELOW the tape. Rosenblatt is telling you the stock is ahead of itself, just not enough to scream sell into a new-CEO + iPhone catalyst stack.
Tim Cook out, John Ternus in (Aug 31). Sept 9 iPhone event with foldable chatter carries the narrative bid. Rosenblatt gives credit for premium pricing and supply chain navigation muting margin pressure — but at 31x FY27 EPS against a LOW DOUBLE-DIGIT EPS CAGR, that's a big premium for an execution story that hasn't printed yet.
"The firm views this as a healthy premium to a low double-digit EPS growth compound annual growth rate."
That's the bull case in one line — "healthy" doing a lot of work. 36x trailing means the market already assumes the narrative works. Evercore's OW reiteration on the Mac mini refresh (M6/M5 Pro, $300 price bumps) is a nice margin sidebar, but this tape is about the phone and the transition, nothing else.
R/r is poor for fresh adds: target below spot, Neutral rating, multiple at the high end. The real PM question — does the foldable put a floor under the narrative even if the fundamental bar is high? That's the trade.
TD Cowen pushes CRM to $300 from $280 on the Anthropic partnership — and honestly, that's just validation of what the tape already told you. Stock's up 25% in a week, trading at the 52-week high ($269). This isn't a fundamentals call; it's a positioning call. The "Claudeforce" deal kills the AI-lab-ate-the-SaaS-world narrative, and that narrative shift is worth more than any single quarter of cRPO.
The bull case is simple: Anthropic choosing partnership over competition removes the existential discount on all of SaaS. TD Cowen sees the joint product strategy as a win-win — AI labs get distribution, SaaS gets AI-native workflows without building their own models. It's a category re-rating masquerading as a single-stock PT hike.
"The new alliance between Salesforce and Anthropic demonstrates that Anthropic intends to partner rather than directly compete in the SaaS market."
The fundamentals aren't bad either — cRPO grew 14% cc, beating the 13% consensus, and management guided similar for next quarter, having already raised full-year guidance twice. 30 analysts revised estimates upward. But the skeptics have a point: Truist flagged Q2 as the crucial test for whether organic growth can sustainably reaccelerate, and KeyBanc's sitting at Sector Weight, seeing the strength but not paying up for it. Argus and Truist both anchor at $300 — the street's clustering there, but nobody's stretched beyond it. This one's a momentum tape now; the AI narrative is the bogeys, and so far it's hitting them.
Hearing PMs argue the DELL fade case: "Decent shot this fades because it's glaringly obvious we're at the part where the cycle should crest." Backlog is real, but if $60.9B orders mark peak, FY28 comps look brutal.
Word on CRDO: beat-and-raise-then-fade is now a tape feature. 22 buys, 1 hold, 0 sells — nothing left to squeeze. The $764.3M cash print vs $1.37B estimate is the tell; Street's balance sheet model was wrong.
Channel checks on MDB: AI workloads are "early momentum," not the driver. Core enterprise drove the beat. Until AI-specific ARR shows up, it's a quality compounder trading like an AI re-rating. SNOW's Cortex Code convergence is a multi-quarter pricing overhang.
Hearing DeepMind's public messaging is hurting GOOG internally. Chief AI architect admits "a little bit below the frontier" with no 3.5 Pro timeline. Gemini 4 is "most ambitious run" but no dates. The market's "does Google care" question just got an awkward answer.
NVDA warned biggest customers of 15% AI server price hikes as memory costs soar. Margin defense, but it creates a demand-elasticity test. Watch hyperscaler acceptance into year-end.
Expert checks on LITE: NPO discussed but not yet implemented or piloted. They're testing customized cable specs to reduce optics demand. Scale-in still has execution risk — NVDA's $2B doesn't make physics instant.
The MRVL warrant scope — "memory interface controllers and near-memory compute" — is the public paper trail for Google's memory pooling play. Largest scale-in exploration running. If it converts, MRVL's AI revenue ceiling moves dramatically.
AVGO prints tomorrow. "That AI semiconductor growth better show up." Street has next quarter at $35B. Event risk is real; the inverted vol curve says the market knows it.
Korea August DRAM export ASP +1.8% MoM, flash +63.3% MoM. Physical and agentic AI creates a new demand leg — one robot generates ~200x an LLM pretraining dataset per year. MU is the highest-leverage US expression.
Custom HBM is the sleeper debate. 5x LLM inference gains, but value may accrue to the accelerator designer, not the commodity DRAM maker. Watch custom HBM awards and mix into HBM4.
Word is the Lambda deal has an "Nvidia twist" — NVDA holds the lease, Lambda deploys. NVDA financing neocloud capacity it then anchors. Circular financing scrutiny will follow; the bull case is that it accelerates demand.
Korea's trillion-dollar sovereign AI framework is framed as an NVIDIA win. CUDA stack consolidates ecosystem value, Hynix and Samsung are relative losers. Sovereignty money flows into the same architecture.
Power/cooling = 27.5% of AI CAPEX. Underappreciated per-token cost driver. Bullish for the whole thermal/power supply chain — SLB's Kelvion deal is the smartest expression of it.
China open-weight token prices dropped below $1/M for the first time. DeepSeek raised prices 3–12x without losing volume. Deflation at the token layer, pricing power at the model layer. Zhipu ARR up 6x since March.
OpenAI quietly folded its Premium tier into Business. Sudden move. Watch ARPU mix implications.
Sarah Guo: "We are 1–2 years away from some sort of exponential intelligence." Says that belief is new within the last 12 months for a lot of researchers. Sentiment datapoint for compute-demand underwriting.
A senior hyperscaler infrastructure leader says nothing moves the needle "at sufficient scale before 2030." Sober counterweight to every alternative-power bull case.
Zhipu runs production inference on ~100,000 domestic accelerator cards. Cost per token down 80%. Another brick in the NVDA China TAM bear case.
MediaTek's $3.5B NVDA convertible likely funds EMIB substrate prepays. Supply-chain read-through — MediaTek becomes the custom XPU bridge into NVLink Fusion.
SK hynix reportedly not using Intel for HBM base die. Removes one INTC foundry bull leg. EMIB-T substrate yield is the 2027 swing; conflicting rumors persist.
SNOW Cortex Code consumption nearly tripled May–July vs Jan–April, still growing 15–20% MoM. Hit 1% revenue contribution in one month — took Snowpark a full year. Projections of 8–10% contribution by year-end. That's the reacceleration engine.
Charter's CFO jumping to the GOOG+BX neocloud JV. Talent flow tells you where institutional money is headed — TPU hosting as an asset class.
GPRO is this cycle's froth marker. $285M stock deal for an optics startup, +79%. Camera company → AI transceivers. No anchor. Terminal-pivot momentum.
No fresh signals on AMZN, ESTC, ADBE, TEM, VSAT, PAYC, TTAN, GRRR, FIVN, RDVT, STDN, SAIC, or SPCX. Clean tape, no news to trade.