Equity research

Meet Lee Seung-ho, a 24-year-old South Korean university student in Seoul, who lost all of his $202,515 earned via 500% margin loan. He plans to borrow again and return to the market the moment he has enough capital.

Within weeks, his account had fallen below his initial investment, leaving him under such strain that he said:

"I literally could not breathe."

Did he learn his lesson? Nope.

Highlights of an expert call from a Senior Director at ARM who was a Former Director at Western Digital regarding memory:

Unique Nature of the Current Memory Upcycle

> AI Data Center Expansion: Unlike past commodity-driven cycles (such as 2017–2018 or 2021–2022), this upcycle is uniquely propelled by sustained, multi-year AI data center buildouts from hyperscalers like Amazon, Microsoft, and Meta.

> Longer Cycle Duration: While past cycles typically flipped every 18 months, strong demand and constrained supply mean suppliers are already largely sold out through 2027, with market softening potentially expected around 2028 as new fabs come online.

Memory Bandwidth & Capacity Constraints (Training vs. Inference)

> Training Phase: Demand for HBM, LPDDR, and storage remains robust to support trillion-parameter models, though the baseline of model creation is becoming more established.

> Inference & The Agentic World: Inference is the primary revenue driver (exemplified by Micron's performance) and the "token economy". As multi-user agent systems scale, demand for faster token generation explodes, creating a "double whammy" of high demand for both GPU-attached HBM and CPU-orchestrated LPDDR bandwidth.

Supply Growth, Market Share, and Competitors

> Current Market Shares:

HBM Share: SK Hynix leads at approximately 58% to 60%, followed by Samsung and Micron at around 20% each.

DRAM Share: Samsung leads overall with roughly 40%, followed by SK Hynix (30%), Micron (20%), and others/China vendors (10%).

> Strategic Allocation: Vendors are balancing high-margin LPDDR capacity (driven by tight markets in smartphones and PCs) alongside HBM.

> The CXMT Threat: China's CXMT is expected to capture growing single-digit market share, primarily focusing on older nodes and LPDDR products for domestic data centers and smartphone brands (Xiaomi, vivo, OPPO).

Long-Term Supply Agreements (LTAs) & Pricing Outlook

> Shift in LTAs: Data center customers are pushing for 3-year or even 5-year LTAs to secure guaranteed supply, giving memory vendors stronger pricing power compared to historical 1-year agreements.

> Pricing Trajectory:

HBM/DRAM/NAND: Broad consensus expects prices to trend upward through 2026 and 2027 due to supply constraints.

HBM Pricing Trajectory: Blended average selling prices (ASPs) per stack are estimated around $330 to $360 for HBM3E in 2026, projected to rise to about $500 for HBM4 in 2027 as it gets introduced, and expected to soften slightly to around $450 for HBM4E in 2028 as supply catches up.

Company Optimism & Outlook

> SK Hynix: Highly praised for its market leadership in HBM, though integrating logic dies for future iterations poses a competitive frontier.

> Samsung: Viewed favorably due to an inherent margin and pricing advantage via its in-house logic foundry (Samsung Foundry), positioning it to potentially claw back some HBM share in the 2027–2028 timeframe.

> Micron: Expected to maintain its steady ~20% market share while benefiting strongly from current supply-constrained margins.

$Micron Tech(MU.US) $EWY $DRAM

Cantor Fitzgerald: Intel & AMD

Intel ($Intel(INTC.US))

> Outlook: Cantor maintains a "Neutral" rating with a $150 price target, noting that the risk/reward profile into earnings is skewed to the upside.

> Drivers: The company is expected to see a "modest beat/raise" in its upcoming results, fueled by demand for agentic AI and expansion in average selling prices (ASP) within its Client segment.

> Key Focus: Analysts are closely monitoring Intel's ability to outperform in a supply-constrained environment and how a potential decline in Client unit sell-through during the second half of the year will impact the company.

> Foundry Progress: There is optimism regarding progress on 18A-P and 14A nodes, alongside developments with both current and new customers. This is expected to lead to a forecast for significantly higher capital expenditure (Capex) in 2027.

AMD ($AMD(AMD.US))

> Upcoming Event: AMD will host its third annual Advancing AI 2026 Event on July 22–23 in San Francisco.

> New Tech: The company is expected to launch its next-generation Instinct MI450-series GPUs and Zen 6-based EPYC "Venice" processors.

> Infrastructure Growth: AMD is anticipated to provide updates on the third-quarter ramp of its "Helios" rack-scale platform and offer more context on a "rapidly expanding pipeline" of large-scale AI infrastructure opportunities.

> Market Projections:

AMD is expected to raise its view on the Server CPU Total Addressable Market (TAM) for 2030 to over $150B (up from $120B).

There is anticipation for an updated view on the AI Accelerator TAM, with potential upside to $600B–$800B in the 2029/2030 timeframe.

> Financial Outlook: Cantor believes AMD is positioned to guide for data center revenues to at least double in 2027. AMD remains a "Top Pick" with an "Overweight" rating and a $700 price target.

J.P. Morgan: TSMC CoWoS

CoWoS Capacity Expansion

> Capacity Increase: J.P. Morgan is raising its CoWoS capacity estimates for 2026–2028 by 1%, 9%, and 11% respectively.

> Targets: TSMC’s CoWoS capacity is expected to reach 115k, 190k, and 225k wafers per month (wfpm) by the end of 2026, 2027, and 2028.

> Strategic Shifts: TSMC is accelerating expansion at AP7 Phase 2 (reallocating capacity from SoIC to CoWoS) and is selectively outsourcing interposer manufacturing to Vanguard.

> OSAT Role: Capacity for CoWoS-like processes at Outsourced Semiconductor Assembly and Test (OSAT) providers is expected to reach 15k, 50k, and 85k wfpm by the end of 2026, 2027, and 2028, primarily to support server CPU packaging.

3D SoIC and CoPoS Outlook

> 3D SoIC: The capacity ramp is now more focused on 2028 (projected 65k wfpm) to align with mass adoption in accelerators. While 2027 may see a slight push-out to accommodate CoWoS demand, key customers like AWS, OpenAI, Meta, and NVIDIA (Feynman versions) are expected to adopt SoIC in 2028.

> CoPoS: This technology is currently in the engineering stage. Risk production is expected in early 2028, with mass production likely in 2029.

Customer Demand and Forecasts

> NVIDIA: CoWoS allocations for NVIDIA are rising, reflecting strong demand for new CPUs (Vera in 2027, Rosa in 2028) and accelerators. All NVIDIA accelerator and Co-Packaged Optics (CPO) products are on CoWoS-L at TSMC.

> AMD: Estimates for AMD are rising significantly in 2027–2028, driven by the Venice server CPU family, which will largely utilize a CoWoS-L-like process from ASE.

> CPUs and ASICs: Server CPUs (e.g., NVDA Vera, AMD Venice) have emerged as a primary upside catalyst for CoWoS demand in 2027. Other programs like Google’s TPU and AWS’s Trainium continue to ramp, though ASICs for companies like Microsoft and Meta remain at a smaller scale.

> Supply Chain Roles: AWS and Alchip are expected to remain the primary suppliers for the Neuronlink and UALink versions of Trainium 4. Marvell may take on some share for specific variants that utilize "NVLink fusion" based interconnects. While Trainium 4 volumes in 2028 are expected to be relatively flat compared to Trainium 3 volumes in 2027, the wafer consumption and average selling prices (ASPs) for these chips are expected to be significantly higher.

> 2027 as a Key Ramp Year: While TPU and Trainium are currently the only large-scale ASIC programs, 2027 is viewed as a key year for others to ramp, with a heavy focus on Meta’s MTIA and Microsoft’s Maia.

Other Technological Notes

> Intel EMIB: Intel’s EMIB-T process is seeing incremental demand, notably for the MediaTek TPUv9 Humufish. However, substrate capacity remains a limiting factor for this process. Intel’s EMIB process is positioned as a cheaper alternative, with costs estimated at approximately 50% of TSMC's CoWoS-L packaging technology. Current substrate-level yields for the EMIB process are around 60%, which is considered acceptable by the researchers for a process 1–1.5 years away from ramp. However, substrate capacity is viewed as a key limiting factor for capacity expansion because the processes for substrate build-up and embedding the silicon bridge steps currently have low yields.

$Intel(INTC.US) $Alphabet(GOOGL.US) $AMD(AMD.US) $NVIDIA(NVDA.US) $Taiwan Semiconductor(TSM.US)

Speedrun | Upcoming Events This Week - [July 20 - July 24, 2026]

Stay informed on events/news for the coming week.

Geopolitical & Market Outlook

> Iran Conflict: Tensions have escalated following an Iranian drone attack on a U.S. base in Jordan that resulted in the deaths of two U.S. service members. U.S. forces have responded with strikes on Iranian military coastal targets for eight consecutive nights.

> Energy Prices: The conflict has pushed crude oil prices above $80/barrel from a low of $68.55 on July 6th. This rise in energy costs poses a challenge to the Federal Reserve’s outlook and may impact future Consumer Price Index (CPI) reports.

> Momentum Unwind: High-beta momentum stocks have experienced a significant sell-off, with Goldman Sachs' High Beta Momentum basket dropping 32.8% in just three weeks since its June 22nd peak. This drawdown is compared to the market environment in late 2021, just prior to the 2022 bear market.

Earnings Calendar

This is a major week for earnings releases. Key companies to watch include:

> Google ($Alphabet(GOOGL.US)): Investors are focused on AI ROI, capital expenditure (CapEx) trends for 2026/2027, and memory cost pressures. There is concern that the market is not properly pricing in higher CapEx spending.

> Intel ($Intel(INTC.US)): As a major semiconductor company reporting this week, Intel will be watched for CPU demand trends related to agentic AI growth.

> Other Notable Releases: Tesla ($Tesla(TSLA.US)), Texas Instruments ($Texas Instruments(TXN.US)), GE Vernova ($GE Vernova(GEV.US)), Nokia ($Nokia Oyj(NOK.US)), STMicroelectronics ($STMicroelectronics NV(STM.US)), TE Connectivity ($TE Connectivity(TEL.US)), IBM ($IBM(IBM.US)), and ServiceNow ($ServiceNow(NOW.US)).

Sector Notes

> Semiconductors: The Philadelphia Semiconductor Index (SOX) is down 19% from its peak and is approaching oversold territory with an RSI of 40.62.

> Strategy: There is a suggestion to shift focus toward "quality names" for long-term survival, as the previous market trend of buying "anything" between April and June appears to have ended.

Check it out 👇

Morgan Stanley: Semiconductor Production Equipment

Market Dynamics by Application

> Segment Dominance: Logic/Foundry is the largest segment of the Wafer Fab Equipment (WFE) market. Within this segment, TSMC accounts for approximately 30–50% of the market.

> Industry Specialization: Among major Japanese equipment makers, companies are typically associated with specific application plays:

SCREEN: Primarily a Logic/Foundry play.

TEL (Tokyo Electron): Primarily a DRAM-related play.

KOKUSAI: Primarily a NAND-related play.

Front-End vs. Back-End

> Market Split: The semiconductor equipment market is split roughly 9:1 between front-end and back-end equipment.

> Trend Rebound: While the back-end equipment share trended downward for over 20 years as front-end tool value-add and unit prices rose, it has recently rebounded due to surging demand in the AI semiconductor market.

Competitive Landscape in Key Equipment Segments

> Lithography: The market is highly concentrated, with ASML holding a 95% share of the CY25 worldwide lithography equipment market.

> Dry Etching: The market for etchers is diverse. Lam Research leads with a 43% share, followed by Tokyo Electron (20%) and AMAT (19%).

In Conductor Etch, Lam Research is the dominant player with a 50% share.

In Dielectric Etch, Tokyo Electron holds the majority share at 55%.

Nvidia Compute Chip Cost Breakdown

> Gross Profit: The gross profit margin accounts for 75% of the total cost.

> Frontend Costs: The wafer cost represents 6% of the total.

> Backend Costs: These total 19% of the cost structure, broken down into:

HBM cost: 13%.

Test cost: 3%.

CoWoS cost: 3%

ASIC Cost Breakdown

> Gross Profit: The gross profit margin is significantly lower at 45% of the total cost.

> Frontend Costs: The wafer cost represents 13% of the total.

> Backend Costs: These total 42% of the cost structure, broken down into:

HBM cost: 24%.

Test cost: 7%.

CoWoS cost: 11%.

$NVIDIA(NVDA.US) $Alphabet(GOOGL.US) $Amazon(AMZN.US) $Applied Materials(AMAT.US) $Lam Research(LRCX.US)

Morgan Stanley: Memory

Agentic AI: Sizable Memory Demand

> Boost to DRAM Bit Demand: The implementation of Vera CPU is projected to increase total DRAM bit demand compared to 2025 levels. Specifically, a 2.5mn Vera CPU configuration is estimated to boost demand by 10%, while a 4mn configuration is estimated to boost it by 16%.

> Increased Memory Capacity: The transition from "GB NVL72" to "VR NVL72" shows a significant jump in LPDDR5X capacity requirements, rising from 17 TB to 54 TB.

DRAM Bit Shipment and ASP YoY Growth

> Decoupled Growth: The divergence between bit volume and Average Selling Price (ASP) shows that recent revenue gains are not driven by massive increases in raw shipment volume, but rather by surging prices for high-value products.

$Micron Tech(MU.US) $DRAM $EWY

Needham: WFE Spending

> Estimates Raised: Needham has increased its adjusted WFE spending estimates for the years 2026, 2027, and 2028;

2026: Adjusted WFE is now estimated at $145.3B (up from $130.1B).

2027: Adjusted WFE is now estimated at $201.0B (up from $149.6B).

2028: New CY28 WFE forecast introduced at $191.6B. Growth Drivers: The upgrades are primarily driven by accelerating DRAM and NAND spending, alongside faster-than-expected growth in leading-edge foundry/logic.

> CY28 Outlook: The forecast for 2028 reflects a "pause in growth" following two years of expansion, with memory spending—specifically DRAM—expected to moderate.

Key Segment Insights

> Foundry/Logic: Spending has strengthened, particularly at leading-edge nodes. For 2027, Needham expects leading-edge growth of approximately 30% year-over-year.

> DRAM: Described as "the true AI bottleneck," DRAM WFE for 2027 is now modeled at $84.1B, a 72% increase from the previous estimate of $49.0B.

> NAND: While pricing has been volatile, long-term bit growth driven by AI is expected to remain in the mid-20s to 30% range.

$KLA(KLAC.US) $Lam Research(LRCX.US) $Applied Materials(AMAT.US)

Mizuho Securities: Optics

Market Trends & Technologies

> Optical Circuit Switches (OCS): OCS is gaining traction among hyperscalers, with Google (GOOGL) leading adoption in its Jupiter Network and other hyperscalers like Amazon (AMZN) and Microsoft (MSFT) exploring its use. Lumentum (LITE) and Huber+Suhner (HUBN) are identified as key partners in this space.

> Co-Packaged Optics (CPO) vs. Near-Packaged Optics (NPO): While interest in CPO continues to grow (with potential deliveries in 2028/29E), industry focus is shifting toward NPO in the near-term because it is easier to deploy and requires less co-integration than CPO, despite being less power-efficient.

> Silicon Photonics (SiPho) & InP: The supply chain for SiPho remains constrained, with demand potentially outpacing foundry capacities at GFS and TSEM. A transition to 6-inch wafers is underway to help alleviate supply tightness by providing ~130% higher wafer area.

> Micro-LED (uLED) Potential: uLED/ALC is emerging as a potential disruptor for networking, offering 20-30m reach at approximately 1/5th the cost of optical pluggables, with no InP supply constraints. Credo (CRDO) is noted as a key player in this space, targeting deliveries starting in 2027-28E.

Company-Specific Updates (Lumentum & Credo)

Lumentum (LITE):

> Growth Drivers: Lumentum management noted that the "best is yet to come" as several major growth drivers—optical scale-up, optical scale-out, and optical circuit switches (OCS)—are just beginning to ramp up and are not yet fully reflected in their numbers.

> 1.6T Transition: The company is seeing a significant tailwind from the transition to 1.6T transceivers, led by NVIDIA and Google, with 200-gig EMLs providing roughly double the average selling price (ASP) of previous generations.

> Scale-Out Business: The partnership with NVIDIA for high-powered lasers in CPO switch platforms is a major focus, with expected incremental revenue of $50 million to $100 million in the fourth quarter of fiscal 2026 alone.

Credo Technology (CRDO):

> Market Outperformance: Credo has significantly outperformed larger rivals like NVIDIA and Broadcom in share price gains during 2026, largely due to its specialized role in high-speed, energy-efficient connectivity.

> Optical Portfolio: Following its acquisition of DustPhotonics, Credo has strengthened its silicon photonics capabilities. The company expects its optical connectivity business to generate over $600 million in revenue in fiscal 2027, supported by a 1.6T networking transition.

$Lumentum(LITE.US) $Credo Tech(CRDO.US) $Broadcom(AVGO.US) $NVIDIA(NVDA.US) $STMicroelectronics NV(STM.US) $Alphabet(GOOGL.US) $Meta Platforms(META.US) $Tower Semicon(TSEM.US) $Marvell Tech(MRVL.US) $Taiwan Semiconductor(TSM.US) $Microsoft(MSFT.US) $Amazon(AMZN.US)

Mizuho Securities: CPUs & GPUs

Market Forecasts & Growth

> Shipment Growth: Industry server CPU shipments are forecasted to reach 35 million units in 2026 and grow to 50 million units by 2027, representing a 40% year-over-year increase.

> Long-Term TAM: The long-term Total Addressable Market (TAM) estimate for 2030 has been raised to $170 billion (up from the previous $107 billion forecast), driven by higher CPU-to-GPU ratio assumptions for AI inference servers.

> CPU-to-GPU Ratios: The CPU-to-GPU ratio on AI servers is accelerating and is expected to approach 1:1 by the end of 2027 or 2028.

Supply Chain & Technical Bottlenecks

> DRAM Constraints: A critical bottleneck exists in DDR5/LPDDR5 supply, with a projected fulfillment ratio of only 70% over the next 12–18 months.

> Demand vs. Supply Gap: Based on current models, the 2027 demand for DDR5/LPDDR5X (over 300 billion 1Gb equivalents) significantly exceeds the projected supply (220–250 billion 1Gb equivalents).

> Potential Risks: The shortage of key materials—DRAM, substrates, and passives—is expected to persist through 2027 and could pose downside risks to downstream server assemblers, potentially leading to lower server rack output.

Key Player Insights (2027 Forecasts)

> Nvidia: Expected to reach 5.0–6.0 million units for the Vera CPU, including 2.0–3.0 million units specifically for agentic AI stack racks.

> Google: Axion CPU production is projected to increase more than 2x year-over-year, aligning with the growth trajectory of TPU units.

> AMD: The N2 Venice CPU is forecasted to exceed 6.0 million units.

GPUs/ASICs

Market Growth Projections

> Rapid Expansion: The total AI ASIC market is projected to grow from 4.1 million units in 2025 to 24.0 million units by 2028.

> Volume Drivers: The growth is driven by substantial increases in deployment by major hyperscalers including Google, Amazon (Annapurna), Meta, Microsoft, and OpenAI.

> External Demand: The market for external (non-Google) AI ASIC units is expected to surge from 0.6 million in 2025 to 7.2 million by 2028.

Key Hyperscaler Activity

> Google (TPU): Continues to be a dominant player, with total shipment units increasing from 2.5 million in 2025 to 7.1 million by 2028.

> Anthropic: Significant ramp-up is forecasted for their "TPU Ironwood/Sunfish" chips, moving from 0.6 million units in 2026 to 6.2 million units by 2028.

> Amazon/Annapurna: Shipments for the Trainium line are projected to double from 1.5 million in 2025 to 3.6 million by 2028.

> Meta: Rapid scaling of MTIA chips is expected, growing from 0.1 million units in 2025 to 2.7 million units by 2028.

Technical Trends

> Advanced Packaging & Nodes: There is a heavy reliance on sophisticated packaging technologies like CoWoS-L and CoWoS-S, and advanced foundry nodes including N2, N3, N4, N5, and A16.

> HBM Integration: Nearly all listed high-performance ASICs utilize High Bandwidth Memory (HBM), with a transition toward newer generations such as HBM3E and HBM4/4E to meet performance demands.

> ASP Variance: Average Selling Prices (ASP) range significantly, from approximately $2,000 for entry-level models to as high as $40,000 for top-tier specialized chips like the TPUv10.

$DRAM $EWY $Micron Tech(MU.US) $Alphabet(GOOGL.US) $Amkor Tech(AMKR.US) $Taiwan Semiconductor(TSM.US) $ASE $NVIDIA(NVDA.US) $AMD(AMD.US) $Broadcom(AVGO.US) $Marvell Tech(MRVL.US) $Intel(INTC.US) $Microsoft(MSFT.US) $Meta Platforms(META.US)

Deutsche Bank: Humanoids

Market Growth Forecast

> Rapid Expansion: The global humanoid robot market is projected to see significant growth, moving from approximately 3,030 units in 2024 to an estimated 700,500 units by 2030.

> Long-Term Outlook: Looking further ahead, the forecast anticipates the market scaling to 70 million units by 2050.

> Market Value: The market value is also expected to climb sharply, reaching $1,050,000 million (or $1.05 trillion) by 2050.

Regional Dominance

> China's Lead: China is currently the primary driver of the market, accounting for 85% of global shipments in 2024. While this percentage is forecast to decrease as other markets mature, China is still expected to remain a dominant player, holding 43% of the market share by 2050.

> United States: The U.S. share is projected to grow from 12% in 2024 to 43% by 2050, effectively matching China's share by the end of the forecast period.

> Other Regions: Europe, Japan, and other regions are expected to grow from a small 3% share in 2024 to 14% of the global market by 2050.

Key Industry Players

> United States: Major players include Tesla, Figure, Agility, Boston Dynamics, and 1X. Tesla is forecasted to be a significant contributor, growing from 100 units in 2024 to 150,000 units by 2030.

> China: Prominent companies include Unitree, AGIBOT, UBTECH, Leju Robotics, Booster Robotics, and Galbot. Unitree is projected to reach 100,000 units by 2030.

Citi: US Communication Equipment

Market Overview: InP Supply Constraints

> Persistent Supply Gap: Despite rapid capacity increases, InP supply is expected to remain below demand for the next three years.

> Strategic Importance: InP is a critical bottleneck in the AI infrastructure buildout, as it is the semiconductor substrate used for high-speed laser diodes in 800G and 1.6T optical transceivers.

> Supply-Demand Model: Citi projects a significant demand gap for EML lasers, estimated at 30% in 2026, narrowing to 20% in 2027 and 11% in 2028.

> 6-Inch Wafer Transition: The transition to 6-inch wafers is a major structural lever, as one 6-inch wafer produces 4.36x more chips than a 3-inch wafer.

Lumentum (LITE) Highlights

> Pricing Power: LITE is identified as the leader in 100G/200G EML and UHP lasers, positioning it to benefit from favorable pricing dynamics.

> Earnings Catalyst: Citi has opened a 90-day upside Catalyst Watch on LITE ahead of its Jun-Q earnings report and ECOC in September 2026.

> CPO Roadmap: Management maintains that Lumentum's Co-Packaged Optics (CPO) roadmap is on track, with laser shipments expected to inflect in Q4 2026 and grow further in 2027.

> Financial Revisions: Citi upwardly revised its FY2027E and FY2028E EPS estimates for LITE by 6% and 5%, respectively, reflecting an anticipated 15% price increase in FY2027E.

Coherent (COHR) Highlights

> Vertical Integration: Coherent is pursuing a strategy of full vertical integration, including growing its own InP substrates and fabricating EML/DFB laser chips in-house.

> 6-Inch Platform: COHR has begun volume production on a 6-inch InP wafer platform and expects to double its InP capacity by the end of calendar 2026.

> Constructive Outlook: Citi remains constructive on COHR, viewing recent stock pullbacks as an attractive buying opportunity given the company's role in the AI optics supply chain.

Competitive Landscape

> Chinese Competition: While domestic Chinese players are scaling up capacity, they are still trailing in qualification cycles and volume/yield stability compared to leaders like Lumentum.

> Export Controls: China's February 2025 decision to add InP to its export-control list caused a supply shock and remains a "tail risk" for the industry.

$AXT(AXTI.US) $Coherent Corp.(COHR.US) $Lumentum(LITE.US)

Morgan Stanley: Humanoid Semis Demand

The Strategic Importance of Semiconductors in Robotics

> Foundational Role: Semiconductors are essential to physical AI. As robots and machines become more intelligent, their semiconductor content increases.

> Market Expansion: Morgan Stanley estimates that the semiconductor market for robotics could expand approximately 800x through 2050.

> Market Projections (Original Equipment Only):

2025: ~$2.2 billion

2030: $24 billion

2040: $570 billion

2050: $1.8 trillion

Types of Required Semiconductors

> Compute & Perception: High-performance processors (CPUs, GPUs) for motion planning, decision-making, and perception, alongside memory and connectivity chips for real-time data processing.

> Specialized Components: Vision processors, sensor interfaces (for LiDAR, cameras, radar, force, and tactile sensors), motor control ICs, and power management/analog chips.

Industry Trends & Developments

> New Design Cycles: The proliferation of robotics is expected to drive demand for a new wave of purpose-built, robotics-specific silicon.

> Major Player Activity: Several semiconductor companies, including Qualcomm, NVIDIA, Infineon, and NXP, have recently unveiled products or frameworks specifically designed for robotics and physical AI.

> NVIDIA's "3 Computer" Approach: NVIDIA views robotics as requiring three distinct computing environments: datacenter (training), simulation, and on-robot inference compute.

> Humanoid Content: Estimates suggest ~$1.5k–$3k in semiconductor content (compute, MCU, sensing) per humanoid robot.

$Microchip Tech(MCHP.US) $Sony(SONY.US) $AMD(AMD.US) $NVIDIA(NVDA.US) $Qualcomm(QCOM.US) $ON Semiconductor(ON.US) $Ambarella(AMBA.US) $STMicroelectronics NV(STM.US) $Allegro Microsystems(ALGM.US) $Synopsys(SNPS.US) $Cadence Design(CDNS.US) $Texas Instruments(TXN.US) $Arm(ARM.US)

Citi: EML Output

> Industry Shift Toward 200G and CW: Across almost all listed companies, there is a marked transition away from 100G EML production by 2028, with a significant ramp-up in both 200G EML and CW output.

> AXT, Inc. Growth: AXT, Inc. shows a substantial increase in its 200G EML and CW output, with its 200G EML output projected to grow from 13.54 million chips in 2025 to 157.82 million chips by 2028.

> Rapid Expansion in China: Companies such as Source Photonics (Dongshan Precision) and the "Other China" category show dramatic growth in CW output. Source Photonics, for example, is projected to increase its CW output from 14.66 million chips in 2025 to 161.77 million chips by 2028.

> Sumitomo Electric: Sumitomo maintains high production levels, particularly in the 200G EML segment, which jumps from 28.93 million in 2025 to 128.58 million in 2028.

(h/t: @yulin98434)

Goldman Sachs: Innolight & Optical Modules

Market Trends and Technology Migration

> Speed Evolution: The industry is aggressively migrating toward higher-speed optical modules. 1.6T modules are moving toward volume production as the next mainstream standard, while 3.2T optical modules are expected to begin volume ramp-up in 2027.

> Silicon Photonics (SiPh) Dominance: Silicon photonics is increasingly displacing traditional EML-based solutions due to lower power consumption, superior performance, and simplified packaging.

> Penetration Rates: By 2026, SiPh penetration is projected to reach 60% for 800G, 80% for 1.6T, and 100% for 3.2T products.

> Revenue Share: Silicon photonics' share of total optical module revenue is expected to surge from 28% in 2025 to 62% by 2028.

Innolight Shipment and Market Share Highlights

> Leadership: Innolight is the leading supplier of 1.6T pluggable modules. It is projected to hold a 50–60% market share of 1.6T modules in 2026. Other industry reports indicate Innolight's market share for 1.6T modules may reach as high as 50–70%.

1.6T and 3.2T Outlook:

> 1.6T: Expected to become the "absolute mainstream" by 2027, with 45.72 million units projected to be shipped that year.

> 3.2T: Following the ramp-up of 1.6T, 3.2T shipments are expected to double in 2028 compared to their initial volume ramp year in 2027.

> Manufacturing & Capacity: To support this demand, Innolight planned a 155% year-on-year increase in capital expenditure for 2026 to expand production capacity.

Market Dynamics: Volume, Revenue, and ASP

> Speed Migration (Volume Mix): The market is shifting rapidly from legacy "Below 400G" speeds toward high-speed AI-focused modules.

While "Below 400G" accounted for 89% of shipment volume in 2025, that share is projected to decline to 77% by 2028E.

Conversely, high-speed modules are capturing significant volume: 1.6T modules are projected to grow from 1% of volume in 2025 to 10% by 2028E, and 3.2T modules are expected to reach 5% of total shipment volume by 2028E.

> Average Selling Price (ASP) Trends: The report explicitly notes an assumed year-over-year (YoY) ASP decline for "like-for-like" products.

For example, the ASP for 1.6T modules is expected to drop from $840 in 2025 to $435 by 2028E.

The 3.2T module, which enters the market in 2027, has a high starting ASP of $1,023 in 2027E, which is then projected to decline to $768 by 2028E.

Technological Shift: SiPh vs. EML

> Silicon Photonics (SiPh) Adoption: The market is undergoing a significant transition from traditional EML-based materials to Silicon Photonics.

SiPh revenue penetration is expected to increase from 28% in 2025 to 62% by 2028E.

Correspondingly, the SiPh penetration rate (by volume) is projected to climb from 6% in 2025 to 18% by 2028E.

Market Opportunity (TAM)

> Total Addressable Market (TAM): The global optical module TAM is expected to grow from $34.2 billion in 2025 to a peak of $72.6 billion in 2027E, before moderating slightly to $69.1 billion in 2028E.

> High-Speed Revenue Growth: The 1.6T and 3.2T segments represent the primary growth engines, with combined revenue for these categories projected to grow from $3.9 billion in 2025 to $44.9 billion in 2028E.

Morgan Stanley: Optical vs Copper

How do you think about the optical scale up opportunity in 2027, 2028, 2029, etc, and how does that relates to the Nvidia ecosystem and roadmap?

A: We believe a small scale out adoption next year, with small volumes accounting for 10-15% penetration with scale out. We do think there are some low adoption with CPO until Feynman in 2029. Small, single digit adoption of CPO in 2028 and really scaling in 2029.

You took your TAM to $73B, 4x from a year ago, talk about the drivers of that and how that compares to consensus and buy side out there?

A: Driven by CapEx data points being 2x we were expecting last year, that is one vector of growth. The other is the cluster sizes being discussed are the biggest rivers of scale up network . We have gone from 8 to 72, then from a couple 100s to 1000s in the next couple generations. This acts as a multiplier to scale up multiplier.

Talk about the exposure among Lumentum, Coherent, and Corning. Are there nuances between who is more exposed, less exposed?

A: For CPO within scale out, it is generally positive for Lumentum. Doesn't necessarily changed a whole lot for Corning. With Coherent, they have to balance it out with transceiver business they have. For Lumentum it is the most positive because of the one customer they have on the transceiver side being google and they are not going to be adopting CPO. For CPO within scale up, it is opportunity for all 3. This is where copper gets replaced with fiber and lit up by lasers and optical engines made by Coherent an Lumentum. In terms of overall exposure, Lumentum has a bigger catalyst because of larger AI/DC revenue. It would be incremental for both Coherent and lumentum. Lumentum #1, Coherent or Corning #2.

Tell me about the Corning Glass Bridge announcement that got the stock excited?

A: This is not a new announcement. It was a paper they released in 2025 and a presentation they did that refocused the market on this technology. It is already part of the $10B photonics target the company discussed couple of months ago.

In terms of catalyst for optical space, OCP (Open Compute Project) coming in October has gone popular. What are you expecting to hear from this?

A: We think this will be a catalyst because it provides more exposure into hyperscaler behavior. OCP will give us a chance to see hyperscalers roadmap, it will focus on CPO, NPO adoption and the vendors involved.

Copper is still the story for 2026 and 2027. We have seen this continued enthusiasm for ASICs. How is copper well positioned for this dynamic? Who are the well positioned beneficiaries?

A: Copper life can be extended with use of retimers and better SerDes. Copper has a lot of runway before switching to optics. We will likely see a hybrid environment before going to 100% optics. A large debate in scale up is what fabric and protocol will become dominant other than NVLINK as we get into new XPUs and alternatives to Nvidia GPUs ramping. A big debate here is between UALink and ethernet. Ethernet is dependent on Broadcom's new ASICs ramping in 2H2027 into 2028, we think Broadcom will have opportunity to bundle XPU solutions with networking portfolio and are very well positioned in ethernet. On UALink, this is high optionality for Marvell and Astera Labs. Astera has the scale up switch for Trainium 3, this is a PCIe based solution. We think customers will want to move to UA Link eventually.

You consider Keysight as an agnostic winner in R&D, regardless of the copper vs optical timeline. Explain your thesis and why do you see it pulling ahead?

A: Keysight really ways in multiple ways when it comes to AI. First, it benefits from different types of architectures being discussed which all need to be tested with their equipment in labs. Second, when you move to higher speeds (400g laser vs 200g), it requires more complicated testing. Third, when architectures change frequently, like every 2 years, that boosts their business. The variety of architectures being tested matters.

$Astera Labs(ALAB.US) $NVIDIA(NVDA.US) $Broadcom(AVGO.US) $Marvell Tech(MRVL.US) $Credo Tech(CRDO.US) $Semtech(SMTC.US) $Amphenol(APH.US) $Keysight Tech(KEYS.US) $Coherent Corp.(COHR.US) $Lumentum(LITE.US) $Corning(GLW.US) $Alphabet(GOOGL.US)

Next week will be huge...many big names reporting.

Texas Instruments $Texas Instruments(TXN.US)

Intel $Intel(INTC.US)

Tesla $Tesla(TSLA.US)

Google $Alphabet(GOOGL.US)

GE Vernova $GE Vernova(GEV.US)

Nokia $Nokia Oyj(NOK.US)

STMicroelectronics $STMicroelectronics NV(STM.US)

TE Connectivity $TE Connectivity(TEL.US)

IBM $IBM(IBM.US)

ServiceNow $ServiceNow(NOW.US)

BofA: Kimi 3

The Kimi K3 Release & The Compute Race

> New Chinese Open-Weight Model: Moonshot has unveiled Kimi K3, a massive 2.8-trillion-parameter Mixture of Experts (MoE) model with a 1-million-token context window.

> U.S. Labs Pressured to Scale: With Chinese open-weight models closing the gap and media reports suggesting Google’s Gemini 3.5 Pro is months behind schedule, U.S. frontier labs (OpenAI, Anthropic, Google) must increase compute. They will need larger training runs, heavier reinforcement learning (RL), synthetic-data loops, and faster release cadences to stay ahead.

> Business Over Leaderboards: Investors are cautioned not to confuse benchmark leadership with sustainable business models. The durable moat in enterprise AI is delivering accurate, low-latency, and high-uptime AI at the lowest cost.

Bullish Outlook for AI Semiconductors

> MoE Architectures Boost Hardware Demand: The shift toward Mixture of Experts (MoE) architectures highlights the critical importance of memory movement, routing, latency, and interconnects.

> NVIDIA's Next-Gen Efficiency: NVIDIA's GB300 NVL72 provides up to a 25x performance-per-watt improvement over Hopper when serving leading open MoE models.

> Expanding Silicon Demand: Open models are inherently bullish for semiconductors. Even as model value commoditizes, the infrastructure demands for GPUs, High-Bandwidth Memory (HBM), networking, and efficient inference will continue to expand.

> EDA Resilience: BofA remains constructive on Electronic Design Automation (EDA) players like Cadence (CDNS) and Synopsys (SNPS). Despite mentions of "open-source EDA at 45nm" in Kimi K3, commercial EDA tools remain entirely mandatory for major foundries like TSMC to manufacture advanced chips.

Surging Token Usage & Enterprise AI Adoption

> Chinese Labs Leading Token Volume: Data from OpenRouter shows that weekly token usage is proliferating rapidly, with daily token usage on Chinese AI models now exceeding that of Western AI labs.

> US Enterprise Adoption Rates: According to the Ramp AI Index, approximately 55% of US businesses now have paid subscriptions to AI tools (significantly higher than the US Census estimate of 21%).

> Market Share & Spending: Anthropic leads enterprise model adoption at 42.4%, closely followed by OpenAI at 39.5%. While the median monthly AI spend per employee is just $11, the top 1% of enterprise spenders are averaging a massive $4,833 per employee monthly.

$Cadence Design(CDNS.US) $NVIDIA(NVDA.US) $Alphabet(GOOGL.US) $Synopsys(SNPS.US)

BofA: Google $Alphabet(GOOGL.US) Earnings Preview

2Q'26 Financial Estimates vs. Consensus

> Net Revenue: Estimated at $102.1bn (up 25% y/y) vs. Street consensus of $101.0bn.

> GAAP EPS: Estimated at $8.38 vs. Street consensus of $2.90.

> Other Income Benefit: BofA's significantly higher EPS estimate includes a $80bn one-time mark-to-market benefit from the revaluation of Alphabet’s stake in Anthropic (Anthropic's valuation reportedly surged from $380bn in 1Q to $965bn in 2Q)

Segment Performance & Model Revisions

> Google Cloud Strength: BofA raised its 2Q Google Cloud growth estimate to 70% y/y ($23.2bn in revenue) driven by accelerating AI demand and a strong backlog. Anthropic has reportedly committed to spending $200bn on Google Cloud over the next 5 years.

> Google Search: Expected to grow 17% y/y to $63.6bn. Strong retail search growth is being slightly offset by softness in travel and consumer packaged goods (CPG), alongside minor FX headwinds.

> YouTube Advertising: Estimated at $10.8bn (up 10% y/y), which is inline with the Street. Growth is decelerating modestly due to tougher comparisons and automated tools (like PMax) shifting client budgets to Search.

> Long-term Revisions (Full Year 2026): Net revenue estimates were raised by 1% to $427bn and full-year EPS estimates increased by 36% to $19.70.

Capex Expansion & Infrastructure Leases

> Capital Expenditures: BofA notes that higher component pricing (e.g., memory/DRAM) and accelerating AI demand could push Google to increase its CY26 capex range by ~5% to $190–$200bn (BofA models $196bn). 2Q capex is modeled at $50.1bn (up 123% y/y).

> SpaceX GPU Lease: On June 5, 2026, Alphabet entered a multi-year agreement to lease 110,000 NVIDIA GPUs from SpaceX, committing to a payment of ~$920mn per month from October 2026 through June 2029 to secure immediate capacity.

> Massive Capital Raise: Alphabet raised $85bn in capital ($44.75bn in equity capital/strategic investment and a planned $40bn ATM equity program) alongside issuing $17bn in Euro and Canadian dollar debt to fund AI infrastructure and tax obligations.

Morgan Stanley: ZAM Memory

> Massive Cloud Capex Exposure: Cloud memory spending is projected to reach US$418 billion by 2030, representing an 8% CAGR from 2026. Memory spending could account for 40% of total cloud capex in 2027 (up from just 12% in 2023, prior to the AI boom).

> Bandwidth Lags Behind Processing: Memory bandwidth improvements (e.g., a +14% increase from DDR5-5600 in 2024 to DDR5-6400 in 2026) are failing to keep pace with explosive token growth, which expanded over 320x between April 2024 and June 2026 on major cloud platforms.

> Absolute Spend Peaking: Absolute cloud memory spending is forecasted to hit its cycle high in 2028e at roughly US$520 billion.

> The Post-2028 Softening: From 2028e to 2030e, even though overall cloud capex (excluding memory) is projected to rise to its highest level of $1.28\text{ trillion}$ by 2030, memory spending falls as a percentage of the total budget back down to 25% by 2030e.

DRAM Supply/Demand (S/D) Imbalance

> Historical Context: The severe supply shortage of 2023 is clearly visible, bottoming out in 4Q23 near -28%.

> The Current 2026/2027 Tightness: Following a brief macro recovery, the market enters a sustained period of deep supply deficit. Through all of 2026e, the supply deficit hovers continuously between -10% and -16%.

> Supply Shortage to Persist: Looking ahead, the green bars indicate that the market remains structurally undersupplied, keeping DRAM conditions historically tight all the way through 4Q28e.

NAND Supply/Demand Dynamics

> The COVID/Post-COVID Glut: The massive NAND oversupply of late 2022 is prominent, peaking in 4Q22 with a +17% surplus.

> Structural Repricing Phase: After hitting a deep shortage in 1Q26e (dipping near -9%), the NAND deficit begins to gradually moderate, though it remains marginally in negative territory (undersupplied) near -3% to -4% through late 2026e.

> Transition to Enterprise Storage: This sustained tightness supports the report's thesis that NAND is moving from cheap consumer sockets up the memory hierarchy to act as massive LLM KV caches.

$DRAM $Micron Tech(MU.US) $Sandisk(SNDK.US) $Western Digital(WDC.US) $EWY

Semiconductor stocks this week and month with a beta > 1.0

Worst Performing (1-Week)

$AXT(AXTI.US): -27.15%

$Astera Labs(ALAB.US): -23.41%

$Marvell Tech(MRVL.US): -22.60%

$Maxlinear(MXL.US): -22.34%

Worst Performing (1-Month)

$AXT(AXTI.US): -58.85%

$Navitas Semiconductor(NVTS.US): -50.46%

$Vishay Intertech(VSH.US): -42.81%

$Atomera(ATOM.US): -40.11%

BofA: 3Q Memory

Key Takeaways from the 3Q Contract Price Check

> Strong Server DRAM Pricing: Up 20–30% QoQ, spearheaded by high-speed LPDDR5 (compared to the market consensus of 20% or less).

> Upbeat Spot Demand: Spot demand is healthy, driven by commodity DDR5 and legacy DDR4, with July prices increasing Month-on-Month (MoM).

> Transition to HBM4: There is a near-term rise in orders shifting toward higher-priced HBM4 over the cheaper HBM3e.

> Lower Long-Term Agreements (LTAs): LTA-based sales represent well below 50% of total DRAM sales (non-LTA portion is at 60–70%), allowing for solid QoQ price increases (e.g., 5–10%) that are often settled even under existing LTAs.

> Rush Orders: Sudden OEM rush orders confirm a QoQ price increase of over 20% for both commodity DRAM and NAND.

> Optimistic ASP Outlook: BofA forecasts a 21% QoQ rise in 3Q DRAM ASP, which is significantly more optimistic than TrendForce's assumptions (13–18% QoQ for conventional DRAM, or only 8–13% including HBM).

Spot Market Dynamics

> Price Rallies: DRAM spot prices have risen for eight consecutive weeks, defying earlier expectations of a price cap at US$35–40 for 16Gb DDR5.

> OEM Buying Behavior: Having previously cut device production to resist the spot-price rally, OEMs are now actively purchasing memory chips to prepare for high sales in September and the peak 4Q season.

> NAND Recovery: NAND spot prices rebounded, driven by a 4% Week-on-Week (WoW) rise in 1Tb wafers, supporting expectations of a 10%+ NAND ASP rise in 3Q.

$DRAM $Micron Tech(MU.US) $Sandisk(SNDK.US) $Western Digital(WDC.US) $EWY

SEMI: Global Semiconductor Equipment Sales Forecast

📈 Record-Breaking Growth & Core Drivers

> Historically High Sales: Global sales of semiconductor manufacturing equipment are forecast to hit a record $165.9 billion in 2026 (up 23.2% year-on-year).

> Five-Year Expansion Surge: The upward momentum is projected to run through 2028, reaching an unprecedented $229.5 billion.

> The AI Catalyst: Accelerating demand for AI infrastructure is driving this historic cycle—specifically forcing massive capital investments in leading-edge logic, High Bandwidth Memory (HBM), advanced packaging, and highly complex device architectures.

🔬 Key Segment Breakdowns

Wafer Fab Equipment (WFE)

> 2026 Projection: Projected to grow 23.1% to $143.9 billion. This represents a sharp upward revision from previous forecasts.

> Long-Term Trajectory: WFE spending is expected to expand 21.8% in 2027 and 14.1% in 2028, surpassing the $200 billion milestone.

Back-End Equipment (Test & Packaging)

> Test Equipment: Surging 31.0% to $15.3 billion in 2026 (following a massive 55.3% spike in 2025). It is expected to hit $20.8 billion by 2028.

> Assembly & Packaging: Projected to grow 9.6% to $6.7 billion in 2026, climbing to $8.6 billion by 2028. This is driven by advanced and heterogeneous packaging needs for complex AI devices.

🌍 Geographic Spending Trends

> China, Taiwan, and Korea will maintain their dominance as the top three destinations for equipment spending through 2028.

> China will hold the #1 spot, though its rapid growth rate is expected to moderate in 2026 following years of heavy, record-setting investments.

> Taiwan & Korea are seeing massive spending specifically targeted at leading-edge foundry nodes and advanced HBM memory capacity.

> Other regions are projected to experience accelerated spending in 2027 and 2028, fueled by supply chain regionalization, government incentives, and specialty capacity expansions.

BofA: ASML $ASML(ASML.US)

Major Estimate Upgrades (CY26–28E)

Following a strong Q2 earnings print, BofA materially upgraded its financial estimates:

> Revenue Estimates: Raised by 13% to 15% across CY26–28E. This is driven by higher Argon Fluoride Immersion (ArFi) units/ASPs, higher Extreme Ultraviolet (EUV) ASPs, and higher Installed Base Management revenues.

> Gross Margin (GM): Upgraded by 2.2 to 3.3 percentage points. This structural improvement is backed by a richer mix of ArFi and EUV tools, higher software revenues, and increased upgrade contributions.

> Earnings Per Share (EPS): Upgraded significantly across the forecast horizon:

2026E EPS: Raised to €41.49 (previously €33.17, up 25%).

2027E EPS: Raised to €58.16 (previously €46.75, up 24%).

2028E EPS: Raised to €75.43 (previously €62.31, up 21%).

> Memory Dominance: ASML’s guided CY26E memory revenue growth of 75% is expected to far outpace US peers, driven by the DRAM industry's structural transition to EUV.

> Rising Tool Prices (ASPs): Low-NA EUV tool average selling prices (ASPs) are projected to climb to €245m in 2H26E (compared to €220–230m in H1), indicating stronger software attach rates.

> Constrained OPEX: Operating expenses are expected to grow materially slower than group revenues due to operational efficiencies.

> Confidence Beyond 2027: While the market is highly focused on ASML's 85 EUV unit target for 2027, management is confident in potentially scaling capacity to 110 EUV units beyond 2027.