Good morning. NVDA's first-ever quantified FY28 guide — +70% VS. THE STREET'S +45-50% — has the AI tape bid into the open. At 14x 2027 EPS, the market prices a 2028 cliff that orders don't corroborate. The ceiling is supply, not demand. Power remains the binding constraint: 2027 North American AIDC DEMAND RUNS ~35GW IT against 16.5-23.4GW deliverable, turbine slots sold out through 2031. That keeps BE, VRT, and the whole onsite-generation complex in the green. TSMC shipping CoWoS back-end to Intel Malaysia validates INTC's packaging optionality just as Section 301 tariff talk (~15% Taiwan) strengthens the onshoring case. CXMT's HBM3E breakthrough diverts commodity DRAM wafer capacity to HBM — bullish for memory pricing and ASMPT as the TCB bonder play. MRVL sits at the 100-day MA with optics carrying the raise and management deliberately saving catalysts for the 10/6 analyst day — asymmetric setup. CIEN reports Thursday: $7.7B backlog, three of four hyperscalers selecting for scale-across, ~32% growth guided at the midpoint. CRM confirms the enterprise SoR rerating — customers flipped from tokenmaxxing to AI token budgeting, and Claudeforce proves LLM vendors want partnership, not replacement. Asia: China MaaS monetization inflecting hard, Zhipu at ~$2B ARR runrate with pricing +101% — growth not from price cuts. We'll hit up NVDA, MRVL, and CIEN first, then get to the power complex and software read-throughs.
Verdict: Not the AI hype name — the AI discipline name. Two reiterations (Piper $320, Citizens $315) and one raise (Evercore $355) all circle the same thesis: AWS capex is the weapon, and Amazon's ROI framework keeps it pointed the right way. The tape is paying 21x for a sub-0.25 PEG. That's the whole pitch in one box.
"Amazon shows the most consistency in return on invested capital among major technology companies."
The numbers back it: ROIC averages ~17% from 2018-2025, projects to 14% in 2026, and sits at 12% LTM. Not spectacular — but durable. Piper's contrast is brutal: Meta and Alphabet see ROIC deteriorate 27pp and 26pp respectively from 2024-2027. Those two are throwing money at frontier models and gross infrastructure. Jassy is on earnings calls talking payback periods. That's a different capital allocation religion.
The DuckDB acquisition fits the same narrative. AWS buying DuckLabs (the open-source analytical database) speeds up analytics workloads, cuts costs, pulls more spend into the platform. Small tuck-in, but it keeps workloads sticky while the GPU buildout lands.
Sweden wind farms (~200MW now, 1GW+ total once fully online) are the boring but necessary part. Power is the binding constraint on data center deployment, and Amazon is securing supply early. Not a catalyst — a de-risker.
Bear: 12% ROIC is a step down from that 17% average. If AI payback periods slip, the market re-rates the whole story. And being on the hook for 2M+ GPUs in 2027-2028 is a lot of committed inventory if demand cracks. The discipline cuts both ways.
Baird reiterating Outperform, $200 PT. Stock at $186.29, $448B market cap. They hosted a small-group tech demo and walked away more convinced: the ontology is the moat. Open weight models are tailwinds, not threats. PLTR's layer on top is what makes agentic AI actually return ROI for enterprises.
Sovereign AI is the rising theme. Bootcamp traction says PLTR is the default vendor. The 85% gross margin is the tell — operational muscle, not narrative. Elite forward-deployed engineers plus ontology equals something rivals can't quickly copy.
Baird said the company can capitalize on the growing popularity of open weight models and continue to lead in helping businesses drive positive ROI from agentic AI.
Recent prints back it up: Q2 rev 6.8% over FactSet, op income 10.5% above estimates. UBS to $220 on 93% growth. Phillip Securities to $215. William Blair flags Maven trending toward a $1B annual run-rate. Truist likes the hiring edge. Benchmark sits alone at Hold — even they admit Rule of 155 improved.
Valuation is the bogey. $448B is rich for any tape. But at this growth clip, scarcity premium wins. Baird's $200 PT reads conservative, not a ceiling.
Distribution just got a price tag. JMP reiterates Market Outperform, $550 PT, on reports China's Moonshot is negotiating K3 revenue share UP TO 30% with MSFT, Amazon, and Google. JMP frames this as the first serious attempt to price AI distribution — if model makers pay for the pipe, Azure's enterprise reach becomes a toll booth (not a commodity utility).
"This represents the first serious attempt to price distribution in the artificial intelligence space."
MSFT at $513.53, ~28.7x. Not cheap, but 17 analysts just revised numbers up. Distribution revenue dovetails nicely with Meta already spending hundreds of millions annually on Azure AI models. Moody's Aaa affirmation, Darwinbox M&A chatter, $3.9B bond 2x oversubscribed — all background noise. The Moonshot negotiation is the signal. Early innings, but the direction of travel matters more than the print itself.
DELL HEADS INTO THE SEPT 1 PRINT WITH STREET SUPPORT IN THE UPPER RANGE — BofA nudges PT to $505 from $500, Evercore already at $550. Stock's up 278% in a year, trading $456, so the easy money is made. The bull case is simple: demand isn't the constraint. Supply is.
BofA models $65B AI server revenue for FY27 vs Dell's own $60B guide. Backlog scale, order cadence, multi-quarter deployment visibility all back the higher number. The limiters are mundane — SSDs, memory, customer datacenter readiness. Not appetite.
"Upside remains primarily limited by availability of SSDs, memory, and customer datacenter readiness rather than demand."
That's the whole earnings play in one sentence. If Dell confirms supply is the chokehold, the stock rips. If demand cracks, it's a 10% haircut. BofA looks for Q3 guide of $42-43B rev / $4.40-4.60 EPS, FY27 at $171-175B / $18.90. Three things to watch: AI demand vs supply, ISS durability, and whether storage is starting to show real AI pull-through. Lenovo's strong print has the complex in a good mood — but bogeys are high, and a supply-narrative stumble gets ugly fast.
Needham starts BW at Buy, $20 PT vs $7.16 handle. The call isn't about the core boiler business — it's about the market pricing ZERO credit for FastPower becoming a repeatable data-center generation platform. Balance sheet is fixed, backlog is ~$2.6B after the $2.4B Base Electron award, and a SECOND FastPower project is supposed to hit NTP before year-end. That's the catalyst cadence that matters.
Stock already ran +221% over the past year, so this isn't some undiscovered name. The re-rating argument is that at $7.16 you're paying for the core business plus the NPV of Base Electron — and getting the $14B+ pipeline (4-6GW of power gen opportunities) for free.
"The potential for FastPower to become a repeatable data-center generation platform as gas-turbine lead times extend and combined-cycle costs rise can drive a re-rating of the equity from current levels."
Q2 supports the momentum: EPS $0.07 vs $0.03 est, REVENUE $319.7M VS $196.97M EST (that's a 60%+ beat), and they RAISED FY26 adj EBITDA guidance to $80-105M. Net income flipped to +$14.3M from -$58.5M a year ago.
The bull case writes itself: gas turbine lead times are stretching, combined-cycle costs are ballooning, and BW has a faster/cheaper path to megawatts. The bear case: one customer concentration (Base Electron is doing a lot of heavy lifting), the stock already 3x'd, and the $20 target implies the market is just wrong on repeatability — not a great risk/reward if the second NTP slips into 2027. Watch the year-end NTP timeline. That's the tell.
Jefferies keeps it at Hold with a $1,020 PT (INR, presumably — the ADR is a different animal). Hosted management for investor meetings and heard the same chorus: discretionary spend stays weak, AI eats the budget, FY27 CC growth stays locked at 1.5-3.0%.
That's the CUT guide — management trimmed from 1.5-3.5% after a quarterly miss ($5.082B vs $5.1B). HSBC then downgraded to Hold and slashed PT to $11.47 from $14.24 on client losses and the softer outlook. STOCK DOWN 31% YTD — most of that is in the price.
The AI narrative is maturing, though. No more enterprise-wide pipe dreams.
Clients have shifted from AI proof-of-concept projects toward more realistic discussions about investment levels, achievable savings and scoped use cases.
Scoped use cases with visible ROI. The blockers: rapidly changing models, data gaps, cloud readiness, and no proof savings persist at scale. Management shrugs off model providers entering IT services and sees downstream AI demand landing medium term. Fine. Medium term is a long wait for PMs staring at a 31% YTD chart.
Canaccord trims LPTH to $15.50 from $16.50, keeps Buy. That's not a thesis break — that's a timing haircut. The Army's Next Gen Short Range Interceptor program just sent out a market survey to scope potential third bidders. Translation: before the procurement phase, the service is kicking the tires on alternatives to the Lockheed/Raytheon duopoly. LPTH is Lockheed's thermal seeker partner, so any new entrant dilutes the path — but the incumbents have a massive head start.
Stock already reacting: DOWN 15% ON THE WEEK TO $11.23. That's the headline repricing. Still up 110% OVER THE LAST YEAR, so this is a profit-taking moment for crowded longs, not a fundamental crack.
"Any competing design will be behind considering that Lockheed Martin and Raytheon each have approximately three years of design, qualification, and testing work already completed."
Steelman the pullback: if you had a big chunk of the Stinger win in your numbers, the delay into a Spring 2027 demo pushes revenue out. That's a real r/r change. But the offsetting news flow stays constructive: $11M follow-on order for BlackDiamond infrared cameras (counter-UAS, repeat customer), plus Piper Sandler just started Overweight at $15. So you've got two houses in the $15-16.50 zone. The actual decision timing slips, but LPTH keeps winning adjacent business. Watch whether the Army actually adds a third vendor — that's the event that would make this more than a one-notch PT trim.
BofA's the bull in the corner — PT to $82 from $80, Buy maintained, into tomorrow's FQ3 print (Sept 2). STOCK UP 146% IN SIX MONTHS, so this thing trades like a momentum name asking for perfection. BofA's above Street on the quarter: $12.3B rev / $0.95 EPS vs $12.0B / $0.93 consensus and company guidance of $11.5-12.1B / $0.88-0.93. They also want a guide raise — low end of FY26 revenue moved up, EPS up $0.05 from the $3.35-3.45 range.
BofA expects management to tighten its fiscal 2026 revenue range by moving up the low end of guidance for 29% to 33% year-over-year growth.
Juniper synergies, structurally higher margins, AI upside — that's the buy narrative. Demand still exceeding supply, which is the key rate-of-change tell. Lenovo's blowout quarter put a bid under the whole hardware complex (HPE +3.5% on the read-through), and the Vultr/NVIDIA AI datacenter deployment gives them a fresh headline into the print.
Caveat: UBS sits Neutral at $65 — a $17 gap from BofA's $82 tells you the dispersion. Options market's pricing ~10% post-earnings move. At 48x earnings, the stock's discounting flawless execution. A beat-and-raise is table stakes — the question is magnitude. BofA thinks the quarter's clean. The tape's already voted.
Cantor hikes PT to $50 from $35, keeps Neutral, and that spread — $44.87 spot vs. a fresh target — tells you everything. Stock's up 71% in six months into the September 1 print. Street's PTs are chasing the tape, not leading it. This is a momentum-wins-into-print setup, but nobody on the sell-side is saying the valuation makes sense yet.
The bull case is real, though. FQ1 FY27 was the first tangible proof the go-to-market reset is working: NEW LOGO GROWTH +30% Y/Y. Duo Agent beat expectations and attached to 4 of the top 10 deals. Paid committed recurring revenue now >$20M, gross margins at an 87% print. The line from management is that AI-driven code gen is pulling through CI/CD engagement — that's the "platform flywheel" narrative carrying the tape.
"Early signals from management suggesting AI is driving higher platform engagement, with usage across CI/CD workflows accelerating as more code is generated."
The bear case is just the numbers. Guggenheim models 19% revenue growth for FQ2 — a hard deceleration from 29% last year. TD Cowen at $42, UBS at $40, Cantor at $50 — every PT is below the current price except Cantor's, and even that's one bad print away. Options market implies a 14% move post-earnings. That cuts both ways, and 71% in six months means a lot of good news is already in the number.
Verdict: into earnings, this is a "right story, wrong price" tape. The AI attach data is genuinely strong, and Duo's traction is the best leading indicator since the GTM overhaul. But the risk/reward into a 14% implied move with three Neutrals and a deceleration guide? That's a coin flip with a heavy bid. If you're long, you're long the momentum, not the multiple.
EVERCORE ISI REITERATES OUTPERFORM / $415 PT — the one analyst print that matters this morning, and it's a good one. Partner surveys show sequential pipeline improvement, large-deal activity ticking up, and AI security urgency going PARABOLIC: 67% of partners now call it urgent, up from 26% in FQ3. That's a rate-of-change signal, not just a level.
No evidence of firewall pull-forward — hardware demand stayed resilient May through July, and April price hikes give FQ4 product revenue a clean tailwind. Evercore sees upside to that line. Platformization is still the core driver, with SASE and XSIAM traction broadening out. This is the "best positioned for integrated visibility and response" argument, and it keeps compounding.
Stock sits at $371.59, ~12% to target, up 102% YTD. The easy money is made; the question is whether the FQ4 print resets the ceiling. The checks say yes.
The rest of the tape is one-way too — Davidson $345, BTIG $380, JPMorgan $384 (raised), BofA $420, all Buy. Collective thesis: platform strength + FCF generation = recurring revenue upside, hardware is the swing factor. No one's short this into the print, so expect volatility on any hiccup in product revenue guidance. But the bull case is steelmanned by the survey data, not just price targets.
"Partners reported no broad-based pull-forward activity on firewall appliances, while most checks suggested hardware demand remained resilient from May through July."
That's the line that matters. Hardware was the bogey heading into this print — and the checks say no cliff.
Verdict: UBS jacks PATH to $19 from $12 — basically marking to market after a 63% run. Neutral stays. The message: demand is better than feared, but the easy money is done at 16x forward FCF.
Checks ahead of the July quarter show a stable demand backdrop — better than UBS and most investors expected. MID-$30M NET NEW ARR target looks achievable, especially with WorkFusion deals in the mix (one seven-figure June deal crossed the tape).
"Upside from current levels looks limited with the stock trading at approximately 16 times its calendar year 2027 free cash flow estimate for what it expects will be roughly 9% annual recurring revenue growth next year."
That's the whole ballgame. 83% GROSS MARGINS, but 9% ARR growth next year doesn't justify multiple expansion from here. UBS wants the late-September investor day as the next catalyst — likely a preliminary FY28 guide, less likely a formal multi-year framework.
Earnings land September 3. OPTIONS IMPLY AN 11% MOVE. PATH has a history of crushing implied — December 2025 printed +35.8% vs. a 12.3% implied move. Not sure we can read too much into that history, but it frames the setup.
Maestro Case launch adds an AI-native case management arrow to the quiver. Nice narrative point for the investor day, not a numbers driver today.
CRM keeps the AI enterprise trade alive. Second guidance raise in as many quarters, an Anthropic partnership, and a Wall Street PT parade to $266-300 — Argus led the way to $300 from $290, Buy maintained, with Truist matching at $300 and TD Cowen at $280. KeyBanc still parked at Sector Weight, the lone holdout.
"Salesforce has potential for sustainable organic growth driven by AI advancements." — Truist
The story is the cRPO print: 14% growth vs 13% consensus, with management doubling down on 2H FY27 acceleration. Margins are doing the heavy lifting underneath — 77% GROSS MARGIN and 10% non-GAAP op income growth on 11% revenue growth. That's operating leverage compounding, not just AI hype.
The Anthropic hookup is the narrative catalyst (enterprise agentic AI integration — the thing every software CEO needs to namecheck this quarter). Argus raised FY27 EPS to $16.68 from $14.41, which makes the setup cleaner: 24x P/E, PEG of 0.38. For a name with this much AI narrative torque, that's not expensive.
Risk: 2H acceleration is guided, not delivered. CRM has burned PMs on guidance cliffs before. But at a sub-0.4 PEG with the AI enterprise bid, the r/r still skews long.
BULLS HAVE THE ROADMAP, AND ROSENBLATT IS HOLDING THE $120 PT INTO THE SEPT 3 PRINT.
The call is straightforward: 5nm Edge AI designs won over the last two years are now ramping into high-volume production. That lifts ASPS and expands the customer base. Rosenblatt goes further — the next nodes (4nm/2nm) lead on performance-per-watt, and the CV architecture's embedded SRAM slashes system DRAM needs. It's a power-efficiency story that plays well in fragmented edge verticals.
"Edge AI will scale across many fragmented verticals rather than through one dominant killer app." — Ambarella customer growth officer Muneyb Minhazuddin
FT has NXP sniffing around. Rosenblatt expects management to stay mum on the call — don't hold your breath for strategic-alternative commentary. But the M&A undertone adds a bid to the story.
Stifel ($106) and Northland ($101) echo the constructive tape, if less aggressive. The 2Q revenue bogey sits near $108M (Stifel's est., +13% y/y). Options are pricing a 14% move; AMBA has beaten the implied range in five of the last eight prints. Buckle up for Wednesday.
ADSK printed a clean beat-and-raise. Q2 REVENUE $2.05B, +16% YoY (~14% cc ex-model) — 2% above Rosenblatt's forecast. EPS $3.30 vs $3.12 est. OP MARGIN 41% vs 39% est — cost actions plus top-line flow-through did the work. (92.5% gross margins: this is pricing power, not a cost-cutting story.)
The AEC vertical is the real narrative: ~15% cc growth from civil infrastructure, data center projects, and large EBA renewals. That's the unsung AI-adjacent angle — hyperscalers can't build data centers without design software. (And the model impact adjustments are getting smaller, which means cleaner prints ahead.)
Billings +10% as reported, +12% cc ex-model — marginally ahead, not a blowout. TTM revenue growth at 18%. FY27 guidance raised to ~15.5% revenue growth (~12% cc ex-model), now including MaintainX. Op margin guide held at ~39% — conservative after the 41% print. Net financing costs on MaintainX hit non-GAAP EPS assumptions, so the raise isn't as juicy as the top line implies.
"Revenue growth was supported by strength in the AEC vertical, which grew approximately 15% in constant currency, driven by civil infrastructure activity and data center projects, as well as large customer EBA renewals."
No dissent. Rosenblatt Buy/$330. UBS and DA Davidson both at $325. Stifel $285. Guggenheim $283 from $277. The $283-$330 PT spread tells you the debate is about multiple, not direction. Everybody's a buyer — the only question is how much to pay for AEC durability. This is the kind of print that resets a narrative.
MIZUHO SAYS THE FY27 GUIDE IS A RESET, NOT A RESULT. Mizuho stays Outperform, $430 PT on a stock at $358 — DOWN 45% YTD, low bar, room to run. The key shift: Global Business/Self-Employed leans off pricing and back onto low-end customer acquisition, while Intuit Enterprise Suite (~$145M ARR at FY26 exit) and mid-market carry the mix.
Consumer guide of 2-3% growth is deliberately conservative — DIY units, lower-priced offerings, and Credit Karma monetization timing all carry cushion. Restructuring savings, headcount discipline, SBC control and buybacks compound into mid-teens EPS growth THROUGH the reset.
Street's split, which is the honest tell here: Wolfe downgraded on strategy, Truist/Freedom Broker cut PTs on the muted guide, Piper/Stifel raised on the reset. Bull case is credibility rebuild. Bear case is growth stall. Investor Day September 17 decides it.
"The fiscal 2027 outlook is a deliberate reset that creates an attractive setup for management to execute and rebuild credibility."
Monness raised to $460 from $415 heading into tomorrow's Sept 1 print — and it's actually the conservative one on the street. Evercore just initiated at $525 Outperform, Citizens at $519, DA Davidson at $465. UBS sits at $460 with a Neutral. That's a crowded bullish tape: software sentiment flipped over the past few weeks, Atlas is now ~75% of revenue with growth tracking 28-29%, and the stock's up 36% in six months — hovering right under the $473 high.
The setup is simple. Beat is the base case. The stock needs a guide-up and a clean AI/database narrative tonight to justify the multiple after this run. Double-click risk is real — everyone's already in the boat.
"MongoDB has an opportunity to participate in secular trends across databases, cloud, and AI, while acknowledging that competition is fierce and the macro environment remains treacherous."
That's the whole trade in one sentence. Bull case works because there's no better pure-play developer database compounder. Bear case is that "treacherous" macro catches up to consumption models right when positioning gets fat. The real catalysts aren't tonight — they're the Investor Day on Sept 29 and MongoDB.local NYC the next day. Tonight just sets the table.
BofA's sticking with Buy and $810 PT — and the Hatch AI agent reports give them a fresh reason to pound the table. The BI story has Meta launching a consumer agent inside Instagram and WhatsApp that browses, buys, and books, with a premium tier up to $200/month. That's real agentic commerce distribution, not another chatbot wrapper.
The bull case is simple: 3B+ users makes adoption a distribution problem, not a discovery problem. BofA's line:
"Meta's large global user base represents a meaningful advantage in driving new product adoption."
The bear case is equally clean: consumer agentic adoption is embryonic, and Hatch needs to be demonstrably better than OpenAI and Gemini out of the gate. One bad launch and the narrative flips from "platform shift" to "another Meta science project."
Also on the tape: the $18B teen-safety settlement is now the known cost of doing business. Analysts are all over the map — UBS $715, Rosenblatt $886, KeyBanc OW, Needham Hold. The bulls see legal clarity and product feature lock-in; Needham sees strategy diffusion. I side with the settlement-as-overhang-removed camp. The Watermelon model launch in October is the next real catalyst — watch whether Hatch integrates smoothly or ship slips.
NVDA — The FY28 guide is the tell, full stop. +70% vs Street +45-50% means the order book is visible and supply caps the upside, not demand. At 14x 2027 EPS, the market embeds a ~50% 2028 revenue cut that orders don't corroborate. Rubin CPX revival is the sleeper: 168GB HBM4, 2,300W, ~1:1 attach with Rubin — HBM per paired inference system +58%, and prefill just became a bigger system BOM.
AVGO — Near-term custom XPU intact. They're designing a 2028 GPU for FOUR OF FIVE frontier labs, claiming 15M units. TPU 8t Superpod (9,600 chips, 2PB shared HBM) confirms custom training silicon is real. But NVDA-MediaTek creates a credible second ASIC lane — out-year volume share no longer assumed. JPMorgan anchors the Street bullish into earnings. That's a crowded long.
TSM — AI compute 5x annually; CoWoS FULLY SOLD OUT with 52-78 WEEK lead times. Capacity allocation, not a demand question. A12 mass production 2029, >1T transistors per package by 2030 — every packaging roadmap extension tightens the scarcity window. Routing $1.3B HBM assembly to Intel Malaysia de-bottlenecks back-end. 10-15% price hikes across nodes; margins beat even if volumes hiccup.
INTC — The Intel Malaysia HBM assembly deal plus SK Hynix weighing Intel for HBM4E base dies is the most consequential packaging second-source catalyst in the window. EMIB-T costs roughly HALF of CoWoS. The barrier is ecosystem lock-in — customer IP validated on TSMC PDK, libraries, PHYs. Manufacturing isn't the issue; customer approval is. Lip-Bu Tan in Washington monthly; foundry optionality real.
MU — Rubin CPX moves memory from GDDR7 to 168GB HBM4 per part at 1:1 — HBM per paired system +58%. HBM SOLD OUT THROUGH 2028. CXMT diverting DRAM wafer capacity to HBM3E tightens commodity DRAM; TrendForce PC DRAM +18-23% QoQ. The offset: custom memory with 3D stacking on N4/N3 cuts commodity DRAM content per AI server ~40%.
HXSCL — SK Hynix weighing Intel foundry for HBM4E base dies to avoid TSMC's 3-4x base-die cost markup. Margin lever and second-source hedge. CPX adds 168GB HBM4 per GPU; HBM sold out through 2028. Japanese fab still under review — infrastructure, not subsidies, now determines memory capacity location. That limits supply growth and extends the upcycle.
SSNLF — Samsung holds the best HBM4 tech and locked ~70% of capacity through 2031 LTAs. Leadership rotation in HBM underway. 400-layer V10 V-NAND targets AI datacenter storage — the memory trade broadens from HBM to enterprise NAND. TrendForce PC DRAM +18-23% QoQ; LTAs lock volume, leave upside on pricing.
SNOW — Cortex Code and Snowpark accelerate consumption. If enterprise agents actually write to Cortex, compute compounds. CRM's Claudeforce read-through supports the data-layer re-rating. AI workload conversion story.
MRVL — FY27Q2 optics carried the raise; custom silicon waits FY29. Stock sits at the 100-day MA with analyst day 10/6 — management explicitly saving key news. Asymmetric into deferred catalysts. Celestial AI at $3.25B+ on ~zero revenue is a platform bet one step from architecture decisions; only works if CPO becomes standard.
LITE — NPO completely additive to CPO, and Lumentum owns the first NPO socket. The laser gets paid twice: once in NPO, once in CPO. Ultra-high-power volume in 2H27, first ELS orders on the 2028 calendar. CEO says InP laser supply gap is WORSE THAN DRAM AND NAND. Rubin CPX disaggregation sends KV-cache over OSFP optical links — scale-out demand.
COHR — Same NPO scale-up and laser path; second seat. Internal laser capacity is the strategic asset given the InP supply gap. CPO volume production by vendors in H2 2026, silicon photonics >50% transceiver share by 2027. Levered to the optical super-cycle, just on a different clock than LITE.
CIEN — Reports Sep 3 with $7.7B backlog; FY26 guide raised to ~$6.3B, roughly 32% growth. CFO says if we had more supply, we'd sell more. THREE OF FOUR major hyperscalers selected Ciena for scale-across AI training. Service provider +28% adds a second gear; India more than doubled on MOFN. Backlog conversion is the test.
CSCO — Silicon One Ethernet plus 800G pluggables is the LIVE non-NVIDIA AI stack in Saudi Arabia with AMD MI355X. Real 2026 revenue traffic before NPO and CPO scale-up. HUMAIN next phase up to 250MW with MI400 in 2027, 1GW by 2030. AI Ethernet is the anti-NVLink bet.
AMD — MI355X-based HUMAIN system live in Saudi Arabia: EPYC CPUs, Cisco Silicon One, 800G optics. First credible non-NVIDIA rack-scale template. MI400 starts 2027 at 250MW, up to 1GW by 2030. NVDA-MediaTek may push AMD, AVGO, and hyperscalers closer together. Software ecosystem still lags CUDA.
GLW — CPO/NPO PM fiber demand is tiny this year: 7,500-10,000 km globally, only ~1,500 km ordered domestically. No PM fiber capacity shortage next year. Corning isn't the gating item, but scale-up volumes create upside.
POET — Lumilens is shipping to a major hyperscaler under a multi-billion agreement; POET is working with Lumilens on next-gen EOI products targeting 2027. That's a breadcrumb map, not evidence. No POET-specific PO disclosed. Headline trader until named sockets appear.
ANET — POET and Lessengers triangulation points to Arista as the likely networking platform for optical-engine insertion. Speculative supply-chain inference, not guidance. If NPO and CPO sockets open, Arista is a platform winner.
VRT — CPX at 2,300W and ~1:1 with Rubin raises absolute power density and MW demand even as cost per token improves. Vertiv sells the power and cooling stack around that density. 2027 NA AIDC demand ~35GW IT vs 16.5-23.4GW deliverable. Power delivery is the binding constraint. Enabler of density, not a follower.
BE — First Oracle deployment completed in 55 DAYS. Grid takes 3-6 years; Bloom compresses to weeks. Time-to-power infrastructure, not a fuel-cell story. 800VDC runs more GPUs within existing power. Manufacturing throughput, not demand, becomes the next bottleneck if this scales.
FCEL — Flagged alongside BE for the >10GW non-turbine power gap. AIDC demand roughly 2x deliverable in 2027. Onsite generation is one of the few levers that compress time-to-power.
NBIS — Signed at 3-4x its own installed-base unit price within 18 months. SCARCITY PRICING in a power-constrained AI cloud market. ARR per GW spans $8.3B to $50B; NBIS sits on the high end. Capital cost is the offsetting risk as credit tightens.
CRWV — Valuation discount may be overlooking the lead in data center capacity, revenue, and capital efficiency. The liquid neocloud comp for the AI buildout. If power-constrained pricing persists, CRWV re-rates.
ETN — Launched 2.5MW UPS, 600kW UPS in 3U, and an MV solid-state transformer at 98.5% efficiency to 800VDC. 800VDC enables more GPUs per watt. Power-density upgrade cycle. Vera Rubin NVL72 rising density adds liquid cooling content via the Boyd partnership.
SLB — Acquires Kelvion for $3.4B cash plus $0.7B debt; pro forma data center revenue EXCEEDS $2B in 2026. Strategic pivot from oilfield services to AI infrastructure cooling. Validates the power and cooling scarcity trade.
SMTC — Q3 guide +54% YoY with GM 58.3%, 15% above consensus; long-term GM >64%. 1.6T TIAs winning at EVERY manufacturer with >50% share; near-package optics engaged with 10-15 manufacturers. Laser fab capacity up 3-4x. Only blemish: Active Copper Cable demand slips to Q4 and needs redesign.
AEHR — 3D packaging makes burn-in structurally critical — one defect die scraps an expensive module. AEHR supplies high-power wafer-level and package-level burn-in. Went +126% then -50% in a month — positioning, not fundamentals. Entry only if test-capacity orders follow the AI packaging ramp.
ASMVF — CXMT HBM3E small-scale production directly benefits ASMPT as TCB bonder supplier. Chinese HBM scaling is an incremental equipment order book on top of the Korean HBM cycle. TCB is the bottleneck for HBM stacking. Any HBM volume surprise goes straight to equipment orders.
LRCX — Investing NT$9.8B (~$310M) into Taiwan for R&D and critical manufacturing design. Modest but consistent with AI wafer capacity expansion.
GOOGL — Only Mag7 down four months: search erosion, antitrust, ad migration. Short-side catalysts getting blunt after four red months. AI Mode now shows inline shopping carousel ads; Search Console AI reports give advertisers measurement. TPU 8i/8t keeps a non-NVIDIA alternative credible. Co-invested in the MediaTek convertible via TPU work.
AAPL — Ternus takes over CEO tomorrow; Schiller steps back. Leadership transition risk is live; 9/10 fall event is judgment day. Trade-secrets case against OpenAI escalated to destroying-evidence tier — an ex-engineer allegedly fed circuit schematics into an AI agent. That poisons the Siri/ChatGPT partnership exactly when AI features need to ship.
CRWD — Falcon IQ uses 50+ AI agents to auto-assess and remediate; launches on Google Cloud and Snowflake. Moves from license to platform, on-prem to multi-cloud. Agentic AI expands the attack surface — structural budget tailwind.
TSLA — Goldman raised humanoid TAM to 6.5M units, $138B market by 2035. Optionality, not earnings. Cybercab launch near: $1.5B single-day options premium shows event-driven flow. Post-launch delivery and regulatory approvals are the real test; volatility structure beats direction.
TTWO — GTA VI leaks caused the largest one-day drop in seven months. BofA: impressive but unlikely to surprise. Leaks stole the reveal option value, not unit sales. Nov 19 release unchanged. Pre-orders, not leaks, are the fundamental signal — a dip into November creates a tradeable setup.
DK — EPA granted full small-refinery exemptions on all four refineries — market saw 770M incremental RIN exemptions. Direct RIN cost relief. Company-specific margin upgrade.
PARR — Full exemption in Montana plus 50% in Washington and Wyoming. Partial RIN relief still reduces compliance cost. Cleaner beat path on refining margins.
CVI — Full exemption for Wynnewood. RIN relief drops straight to refining margin. Independent refiner group is the regulatory winner.
CLMT — Full exemptions at Montana and Shreveport, but Montana Renewables RVO obligations offset part of the benefit. Net positive but less clean than peers.
DINO — Mixed results: one full, two partial, one denial, one ineligible. RIN relief real but messy. Market parses the exact dollar benefit.
HWM — Turbine blade duopoly with Berkshire's Precision Castparts. Large gas turbine slots SOLD OUT THROUGH 2031. The power bottleneck is a multi-year tailwind for blade content.
DPC — Direct single-crystal turbine blade play via recent IPO. Same sold-out cycle with a 4-5 year order-to-COD. High beta to the AIDC power buildout.
QCOM — In Goldman's humanoid robot supplier map. Edge-compute and AI connectivity BOM. Optionality, not near-term revenue.
MBLY — In Goldman's humanoid map. Autonomy stack transfers to humanoid navigation. Long-dated optionality; no change to the automotive cycle.
HSAI — Hesai in Goldman's humanoid map. Lidar content per humanoid could exceed automotive. Early-stage TAM expansion.
OUSTR — Ouster in map. Solid-state lidar cost advantage for robot navigation. Speculative but real content-per-robot upside.
INVZ — Innoviz in map. Unproven humanoid revenue. Satellite.
AEVA — Aeva in map. FMCW lidar differentiation matters for robotics. Long-dated optionality.
BYDDY — BYD in map. Manufacturing scale and vertical integration could carry into humanoids. High optionality, no humanoid revenue yet.
XPEV — XPeng in map. EV platform plus robotics R&D optionality. Marginal to the current thesis.
XIACY — Xiaomi in map. Consumer ecosystem and manufacturing scale are the bull case. Still satellite.
AAOI — Pulled well off recent highs while Taiwan InP names made new highs. InP laser supply gap is worse than DRAM and NAND, so optical component supply is the choke point. Directionally positive, but the tape favored Taiwan names over US opticals.
MRAAY — Murata follow-through on SEMCO MLCC hikes is the tell. SEMCO has server X6S +10-20% and consumer X5R +25-30%. If Murata follows, high-end MLCC pricing power is industry-wide.
SEMUF — FC-BGA utilization 89% in H1, demand >50% above capacity, ASPs +18.7% YoY. Package Solution sales +41%, operating profit +237%. Substrates are a hard gating item on Vera Rubin volume.
LGINY — Substrate utilization 94%: 437,000 sheets against 465,000 capacity. Package Solution sales +18%, operating profit +94%. Capacity effectively full; ASPs and profitability inflecting.
UMICF — Fell 10% limit down after prosecutors raided a plant over alleged relabeled China-made PCBs as Made in Taiwan. Closed at NT$999, out of the 1,000-dollar club. Company says the investigation targets PCBs only, not ABF substrates. If the cut holds, ABF thesis is contained; if it expands, Taiwan electronics origin compliance gets repriced.
IBIDY — 2Q26 contract liabilities nearly doubled in one quarter — reads like Intel and Nvidia putting cash down to reserve future ABF capacity. The question is the order book behind the deposits. ABF tightness is becoming prepaid.
MDTKF — NVDA invests $3.5B in the MediaTek convertible; Google co-invests; NVLink Fusion on the table. Cements MediaTek as Nvidia's custom XPU arm — CSPs get an alternative ASIC lane that still feeds Nvidia's ecosystem. MediaTek committing to FOPLP for seven years. The convertible is strategic, not financing; a sell-off on good-news-as-priced is the cleaner entry.
HNHPF — Foxconn is a major Taiwan AI stack assembler; Section 301 likely settles near Taiwan's 15% negotiated rate due to >$300B US commitments. A higher rate would hit ODM margins; onshoring cushions policy risk. Part of the Taiwan CPO manufacturing stack for switch and system integration.
QUCCF — In the Taiwan AI hardware stack. 301 at ~15% manageable. AI server demand and power constraints are bigger levers than tariffs.
WICOF — In the Taiwan AI stack. 301 at ~15% largely priced. AI server mix and power constraints matter more than tariff headlines.
DELTY — Vera Rubin PSU capacitor issue is an R&D-phase quality problem; fix is capacitor replacement and rework over ~a month. Durable read: customers harden multi-sourcing at higher power densities — mainland suppliers accelerate from second-source to core. Delta still leads 18kW PSU and HVDC, but share-shift risk is rising.
No new signals in the feed for NET, WDAY, RBRK, ESTC, S, MXL, PD, RNG, CEVA, MPNGF, PINS, AFRM, DASH, BILI, ROC. Names unchanged.