TMT Breakout晨报复盘PLTR、ON财报及美国限制中国光收发器传闻

Good morning. **QQQs +1% (SPX +35bps)** as futures rallying after [CNBC ](https://www.cnbc.com/2026/08/04/bessent-says-there-may-be-deal-tuesday-or-wednesday-to-open-strait-of-hormuz-with-freedom-of-movement.html)reported Bessen said there may be a deal Tuesday or Wednesday to open Strait of Hormuz with “freedom of movement.” Semis being helped by news that Trump banning Chinese optical transciever products which has **LITE/COHR/CIEN +10%** although some pushback on how feasible it is (we go into it below). **Semis +5%** **as a group while software is flat** (although ex-PLTR, down 1%). Hyperscalers giving some back early after a few days of outperformance.

Overnight, Asia was mixed: TPX +0.04%, NKY +0.32%, Hang Seng -0.6%, HSCEI -0.9%, SHCOMP +0.33%, Shenzhen +2.73%, Taiwan TAIEX -0.06%, Korea KOSPI +1.62%. Memory names up with **SK Hynix +6% and Samsung +4%.**

Things begin to ramp up on the earnings front today. We’ll be here covering it all.


We’ll hit **PLTR ON SPOT DOCN TSEM** earnings recaps first the onto the usual.

Let’s get straight to it…


### **PLTR +14%: A U.S.-led blowout, 2nd biggest beat on record, & accelerating revs helps keep accelerating narrative alive.**

Stock had been in the narrative dog house and positioning skewed short given some mediocre commercial checks heading in (remember Clev?) so these numbers much better than expected. It’s still unclear to me whether this print shifts the bull vs. bear debate in a material way (see below), but for the day definitely good enough.

Karp gave his usual enthusiastic spiel, positioning PLTR as the sovereign AI control layer for Western enterprises, letting customers orchestrate and fine-tune multiple models while keeping their data, prompts, reasoning traces and model weights inside their own security boundary, reiterating his usual critique of frontier labs saying that enterprises are paying open-ended token bills while effectively handing proprietary know-how to third-party models, whereas PLTR claims it can convert those models into measurable workflows while customers retain ownership of the resulting “alpha.”

Key quote on the growth rate: *“I am driving the business to grow at a rate equal or above to what we have in US commercial for the next 18 months, which is a very high goal, but it is one we can actually get to*” although on the call back mgmt said that was aspirational and not formal guidance and said said investors should interpret it as targeting 100%+ growth for combined U.S. revenue, said investors should interpret it as targeting 100%+ growth for combined U.S. revenue,

### The #s:

**Q2** Revenue was $1.936B, +93% y/y (last q +85%) vs Street $1.811B, +~80%; adjusted EPS was $0.41 vs Street $0.34, or roughly $0.39 excluding the $0.02 SpaceX mark-to-market benefit.

US Commercial was $764M, +149% y/y (last q +133%) vs Street ~$716M, +~134%, above the ~135–140% buyside bogey; US Government was $809M, +90% vs Street ~$733M, +~72%

Non-GAAP OM reached 61.7% vs Street 59.6%, adjusted FCF margin was 63.1%, NDR rose to 157%, and US Commercial TCV reached $2.132B, +153% y/y

**FQ3 revenue guidance** was $2.160–2.164B, +83% vs Street ~$2.000B, +~69%;

**FY26 revenue guidance** was raised to $8.150–8.158B, +82% vs Street ~$7.722B, +~73%, with US Commercial raised to >$3.424B, +134%+.

### **Key Takeaways:**

-

**AI** The new narrative is “sovereign AI”: customers retain control of their data, logic, actions, security, model weights and accumulated institutional intelligence. PLTR is extending AIP into model evaluation, orchestration and post-training through supervised fine-tuning and reinforcement learning, initially including classified deployments with NVIDIA’s open-weight Nemotron models. This positions PLTR above individual models rather than tying its fate to any one frontier lab.

-

**Competition** . Mgmt described a head-to-head engagement in which a frontier lab failed to generate value while PLTR’s AIP/FDE deployment converted into a $10M ACV contract. Customer checks also indicate large customers are not currently moving PLTR workloads or planned use cases to AI-native alternatives. The remaining risk is that OpenAI, Anthropic, AWS and MSFT are all investing heavily in FDE-style deployment organizations, while Databricks and Snowflake remain lower-cost alternatives for less complex workloads.

-

**Demand** The demand commentary was unusually strong. Mgmt described “massive demand” from CEOs and operating leaders seeking measurable AI returns and greater control over proprietary data, prompts, workflows and reasoning traces. The message was that enterprises are not cutting AI budgets broadly; they are becoming less willing to fund open-ended token consumption without demonstrable ROI.

-

**Growth **The growth engine is increasingly installed-base expansion. NDR rose to 157% from 150%, average trailing-12-month revenue from the top 20 customers increased 67% to $124M each, and customers that initially entered through pilots are converting into multi-year, enterprise-wide relationships. US Commercial RDV reached $6.24B, +124% y/y and +27% q/q, providing substantial contracted growth ahead.

-

**Customers** The clearest blemish was slower new-logo formation. PLTR added 42 total customers vs 53 last quarter and 80 a year ago, including 38 US Commercial additions vs 44 last quarter. US Commercial customer growth slowed to 35% from 42%, and the 38 additions were the lowest since 3Q24. The positive interpretation is that the customers being added are much larger and expansion is accelerating; the bear interpretation is that the long-term growth model is becoming more dependent on a relatively concentrated installed base.

-

**Government** US Government accelerated to 90% from 84% last quarter, driven by existing-program execution and new awards. Maven secured its first program of record that selected Maven as the operating platform for the entire program, its builder ecosystem exceeded 25,000 users, and trailing-12-month Department of War revenue remains less than 25 basis points of the Pentagon budget. That supports the argument that government scale is large in absolute terms but penetration remains very low.

-

**Guide **The quarter materially weakened the immediate competition bear case. Mgmt described a head-to-head engagement in which a frontier lab failed to generate value while PLTR’s AIP/FDE deployment converted into a $10M ACV contract. Customer checks also indicate large customers are not currently moving PLTR workloads or planned use cases to AI-native alternatives. The remaining risk is that OpenAI, Anthropic, AWS and MSFT are all investing heavily in FDE-style deployment organizations, while Databricks and Snowflake remain lower-cost alternatives for less complex workloads.

-

**Guide** The FY26 revenue raise of roughly $498M was considerably larger than the quarter’s dollar beat, while OM and FCF guidance were also raised. Karp’s 18-month statement was important but should not be treated as formal guidance: the callback clarified it as an aspirational goal for overall US revenue growth to exceed 100%, not a formal commitment for total company growth to remain at the current 149% US Commercial rate. Mgmt also cautioned that the 7.5% quarterly revenue beat is not the new baseline.

-

**Margins** Non-GAAP OM expanded to 61.7% from 60.2% last quarter and FCF margin reached 63.1%, showing exceptional incremental economics even as PLTR invests aggressively. Gross margin fell to 86.3% from 87.9% and missed Street by roughly 50 bps because PLTR assumed cloud-hosting costs for a government customer. Mgmt expects a meaningful FQ3 expense ramp from new-hire starts, R&D, product and marketing investments, which explains the modest sequential OM step-down in guidance.

-

**International** International Commercial remained at 26% growth, while International Government slowed to roughly 43% from 51%. Mgmt expects international results to remain relatively muted, and European political concerns around dependence on US technology remain a headwind in France, Germany and potentially the UK. This matters less while the US is growing more than 100%, but it limits the breadth of the current acceleration.

### **Bull vs. Bear Debate**

**The bull case** is that PLTR is not merely another application-software vendor adding an AI feature. It is becoming the enterprise operating and control layer that connects data, business logic, permissions, actions, models and employees. The Ontology supplies the semantic context that models generally lack, while AIP handles model routing, evaluation, observability, cost controls, post-training and deployment inside the customer’s security boundary. That makes PLTR model-agnostic. More open-weight models could therefore benefit PLTR by increasing model choice while raising the need for a governance and orchestration layer.

This quarter supplied unusually strong evidence for that thesis. Revenue accelerated for a 12th consecutive quarter, US Commercial accelerated to 149%, US Government accelerated to 90%, NDR reached 157%, and large-deal activity hit records. The most important qualitative proof point was the competitive engagement in which a frontier lab failed to deliver a useful outcome while PLTR’s AIP and FDE team generated a $10M ACV contract using the same customer, timeline and underlying models. Bookings and RDV also indicate that current revenue growth is not simply recognition timing: US Commercial TCV rose 153%, US Commercial RDV rose 124%, and cRPO rose 105%.

The government opportunity adds a second, largely independent growth engine. Maven is transitioning from an individual tool into a platform on which full programs are operated, while the 25,000-person builder ecosystem provides leverage beyond PLTR’s own headcount. With Department of War revenue still below 25 basis points of the Pentagon budget, a multiyear defense-modernization cycle could sustain growth even if commercial AI adoption eventually normalizes. At the same time, 61.7% OM and 63.1% FCF margin suggest that growth does not require sacrificing profitability

**The bear case** begins with the composition of growth. US Commercial customer additions declined to 38, the lowest since 3Q24, and customer growth slowed to 35% from 42%. Much of the current acceleration is coming from customers acquired two or three years ago expanding rapidly, not from an accelerating flow of new customers. That can be powerful for several years, but NDR cannot increase indefinitely, and a slowdown in a relatively concentrated installed base could produce a faster deceleration than the headline 149% growth suggests. The spread between $124M of average revenue among the top 20 customers and approximately $3.5M among the broader commercial base illustrates both the expansion opportunity and the concentration risk.

Pricing is another unresolved issue. Large customers report that PLTR is expensive enough that they must choose which workloads justify AIP and which can run on Databricks, Snowflake or another lower-cost platform. The direct frontier-lab win is encouraging, but OpenAI, Anthropic, AWS and MSFT are all building or financing deployment organizations that increasingly resemble PLTR’s FDE motion. Even when these efforts do not displace PLTR outright, they could pressure pricing, narrow the talent advantage or reduce the number of workflows for which customers require the full PLTR stack.

The setup also becomes mathematically harder. PLTR faces blended revenue-growth comparisons of roughly 67% in 2H26 and 89% in 1H27. International growth is already substantially below the US, European political sovereignty remains a headwind, and the company has acknowledged that scaling delivery capacity is a constraint. Gross margin declined sequentially, operating expenses are expected to ramp in FQ3, SBC is expected to run near 15% of revenue, and $0.02 of the quarter’s adjusted EPS came from an unrealized SpaceX gain. The callback additionally cautioned that investors should not treat the 7.5% revenue beat as a recurring cushion. Bears will still harp on valuation despite it coming down significantly

### **ON +6%: ON delivered an AI data center-led beat/raise and a clear gross-margin inflection, but main nit is top-line beat was narrow because Auto and Industrial both missed Street. Expectations had been lower after NXPI/STM misses, so this should be good enough**

Revenue was $1.604B, +9.2% y/y (last q +4.7% y/y) vs Street $1.587B, +~8.1%; non-GAAP GM was 39.3% vs 39.1%; and EPS was $0.74 vs $0.71. Utilization rose to 83%, FCF reached $425M, channel inventory fell, and mgmt said the utilization benefits have not yet fully flowed through the P&L.

Q3 revenue was guided to $1.65-1.75B, or $1.700B at the midpoint, +9.6% y/y vs Street $1.669B, +~7.6%; GM was guided to 41.0% vs 40.3%, and EPS to $0.87 vs $0.84.

AI data center was the clear positive, driving Other revenue +34% q/q and prompting ON to raise its CY26 outlook from doubling to more than doubling, while Auto and Industrial remained softer than expected.

### **Key Takeaways:**

-

**AI DC** ON’s AI data center business is accelerating faster than expected, with CY26 revenue now expected to more than double vs the prior expectation to double. Other revenue was $400M, +34% q/q vs Street roughly $333-346M, and Q3 Other is expected to rise high-teens q/q. Wins include an expanded role in NVIDIA MGX, AWS power-supply and battery-backup content, and two Great Wall power-supply platforms in China. Mgmt characterized the strength as sustained rather than a one-quarter spike.

-

**Demand** The broader cyclical signals improved meaningfully: book-to-bill has remained significantly above 1 and strengthened further, average lead times extended from roughly 27 to 32 weeks, orders inside lead time and customer escalations increased, visibility now extends into 2027, and some customers are placing orders into 2028 to secure supply. These are stronger signals than the actual Auto and Industrial revenue results would suggest

-

**Core** The main negative was that almost all of the upside came from data center. Auto revenue of $781M, -2% q/q, was below Street around $800M, while Industrial revenue of $423M, +1% q/q, was below Street around $440-452M. Q3 Auto is guided up only low-single digits and Industrial flat, both materially below prior Street expectations, while Other is guided well above. This leaves investors debating whether ON is experiencing a broad recovery or an AI-specific surge masking a still-soft core.

-

**Allocation** Mgmt acknowledged that constrained power capacity was prioritized toward AI data center, taking revenue away from Auto and Industrial in the short term. Mgmt expects manufacturing output to catch up during Q3 and Q4 and said the allocation decisions did not create a meaningful near-term customer impact. Bulls see this as evidence that the Auto/Industrial misses were supply-driven; bears want to see the deferred revenue actually return.

-

**Margins** Non-GAAP GM rose 80 bps q/q to 39.3%, and Q3 GM was guided to 41.0%, 70 bps above Street and 170 bps above Q2. Importantly, Q2 still reflected the lower 4Q25 utilization level because manufacturing utilization takes roughly two quarters to reach the P&L. Utilization increased from 68% in 4Q25 to 77% in Q1 and 83% in Q2, with each utilization point worth roughly 25-30 bps of GM. That gives ON a fairly visible margin tailwind into Q4 even without another utilization increase.

-

**Pricing** Q2’s margin improvement was driven primarily by mix and manufacturing performance, not pricing. ON is implementing a second round of price increases across end markets, including Auto, to offset higher costs for substrates, gold, raw materials and external manufacturing. Much of the benefit should arrive in Q4 and subsequent quarters, and mgmt does not currently expect a softer pricing environment or normal annual price reductions in 2027.

-

**Auto** Auto was weaker sequentially because of European customer seasonality and AI DC supply prioritization, but China remained strong. China Auto revenue grew 13% in 1H26 despite total vehicle sales declining 4%, and China Auto SiC revenue is expected to grow 60-70% in CY26 through share gains and platform ramps. ON also added power content on Rivian’s R2 platform across zonal control and onboard charging. The debate is whether these company-specific wins can offset a still-muted global Auto production environment.

-

**Industrial** Industrial strength remained concentrated in energy infrastructure, medical and factory automation, while traditional Industrial stayed weak. ON expects ESS revenue to grow approximately 40% y/y in CY26, helped by North American microgrids and AI-related power infrastructure. However, the Q3 flat sequential guide remained well below Street and was arguably the weakest part of the outlook.

-

**Inventory** The demand quality looked healthy. Total inventory declined nine days to 192 days, strategic inventory fell eight days, base inventory declined to 125 days, and channel inventory fell to 10.1 weeks from 10.8 weeks as sell-through exceeded sell-in. Mass-market revenue rose 20% q/q, which mgmt views as a leading indicator, and mgmt said Auto and Industrial shipments are aligned with end demand rather than inventory restocking.

### **Bull vs. Bear Debate**

**The bull case** is that ON is transitioning from a cyclical Auto/Industrial analog supplier into a broader AI power platform just as its legacy businesses and margins begin recovering. AI DC revenue is expected to rise from approximately $250M in CY25 to more than $500M in CY26, with some forecasts closer to $600-750M. The opportunity is not dependent on one SiC component: ON has content across high-voltage power conversion, silicon carbide MOSFETs and JFETs, silicon products, battery backup, power supplies and lower-voltage delivery near the XPU. That breadth reduces product concentration and means ON can benefit both from more AI racks and sharply higher power content per rack.

This quarter supplied the first concrete evidence that the thesis may be working. Other revenue grew 34% q/q, Q3 Other is guided up high-teens, AI DC guidance was raised, and wins are now converting into revenue. The longer-term 800V transition could move ON’s content per rack from roughly $15K to $115K, with high-voltage content rising from approximately 30% of the current rack to 50% of the future rack. Mgmt expects that transition to begin around late 2027 or early 2028, creating a multi-year runway rather than a single-product cycle.

The second part of the bull case is earnings leverage. ON has already removed approximately 12% of capacity, completed $900M of annualized non-core revenue exits, constrained capex and taken utilization from 68% to 83%. Because utilization reaches the P&L with a two-quarter delay, the Q2 loading increase should support Q4 margins. Pricing, Fab Right savings, lower depreciation and richer AI mix add further upside.

**The bear case** is that AI DC is doing all the work while roughly three-quarters of the business remains underwhelming. Auto and Industrial both missed Street, and their Q3 guides were substantially below prior expectations despite extremely positive commentary on orders, lead times and the recovery. Even assuming AI DC reaches $600M in CY26, it would represent only around 9% of projected annual revenue. That may not be large enough to offset weakness in Auto, traditional Industrial and the non-AI consumer and communications businesses embedded within Other. Furthermore, mgmt acknowledged that not all of Other’s 34% q/q growth came from data center, making the precise size and trajectory of AI DC difficult to verify.

Bears also question whether supply allocation fully explains the Auto and Industrial shortfalls. If demand was as strong as book-to-bill, lead times and escalations imply, those businesses should eventually show much better revenue. Until that happens, capacity prioritization can be viewed as a convenient explanation for a core recovery that is lagging peers. Auto remains exposed to global production, China EV competition and SiC share dynamics; Industrial remains exposed to a weak traditional factory and distribution backdrop. ON’s inventory also remains high at 192 days even after the sequential reduction.

The margin bridge is attractive but demanding. Moving from 39.3% GM to the long-term 53% model requires roughly 1,370 bps of expansion. Pricing is currently offsetting costs rather than adding margin, Q3 Opex is around $10-15M above Street, and input costs are not expected to decline in 2027. The Synaptics acquisition adds debt, integration risk and strategic complexity, with some analysts expecting the meaningful benefits to take until approximately 2029. Bears therefore argue that ON is asking investors to underwrite several things simultaneously: an Auto/Industrial recovery, sustained AI DC growth, manufacturing execution, higher pricing and a large acquisition.

### **SPOT -4%: Mixed with main concern softer user growth as MAUs missed street and Q3 guide came in 6M below**

SPOT’s print was mixed, with the main concern centered on engagement. Q2 MAUs missed Street and the 3Q guide came in 6M below, while premium subscriber adds are expected to slow to 5M versus Street at 6M. Paid conversion remained healthy, with Q2 premium subscribers modestly above Street, but weaker MAU momentum likely weighs on the stock given positioning that leaned slightly long heading in.
On the bright side: GM and op income beat Street, while 3Q revenue guidance was modestly above expects, largely due to FX. However, 3Q operating income came in below Street and gross margin is expected to decline 50bps q/q. Call ongoing, but they will need need to provide confidence around MAU trends, opex discipline and the timing of any financial contribution from Spotify’s AI initiatives, including the new Merlin agreement.

The #s:

**Q2 MAUs:** 777M, +11% y/y vs Street 778M
**Q2 Premium Subscribers:** 300M, +9% y/y vs Street 299M; net adds 7M
**Q2 Revenue:** €4.78B, +14% y/y (+15% FXN) vs Street €4.80B
**Q2 Gross Margin:** 33.4%, +193bps y/y vs Street 33.1%
**Q2 Operating Income:** €655M vs Street €639M; €646M excluding favorable social-charge adjustment
**Q3 MAU Guide:** 788M vs Street 794M; net adds 11M
**Q3 Premium Subscriber Guide:** 305M vs Street 305M; net adds 5M vs Street 6M
**Q3 Revenue Guide:** €5.0B vs Street €4.93B, including an approximately 200bps FX tailwind
**Q3 Gross Margin Guide:** 32.9%, -50bps q/q vs Street 32.8%
**Q3 Operating Income Guide:** €670M vs Street €678M

### **DOCN -11%: Beat-and-Raise, Though Much of the Upside Was Already Expected**

DOCN delivered strong Q2 results and raised FY26 revenue, margin and EPS guidance, supported by accelerating AI demand, larger contracts and record incremental ARR. AI ARR grew 212% y/y, RPO increased 12x and management flagged 50%+ revenue growth for 2027, but the quarter was preannounced and investors had already been discussing similar FY26 targets

The #s:

**Q2 Revenue:** $281M, +29% y/y vs Street $279M
**Q2 Adjusted EBITDA:** $113.6M vs Street $106.4M
**Q2 Adjusted EPS:** $0.45 vs Street $0.26
**AI ARR:** $259M, +212% y/y
**RPO:** $894M, up 12x y/y
**Q3 Revenue Guide:** $304M-$307M vs Street $300M
**Q3 Adjusted EBITDA Margin Guide:** 38%-39% vs Street 38.4%
**Q3 Adjusted EPS Guide:** $0.28-$0.30 vs Street $0.29
**FY26 Revenue Guide:** $1.17B-$1.18B, +30%-31% y/y vs prior $1.13B-$1.15B and Street $1.16B
**FY26 Adjusted EBITDA Margin Guide:** 38.5%-39.5% vs prior 37%-39% and Street 38.6%
**FY26 FCF Margin Guide:** 11%-13% vs prior 9%-12% and Street 10.1%
**FY26 Adjusted EPS Guide:** $1.35-$1.40 vs prior $1.10-$1.20 and Street $1.26

### **TSEM -1%: Clean Beat, Big Q3 Guide and Higher SiPho Targets**

TSEM delivered record revenue and profitability, with Q2 revenue and adjusted EPS above Street and Q3 revenue guided roughly 6% higher than expectations. The main upside driver is Silicon Photonics, where revenue reached a $680M annualized run rate, up from $180M a year ago, with management targeting more than $1B by 4Q26 and continued strong growth through 2027.

Management also raised its 2028 model to $3.6B of revenue and $1.2B of net income, saying the capacity is fully committed by customers. The trade-off is elevated investment: Q2 capex of $187M exceeded operating cash flow of $177M as TSEM expands capacity against demand.

**Q2 Revenue:** $460.1M, +24% y/y vs Street $453.3M
**Q2 Adjusted EPS:** $0.88 vs Street $0.76
**Q2 Gross Profit:** $138M, +72% y/y; gross margin 30.0% vs 21.6% last year
**Q2 Operating Profit:** $90M, +125% y/y; operating margin 19.6%
**Q2 Net Profit:** $91M, +95% y/y; net margin 20%
**Q3 Revenue Guide:** $520M, +31% y/y and +13% q/q vs Street $490M
**SiPho Run Rate:** $680M vs $180M last year; targeting $1B+ in 4Q26
**2028 Target:** $3.6B revenue and $1.2B net income

# **TECH RESEARCH/NEWS**

### **AMZN: AMAZON'S BEZOS FILES TO SELL $4.07B OF SHARES**

### **Optical: Trump administration drafting ban on Chinese data center devices**

**[Reuters](https://www.reuters.com/world/trump-administration-drafting-ban-chinese-data-center-devices-sources-say-2026-08-04/):**
**

The Trump administration is drafting a ban on U.S. imports of **new** models of Chinese data center components, four people familiar ​with the matter told Reuters, as it seeks to protect the infrastructure that undergirds the AI boom.

The Federal Communications Commission, which oversees the U.S. telecom ‌industry, is working on the measure to bar imports of new Chinese optical transceivers, which allow data to travel over fiber-optic cables at the speed of light within data centers. Officials hope to publish it this year, when it would take effect.

**LITE +11%, COHR +13%, FN +13% and AAOI +15**% are the main beneficiaries here, but the rest of the space is also up. Remember, LITE’s CEO has been on podcasts arguing that “it’s unfair China can sell here, but we can’t sell there,” while COHR’s CEO recently went to China. The ***new*** models language imply 1.6T instead of 800G. The pushback is that there may not be enough optical capacity to serve the data center market without Chinese supply—Innolight alone has roughly 50% of NVDA’s transceiver wallet share, and it is unclear what happens to InP supply. It is also unclear whether any restrictions would be company-based or country-based, relevant bc LITE and COHR have manufacturing exposure in China. Memory bulls hoping this is just a precursor for CXMT/YMTC getting banned.

### **CRWV/NBIS: Piper Initiates CoreWeave Overweight and Nebius Neutral as Neocloud Demand Outpaces Supply**

Piper initiates CRWV at Overweight with a $151 price target and NBIS at Neutral with a $224 target, saying both should benefit from an AI compute market expected to grow at a 56% CAGR to $109B by 2029 as demand remains ahead of supply. Piper prefers CRWV because its engineering stack, multi-cloud capabilities and financing structure reduce cost and execution risk, while Piper says NBIS’s asset-light model and full-stack platform are attractive but carry more near-term estimate risk around the Vineland data center. Piper adds that both neoclouds can price more than 40% below hyperscalers, though Piper notes customer concentration and heavy capital requirements remain the key risks.

### **SKHY: RBC Initiates Outperform with $200 PT as HBM Leadership Extends the Memory Upcycle**

RBC initiates SK Hynix at Outperform with a $200 price target, saying generative AI has structurally strengthened the memory cycle and should extend the upturn through 2027. RBC says SK Hynix’s sustainable HBM leadership provides an added advantage, while 2027 contract repricing should offset potential moderation in broader DRAM pricing. RBC also believes the stock’s meaningful valuation discount to U.S. peers creates an attractive risk-reward.

### **SKHY: Stifel Initiates Buy with $240 PT as HBM Leadership Supports a Multi-Year Memory Upcycle**

Stifel initiates SK Hynix at Buy with a $240 price target, saying AI has structurally strengthened memory demand and made DRAM a critical constraint for global data-center buildouts. Stifel says SK Hynix’s roughly 60% HBM share and technology leadership should support continued gains, while clean-room and equipment constraints could keep 2027 industry bit growth near 20%, well below unconstrained demand. Stifel also says SK Hynix remains undervalued despite its recent U.S. listing and sees room for a valuation rerating toward 7x NTM earnings as HBM leadership and durable supply tightness become better appreciated.

### **MU: BofA Reiterates Buy and $1,550 PT as the Pullback Discounts an Overly Harsh Memory Downturn**

BofA reiterates Buy and a $1,550 price objective, saying MU’s recent pullback creates an attractive entry despite memory pricing likely normalizing from mid-2027 onward. BofA models CY28 EPS near $150 under its base case and says even a severe downturn with DRAM/NAND ASPs down 30%/40% could still leave EPS around $100, far above the prior-cycle peak near $12. BofA also says multi-year LTAs should reduce volatility, CXMT remains largely irrelevant for HBM and U.S. AI demand, and near-record GPU rental rates and rising hyperscaler capex confirm that AI memory utilization and pricing power remain strong.

### **Memory: 2027 memory capacity reportedly sold out as buyers quietly lock in supply**

**[Digitimes](https://www.digitimes.com/news/a20260804PD217.html?mod=2&utm_source=newsletter&utm_medium=email&utm_campaign=daily_picks_paid&utm_id=DailyPicksPaid&utm_content=article): **
**

According to the sources, 2027 DRAM and high-bandwidth memory (HBM) capacity at the three leading memory suppliers has effectively been fully allocated ahead of schedule. NAND Flash is not yet as constrained as DRAM, but much of next year’s available capacity is expected to be reserved by the end of August 2026.

Whether or not buyers have signed formal long-term agreements, suppliers are increasingly requiring advance deposits to secure allocations. The trend supports industry expectations that 2027 could become the most severe year of the current memory shortage, although the pace of price increases may moderate.

### **Memory: SK Hynix, Sandisk release first HBF specs with up to 512GB and 3TB/s bandwidth**

**Digitimes:**
**

SK Hynix and Sandisk released the first standard specifications for High Bandwidth Flash (HBF) on August 4, defining a NAND-based memory technology with a capacity of up to 512GB and bandwidth of up to 3TB/s.

According to the companies, the specifications cover two configurations using 8-high and 16-high NAND die stacks, with bandwidth divided into three grades ranging from about 0.4TB/s to 3TB/s.

HBF uses Universal Chiplet Interconnect Express (UCIe) to connect to processors. UCIe is an open interface standard that enables high-speed connections between different semiconductor chiplets, allowing HBF to work with processors including GPUs and CPUs.

### **Datacenters: Data center approvals in Texas halted until audits completed, Gov. Greg Abbott says**

**[TexasTribune:](https://www.texastribune.org/2026/08/03/texas-data-center-project-audit-greg-abbott/)**
**

Amid growing public backlash to the rapid buildout of data centers across Texas, Gov. Greg Abbott on [Monday announced a moratorium](https://gov.texas.gov/news/post/governor-abbott-directs-comprehensive-data-center-audit) on the approval of data centers until regulatory agencies can audit proposed data centers seeking connection to the state’s electric grid.

It is unclear how long the audit will take, but the effort is Abbott’s strongest effort yet to slow the development of these data centers statewide and underscores how little information members of the public, including state leaders, know about the facilities.

### **LRCX/KLAC/AMAT: BofA Raises WFE Outlook as Earnings Leverage Remains Underappreciated**

BofA raises its CY26 WFE forecast to $150B–$160B, +30% y/y, from roughly $140B–$145B, citing DRAM/NAND conversions, 2nm/3nm logic and advanced packaging. BofA says the group can compound EPS at roughly 30% through CY28, yet many semicap and subsystem names remain 20%–40% below their highs and trade near a 0.7x three-year PEG. BofA adds that higher NAND layer counts could lift equipment intensity per wafer toward 2x, while conversion-driven additions are unlikely to create excess supply through CY28; BofA’s preferred names are LRCX, AMAT, MKSI and TER.

### **LITE: Mizuho Reiterates Outperform and $1,100 PT as InP Supplier Checks Point to September-Quarter Upside**

Mizuho reiterates Outperform and a $1,100 price target, saying supplier checks from AXTI, Landmark and SEI suggest its $1.13B September-quarter revenue estimate, +15% q/q, may prove conservative. Mizuho says InP capacity is expanding rapidly while demand for 1.6T/3.2T EMLs, NPO, OCS and CPO lasers supports a multi-year growth runway and improving mix, with gross margin potentially moving above 50%. Mizuho views LITE’s roughly 27% pullback as a tactical entry ahead of stronger 2H26 momentum, while Mizuho says valuation remains attractive relative to projected FY26/FY27 EPS growth.

# **SOFTWARE:**

### **INTU: Truist Downgrades to Hold and Cuts PT to $350 as Growth Headwinds Persist**

Truist downgrades INTU to Hold from Buy and cuts its price target to $350 from $410, saying management commentary points to a J-curve in growth, persistent DIY tax pressure, flat Mailchimp trends and low-single-digit declines in the GBS desktop business. Truist lowers F4Q26/FY27 revenue estimates to $4.25B/$23.74B from $4.27B/$23.85B, while Truist raises FY27 EPS modestly to $27.34 on cost discipline and buybacks. Truist still views Intuit’s assisted tax, money portfolio and mid-market initiatives as strong long-term assets, but Truist says the lack of a near-term catalyst and limited visibility into reacceleration justify moving to the sidelines.

### **TEAM: Hearing Clev saying Q4 appeared in line (better vsprior) as headwinds from macro/AI/cuts to partner program stabilized**

### **HUBS: 3p said enterprise net adds remain weak while SMB seat growth remains healthy**

### **TTD: Hearing Clev negative saying share loss expected to worsen next 6-12 months**

### **NET: Barclays Says Traffic Inflection and PoF Ramps Support 2Q Upside, with Monetization Catch-Up Still Ahead**

Barclays models roughly 30% 2Q revenue growth and 31% cRPO growth, saying rapidly rising bot and agent traffic should eventually translate into upside as customers exceed request caps and upgrade contracts. Barclays says the lag between traffic and revenue is beginning to narrow, while the large 4Q25 Pool of Funds cohort should lift both revenue and DBNR, which Barclays expects to improve toward 119% from 118% in 1Q. Barclays also expects gross margin to stabilize near 72% and EBIT margin near 14%, and Barclays believes Cloudflare remains on track for a Rule of 50 profile by 2027 despite continued mix pressure from lower-margin Act 3 products.

### **Interesting X post ([link](https://x.com/bradmenezes/status/2084328970649026710?s=20)) getting flagged below (h/t John in Slack post-close, who made a good point that “months ago something like this would have wrecked SaaS”)**


# **INTERNET**

### **EBAY: Citizens Downgrades to Market Perform as Vinted’s U.S. Push Raises Depop Profitability Risk**

Citizens downgrades EBAY to Market Perform from Outperform and removes its price target, saying Street estimates look too high as Vinted sharply increases U.S. investment around eBay’s Depop acquisition. Citizens notes Vinted’s paid-search spend rose roughly 15x from February to June and its App Store ranking improved from around 200 to No. 5, while Citizens says Vinted’s dominant U.K. position shows the potential share risk. Citizens acknowledges eBay’s core marketplace and Collectibles execution remain solid, but says the likely cost of defending Depop’s growth makes the risk-reward more balanced.

### **DUOL: BofA Downgrades to Underperform and Cuts PT to $93 as Valuation Outruns Product Validation**

BofA downgrades DUOL to Underperform from Neutral and cuts its price target to $93 from $103, arguing the stock’s roughly 20% rally since early June appears driven more by a one-time promotion than durable product improvement. BofA expects a 2Q bookings beat but says 3Q should be roughly in line, while BofA cuts CY27 bookings and EBITDA to $1.34B/$338M from $1.38B/$388M. BofA says DUOL’s roughly 16x CY27 EBITDA multiple is difficult to justify versus subscription peers near 10x until the company proves its broader product strategy can sustain long-term acceleration.

### **BKNG/ABNB/EXPE/TRIP: Bernstein Says Weak July Traffic Could Pressure 3Q OTA Guides**

Bernstein says July traffic slowed from 2Q across most OTAs, which should drive cautious 3Q guidance even if 2Q room nights and EBITDA beat. Bernstein notes BKNG app engagement slowed to 4% growth, ABNB web and app trends decelerated, and EXPE app engagement fell 4% y/y, with VRBO DAUs down 13%; Bernstein says late-July improvement may soften but not eliminate the concern. Bernstein highlights TRIP’s Viator as the outlier, with July app engagement growth accelerating to 46%, and says World Cup timing, renewed Middle East conflict and severe European weather may explain part of the slowdown.

### **APP: Citi Says U.S. eCommerce Store Growth Softened as Hong Kong and Germany Accelerated**

Citi says AppLovin Ads reached 10,837 eCommerce customers as of July 31, up 1.2% w/w, but Citi notes U.S. stores declined 0.2% while Germany and Hong Kong accelerated. Citi says 3Q-to-date weekly store growth is averaging roughly 1.0%, implying about 11.1K stores for the quarter, below the 2Q pace. Citi also cautions that part of the Hong Kong growth may reflect concentrated dropshippers repeatedly installing and uninstalling AppLovin Ads rather than broad-based customer expansion.

### **PDD: BNP Says EU Parcel-Tariff Changes Are Hitting Temu Harder and Could Create a Longer-Lasting Drag**

BNP says the EU’s removal of small-parcel exemptions drove July MAUs down 31% y/y for Temu, 37% for SHEIN and 8% for AliExpress, with Temu showing the sharpest deterioration through the month. BNP says the impact is also visible in logistics, with Hong Kong-to-Europe air-cargo volumes down 22%–23% y/y in late July and APAC-to-Europe spot rates falling 12% since June. BNP believes Temu and SHEIN are more exposed than AliExpress and expects the EU impact to persist longer than the comparable U.S. de minimis change because of greater regulatory complexity.

### **AAPL: Morgan Stanley Sees a Potentially Conservative September Guide but Flags Rising AI Costs and Services Risk**

Morgan Stanley says Apple’s September-quarter outlook may be intentionally conservative, with management’s cited 3.5-point supply headwind potentially overstating near-term Mac constraints and creating room for an upside earnings beat. Morgan Stanley also says Apple’s new $1/day leasing model could accelerate annual upgrades and support a higher-LTV subscription narrative. However, Morgan Stanley notes inventory rose sharply as Apple secured supply amid semiconductor and memory inflation, while slowing App Store growth and rising AI infrastructure costs could pressure margins if monetization lags.

### **AAPL: Apple downgraded to Hold from Buy at DZ Bank**

### **AAPL: Apple downgraded to Hold from Buy at China Renaissance**