Good morning.
SPX futures flat, NDX +0.2%. MU +8% getting close to $1T — memory is the binding constraint again. Foxconn print: NT$2.51T revenue, +39.8% YoY, 6% above consensus. Cloud & Networking now nearly 50% of revenue. The margin debate? We'll get there. Asia: TSMC guiding revenue to near 40% USD growth, up from "over 30%". That's a bogey-shifting move. July 16 earnings is the next catalyst.
Three themes framing the day:
1. Memory crisis, real. MU sees DRAM tightness through 2028. Lenovo now shipping YMTC SSDs globally. Every Rubin platform takes 3x the memory of Blackwell. Buy the dip on any Apple-China exemption scare — that's noise.
2. NVDA's performance moat eroding. Kyber NVL144 delayed 12+ months to 2028. NVL72x2 rack cancelled. Rubin Ultra's scale-up domain just got smaller. Spec downgrade = bearish CPO, bullish NPO, and big tailwind for AMD and the TPU ecosystem. We need to diligence the SemiAnalysis report.
3. Hyperscalers go internal. AMZN raised Trainium3 server shipments 20-30% above plan for Q3. Trainium4 acceleration is under consideration. Also designing in-house end-to-end silicon for devices. This isn't a fringe story — it's the core of the ASIC thesis hitting an inflection point.
We'll hit up MU, TSM, and NVDA first, then get to AMZN and the broader memory / networking complex.
Beat, raised, but still a show-me story. The stock rips 17.6% in a week — but that's off a 44% annual drawdown. Two firms lifted PTs ($275 at BMO, $253 at Jefferies) but neither changed ratings (Market Perform / Hold). The collective read: subscription acceleration is real, margins are not.
Fiscal Q3 top and bottom line beat (EPS $4.53 vs $4.45e, rev $622.9M vs $617.3M). The headline number everyone is watching — organic ASV growth (+7.1%) — hit its fifth straight quarter of acceleration. Management reaffirmed guidance across all metrics and said they're tracking toward the high end.
But here's the push-pull: margins missed due to timing-related performance comp and tech investment. Jefferies notes adjusted op margins have trended down for two years. BMO says the AI spend is early stage. The market is buying the "investment phase" narrative for now.
"Early AI initiatives continue to show positive traction, supporting faster growth within the installed base and accelerating the shift toward larger enterprise agreements." — BMO Capital
TRUIST GOES TO $155 ON RUCKUS CLOSE. BUY RATED. THESIS: THE DEAL CHANGES THE REVENUE MIX TOWARD SOLUTION SELLING – FASTER GROWTH, BETTER MARGINS.
New model post-close (July 1) yields CY27 EPS estimate bump to $9.95 from $9.20. PT at 15.5x that – a 5.5x discount to the S&P 500, which is the historical norm. Not pricing in much multiple expansion yet, but the earnings power is the story.
The key line:
"Truist expects the combined company to generate more revenue from solution selling, resulting in faster total sales growth and higher profitability."
That’s the whole bull case in one sentence. BDC not a sexy name, but the Ruckus assets were a good fit for their industrial/broadcast/commercial networking stack. $1.75B net proceeds to Vistance confirms the price paid was real. Now it’s about execution and showing the cross-sell.
R/R: At $120, 15.5x forward earnings on a growing base. Not egregious. Not going to double overnight, but for a PM looking for quality compounders in the mid-cap wireline space, this is a solid hold.
The PT cut to $169 is a nothingburger — pure mechanical adjustment for the 4:1 split. Barclays kept Overweight, same 42x FCF multiple on FY31 FCF of ~$4.2B. The only change: share count went from ~259M pre-split to ~1.036B post-split. Stock is already trading at $195.28, which is ABOVE the split-adjusted Cantor PT of ~$181 (that was $725 pre-split). So you've got a split-adjusted divergence — Barclays at $169, Cantor at $181, street consensus somewhere in between. Meanwhile the tape is doing its own thing: +70% over six months, near the 52-week high of $196.50.
The real story is the operational momentum, not the analyst target shuffle. Earnings beat consensus by 1.7% (a nice bounce back from the prior miss), and the AWS AI security expansion + Project QuiltWorks extension are incremental positives. The identity security launch for AI agents is more narrative than near-term revenue, but fits the "CrowdStrike owns the AI security stack" storyline.
"The lower price target accounts solely for the increased share count resulting from the stock split." — Barclays, reminding everyone this is a math exercise, not a thesis change.
Bottom line: Split-adjusted targets are noise. The stock is pricing in continued execution on the AI security cross-sell. At ~$195, you're paying ~42x FY31 FCF — same multiple Barclays uses. That's not cheap, but the rate of change on AI agent security spend is still accelerating. Worth holding unless you think the FY25 revenue deceleration (guidance implied ~25% growth) is a canary.
Verdict: Buy initiation at $41 from Freedom Broker, but the stock has been getting clocked — down 7.7% in the past week, trading at $30.12 with a P/E of 11.2x and a PEG of 0.28. The thesis is pure capacity-cycle torque: PLAB is one of only two independent merchant photomask players globally, and it’s in the middle of its largest-ever investment cycle — ~$330M in FY26 capex — expanding in the US and Korea for 28nm-and-smaller IC nodes and AMOLED. The recent pressure is the entry point, assuming the tape-out recovery that management flagged starting in May actually materializes.
“The stock has experienced recent pressure, creating what the firm views as an attractive entry point at current valuation levels.” — Danial Yermakhan, Freedom Broker
Near-term pain is real. Q2 (Apr) missed: EPS $0.42 vs $0.53 est, revenue $209.9M vs $216.7M. Q3 guidance midpoint of $211M vs $218.5M consensus. Craig-Hallum cut PT from $48 to $42 but kept Buy, citing IC softness, memory supply constraints, and geopolitical delays on design releases. The offset? Management says tape-outs started recovering in May. That’s the leading indicator for photomask orders 2-3 quarters out.
Bull case: dominant oligopoly position + high-end capacity coming online just as demand for leading-edge masks (below 28nm, AMOLED) accelerates. PEG at 0.28 is absurd for a semi-equipment-like play with structural tailwinds. If the recovery is real, this is a $40+ stock in a year.
Bear case: guidance implies no recovery in the current quarter. Capex is massive relative to current revenue (~$215M/qtr vs $330M annual capex). If the cycle doesn’t turn, ROIC gets ugly. And at $30, the market is pricing in continued weakness — you’re betting on the timing of the turn, not the structural story.
Worth a look for PMs who can handle a 3-6 month tape-out watch. The risk/reward skews positive at these levels, but don’t size big until you see the Q3 print confirm the recovery signal.
UBS doubles down. Raised PT to $110 from $92, keeps Buy. Thesis is straightforward: demand backdrop is solid, cloud revenue (roughly 50% growth) is sustainable, AI tailwinds are real, and competition is MIA. The firm’s own industry checks (15 of them) back it up. Stock’s already up 119% in the past year, trading just 1% off its 52-week high. Multiple at 48x CY27 FCF is priced for perfection — but the narrative keeps improving.
“The setup for the stock remains attractive despite its recent run… We see potential for material upward estimate revisions.”
UBS projecting 26% and 25% revenue growth in FY26 and FY27. The only real pushback: elevated multiple and positioning risk after that run. But the rate of change is still positive, and the bull case (AI-driven software artifact demand) is intact. Not much else to say with one article — watch the Q2 setup and whether that ~50% cloud clip holds.
HSBC just went all-in. Raised PT to $200 from $100 (that's a double) — now explicitly including Intel Foundry in SOTP after capacity bottlenecks are "widely acknowledged" and external customer engagement is building. Stock already up 481% over the past year to $123.47. The bet: design commitments hit H2 2026, then server CPU shipments accelerate.
"HSBC raised its 2026 server CPU shipment growth estimate to 25% year-over-year from 20%, driving its data center and AI revenue estimate to $24.1 billion, 4% above consensus."
The bigger swing is 2027: 30% YoY server CPU shipment growth vs prior 20%, with 18A ramp and internal capacity reallocation. That puts DCAI revenue at $33B — 20% above consensus. Mizuho and Cantor also raised PTs recently ($135 and $150 respectively), echoing the same narrative: advanced packaging tailwinds, compute strength, but also flagging supply constraints that could persist through 2027 as hyperscaler capex keeps pumping. Not all bullish—Mizuho remains Neutral on valuation. But the rate of change in foundry narrative is accelerating. Intel turns profitable this year (consensus $1.12 EPS). PMs should watch for any customer announcements from IFS — that's the next catalyst.
Verdict: TD Cowen just threw cold water on a 38% YTD rocket. Raised PT to $320 (from $275) but kept Hold — and that’s 13% below current close of $367.53. They see the data center story (mid-single-digit % of rev) and even concede upside to Street estimates, but at 34x P/E and 22x EV/EBITDA they call the move “excessive.” Not a fundamental call — a valuation call.
“The recent sharp move higher is excessive in our view.”
On the other side, Barclays upgraded to Overweight with a $414 PT last month, leaning into the data center capex cycle narrative. So you’ve got a clean bull/bear split: thesis = same, timing and multiple = contested. TD Cowen’s target implies 15x FY27 EBITDA and ~25x FCF — conservative relative to current expansion. For PMs: this is pricing in perfection. Any stumble on datacenter ramp or multiple compression will sting. If you’re long, you’re betting growth accelerates. If you’re flat, the r/r looks asymmetric to the downside from here.
JPMORGAN RAISES PT TO EUR190 FROM EUR170 — maintaining Overweight after bumping 2026/27 EBITDA by 3-5% on fiber price upside. The stock has already ripped 132% in the past year, but the PEG ratio sits at 0.34, so the multiple hasn't re-rated with the earnings growth. That's the bull case: still cheap if the hyperscaler framework agreements land.
Next catalyst is July 30 Q2 print (they model EUR713m adjusted EBITDA, 6.7% organic growth — well above NKT/Nexans). But the real trigger is this:
"The formal announcement of framework agreements with U.S. hyperscalers and capacity expansion as the next catalyst."
JPMorgan's new target uses December 2027 forward estimates (same multiple, higher base). Bottom line: momentum is real, but the easy money is already in the price. Need the hyperscaler pen to push the stock past EUR190.
MU — The supercycle vs. topping pattern debate is the defining call. DRAM tightness through 2028 is structural — Vera Rubin requiring 3x+ memory vs Blackwell is a demand step-function, not a cyclical blip. Pricing accelerating: Samsung reportedly raising 20% QoQ, UBS sees DDR4 Q3 +32%. The SK Hynix ADR listing on July 10 breaks MU's monopoly as the only US-listed HBM name. Relative valuation arbitrage creates a spread to watch. Lenovo shipping YMTC SSDs globally is a competitive headwind, but DRAM tightness overwhelms it for now.
SNDK — NAND follows DRAM's tightness, but the cycle is on a plateau, not flat ground. Kioxia director says AI raised the cyclical floor. Pricing raised 30% QoQ per UBS. The new demand vector: AI analyst agents generating staggering telemetry and log data. This is a volume story not priced into consumer or enterprise cycles. Lenovo shipping YMTC SSDs globally introduces Chinese NAND competition — a new dynamic to monitor. Target calls for SNDK > $2,200 but bears argue memory stocks already expensive at 22 P/E for MU.
NVDA — The Kyber NVL144 delay 12+ months to 2028 and NVL72x2 cancellation is a massive execution setback. Rubin Ultra scale-up domain now limited. This erodes the performance moat the market priced in. Read-through: bearish CPO supply chain, bullish AMD and TPU ecosystem. The monopoly premium is eroding. Despite $725B aggregate hyperscaler capex in 2026, the product execution issues raise questions about future growth trajectory. Multiple compression risk as investors re-rate for competition.
AMZN — AWS raised Trainium3 server shipments 20-30% above plan for Q3. Mass production in July. Trainium4 acceleration under consideration. This is a major upside surprise on ASIC momentum. Amazon designing end-to-end in-house silicon for devices — extending beyond cloud to edge. Less future GPU demand from NVDA, but net positive for Amazon's capex efficiency. The custom silicon narrative is accelerating faster than consensus expects.
TSM — Guide-up to near 40% revenue growth in USD from 'over-30%' previously. July 16 earnings is the catalyst. Strong orders across Nvidia, AMD, Apple. Even with NVDA's Kyber delay, TSMC benefits from AMD and Amazon ASIC ramps. TSMC is the common bottleneck. SoW-X packaging redefines logic-memory integration. The foundry leadership gap is not closing — Intel and Samsung still chasing.
AMD — Primary beneficiary of NVDA's Kyber delay and Rubin Ultra downgrade. MI450 and Helios becoming real second-source options. Analyst projects approaching $100B annual revenue by 2028. EPYC at 46% server CPU revenue share — massive gain from near zero. Venice ramping on TSMC 2nm keeps process competitive. The second-source narrative gaining traction. Key debate: can AMD capture 20%+ of AI GPU market?
MSFT — Mixed signals. Added $41B in data center leases in Q1 2026 (+26% QoQ), total ~$197B. Massive commitment. But Azure engineer says neoclouds' 'we have GPUs' pitch is not durable — Microsoft preparing bare-metal AI access. Curbed an AI coding assistant after costs became hard to justify — warning on enterprise AI monetization. Targeted layoffs at Xbox signal cost discipline.
META — Leading net new ad dollars for 4 consecutive years including YTD 2026. On current trajectory, larger than Google Search within two years. Added $79B in data center leases in Q1 2026 (+76% QoQ). First big tech to massively increase GPU capex in 2022-2023 — early bet paying off. The AI killer-app thesis is proven: recommendation and ad systems rebuilt on GPU infrastructure.
ORCL — Leads total data center lease commitments at ~$250B, securing sites for the OpenAI contract. This is hyperscaler scale, not just database company. OpenAI as anchor tenant drives sustained revenue. But liabilities are massive — overexposed if AI demand slows.
ADI — Lead times extending to 6 months, asking for earlier orders. Classic cycle bottom signal. Demand recovering, not just channel fill. Combined with tight 8-inch wafer capacity, pricing power returning. If cycle recovery confirmed, multiples can expand meaningfully from mid-cycle levels.
LITE / COHR / MRVL / GLW / FN / AAOI / SIVE — All in the VCSEL-based CPO supply chain discussion. InP laser supply bottleneck is driving shift to VCSEL. Ashkan Seyedi (ex-NVDA photonics lead) moving to ams OSRAM highlights talent shift. If VCSEL CPO gains traction, existing manufacturing footprints become valuable. Adoption timeline uncertain but the demand driver from AI clusters is strong.
TEL — Underlayer analysis shows ~50% dose-to-size reduction — EUV absorption may not be the primary exposure mechanism. Stochastic electrons from underlayer dominating has implications for High-NA EUV adoption economics. If throughput limited by underlayer effects, alternative lithography approaches benefit.
IFX — Raised prices 10-25% across product categories. Expects AI power solutions revenue to reach €1.5B. This is pricing power driven by AI demand for power management. Global semiconductor prices rising across categories. Supply-demand imbalance in power semiconductors.
Hearing Foxconn's record Q2 revenue (NT$2.51T, +39.8% YoY) is real — AI server demand now large enough to override normal ICT seasonality. The key debate shifts from demand to margin conversion and ROIC. Cloud & Networking nearly 50% of revenue now.
Word is Databricks' acquisition of Neon (managed Postgres) is a strategic fit for AI-native workloads — "the managed database of choice for the world's smartest model." Not yet in most estimates.
Channel checks suggest AI agent telemetry and log data volume is "staggering" — a new demand vector for enterprise storage that's not in any sell-side models. Disproportionately benefits SSD/HDD ecosystem.
Hearing NetApp (NTAP) wins are accelerating as agents flood customers with observability requirements. Pure Storage (PSTG) seeing similar phenomena. The enterprise storage cycle may have a structural AI component, not just replacement.
Word is Microsoft's Copilot cost issues are real — curbed an internal coding assistant after costs became hard to justify. This is a warning signal for enterprise AI monetization across the stack. Copilot adoption may face hurdles if cost per user exceeds value.
Hearing Microsoft's planned Xbox layoffs next week are a signal of cost discipline ahead of AI capex commitments. Not large scale but sentiment-negative for Xbox stakeholders.
Word is the neocloud 'we have GPUs' pitch is not durable — Azure engineer says Microsoft preparing bare-metal AI access that directly threatens neocloud valuations. If scarcity eases, the neocloud thesis breaks.
Channel checks suggest power architecture may deliver more marginal value than chip design at this stage. If you're an EE grad, go into power. Career and startup opportunity in data center power is underappreciated.
Hearing Western nuclear build costs are structurally 5-7x historical benchmarks. France/Korea at ~$2,000-2,500/kW are the only modern counterexamples — due to standardization and continuous orders the West no longer has. This matters for data center power timelines.