Equity research

Jensen Huang: "no HBM for you, I need HBM!"

Overall Market & Demand Growth

> Massive Expansion: Total HBM demand is projected to scale dramatically, reaching over 6 million k GB (approx. 48,618 mn Gb) by 2027.

> Rapid Multi-Year Trajectory: Historical and projected growth shows a steep upward curve from 2023 through 2027, driven heavily by increasing deployment in AI GPUs and AI ASICs.

NVIDIA's Market Dominance

> Leading Consumer: NVIDIA remains the single largest consumer of HBM supply, commanding the largest share of the demand stack in 2027.

> Key Products: High-demand products like the Rubin R200 stand out with massive total HBM demand (projected at over $1.7 million k GB), utilizing advanced HBM4 12hi configurations.

> CoWoS Allocation: NVIDIA dominates advanced packaging allocations, notably with massive wafer distributions for products like the Vera CPU and Rubin series.

Key Competitors and Ecosystem Players

> Google: Google represents a massive portion of the AI ASIC HBM demand, particularly driven by TPU lines such as the TPU v8i (Sunfish), which demands over 1.1 million k GB of HBM.

> AMD: Competes strongly with advanced offerings like the MI455 and MI350 series, utilizing HBM3e and HBM4 configurations sourced primarily through Samsung and Micron.

> Cloud Service Providers (ASICs): Other major tech giants like AWS (Trainium 3) and Microsoft (Maia 300) account for significant shares of the total specialized hardware demand footprint.

> NVIDIA holds the largest single share of projected HBM consumption at 36%.

> Google represents a massive segment of demand at 35%, nearly matching NVIDIA due to its heavy reliance on custom TPU deployments.

> AMD follows as the third-largest consumer with 16% of the market share.

> AWS accounts for 5%, GUC's customers make up 3%, while Microsoft and Meta each take 2% of the projected 2027 consumption.

$NVIDIA(NVDA.US) $Meta Platforms(META.US) $Microsoft(MSFT.US) $AMD(AMD.US) $Alphabet(GOOGL.US) $Micron Tech(MU.US) $DRAM $EWY

Nvidia's announcement of co-packaged optics entering mass production provides a boost to optical communications.

"Gilad Shainer, Senior Vice President of NVIDIA, announced that Co-Package Optics (CPO) has entered mass production."

"According to foreign media reports, Gilad Shainer announced at a recent technology forum that CPO has begun mass production. Switches developed by Nvidia in collaboration with its supply chain have been delivered to close customers and are also being deployed at Nvidia's own facilities. It is expected that a large number of switches with CPO technology will be introduced into AI factories around the world this year."

CMB International: CPO

> CPO Market Growth: Co-packaged optics (CPO) are projected to experience rapid expansion with a 219% revenue CAGR over 2025–30E, including 300%/350% YoY growth in 2026E/27E, driven primarily by early adopters.

> Pluggable Optical Modules: Traditional pluggables will continue steady volume and technology growth, posting a 30% CAGR over 2025–30E alongside 58%/33% YoY growth in 2026E/27E.

> Market Dynamics & Roles: CPO is not an immediate replacement for pluggables. Pluggables remain the near-term volume engine, while CPO/NPO act as faster-growing extensions addressing escalating AI network power, density, and copper-reach constraints.

$Lumentum(LITE.US) $Coherent Corp.(COHR.US) $Applied Optoelectronics(AAOI.US) $Marvell Tech(MRVL.US) $Broadcom(AVGO.US)

GF Securities is forecasting global WFE to reach $236 billion in 2027 and then $295 billion in 2028.

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

BofA: Hyperscalers

Massive Capital Expenditure (Capex) Surge

> Record Growth in 2025–2026: Combined capex for the top US tech companies (Amazon, Microsoft, Alphabet, Meta, and Oracle) is projected to grow ~65% YoY in 2025 to reach a massive scale, followed by an additional ~100% YoY jump in 2026 to hit $730 billion.

> Hitting the $1 Trillion Milestone: Combined annual capex is expected to surpass $1 trillion per annum by 2027 and 2028.

> Individual Hyperscaler Breakdown (2026E): Oracle and Alphabet lead the YoY percentage growth chart for 2026E with triple-digit increases (around 120% and 118% respectively), closely followed by Microsoft and Meta.

Robust Cloud Revenue Growth

> Consistent Expansion: Cloud revenue across major providers (AWS, Microsoft Azure, and Google Cloud) remains structurally strong, maintaining a solid 35–45% YoY growth trajectory through 2026–2028.

> Revenue Trajectory: AWS continues to hold the top spot in total revenue terms heading into 2028, with Microsoft Azure and Google Cloud tracking closely behind on steep upward curves driven by surging AI infrastructure demand.

Temporary Free Cash Flow (FCF) Pressure

> Short-Term Dip: Driven by unprecedented AI infrastructure investments, several hyperscalers are expected to experience temporarily negative Free Cash Flow (FCF) through 2026 and 2027 as spending outpaces near-term cash flow growth.

> Quarterly Volatility: Quarterly trends highlight notable dips into negative FCF territory for companies like Amazon and Alphabet during peak spending quarters before stabilizing.

$Oracle(ORCL.US) $Microsoft(MSFT.US) $Meta Platforms(META.US) $Alphabet(GOOGL.US) $Amazon(AMZN.US)

BofA: Memory

DRAM Pricing Trends

> Record Highs: DRAM spot and contract prices climbed to record highs of approximately US$35–40 following a strong rally in 4Q25 and 1Q26.

> Growth Trajectory: After the sharp surge, prices are exhibiting moderate growth moving into 2Q26.

> Spot vs. Contract: Both 16Gb DDR5 spot and contract prices remained relatively flat between 1Q23 and 3Q25 before experiencing a steep vertical breakout starting in late 2025.

NAND Wafer Pricing Trends

> Record-High Levels: Both 512Gb wafer spot and contract prices are currently at record-high levels.

> Contract vs. Spot Dynamics: NAND wafer contract prices have surged significantly ahead of spot prices during the recent upswing.

> Historical Baseline: Prices hovered stably between US$2 and US$5 from early 2023 through late 2025 before skyrocketing sharply into mid-2026.

DDR4 vs. DDR5 Contract Price Comparison

> Disappearance of Price Premium: DDR4 and 16Gb DDR5 contract prices have converged at similar levels (US$35–50 by mid-2026); the traditional DDR5 price premium no longer exists.

> Driver of Convergence: The collapse of the price gap is driven primarily by an acute DDR4 shortage.Timeline: Both technologies tracked closely below US$10 until mid-2025, after which they escalated rapidly through July 2026.

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

Lumentum leads in market share of laser chips

$Lumentum(LITE.US) $Broadcom(AVGO.US) $Coherent Corp.(COHR.US) $Macom Tech(MTSI.US)

Kioxia $258A.T earnings:

Consolidated Financial Performance (Q1 FY2027)

Driven by robust demand centered on generative AI applications and strong average selling prices, the company achieved significant year-over-year growth:

> Net Sales: 1,767,117 million yen (an increase of 415.5% or +1,424.3 billion yen compared to Q1 FY2026).

> Operating Profit: 1,270,017 million yen (compared to 44,899 million yen in Q1 FY2026).

> Non-GAAP Operating Profit: 1,326,216 million yen. Profit Before Tax: 1,234,720 million yen.

> Quarterly Profit Attributable to Owners of the Parent: 842,165 million yen (a massive surge from 18,284 million yen in the same period last year).

Sales Breakdown by Application

> SSD & Storage: 1,174,719 million yen (up from 217,411 million yen in Q1 FY2026).

> Smart Devices: 525,656 million yen (up from 79,040 million yen in Q1 FY2026).

> Other: 66,742 million yen (up from 46,348 million yen in Q1 FY2026).

Stock Split Resolution: The Board of Directors resolved on July 31, 2026, to execute a 3-for-1 stock split of common stock, with an effective date of October 1, 2026.

Q2 FY2027 Outlook: Anticipating continued strong demand from data centers, Kioxia projects further sequential growth for Q2 (July 1 to September 30, 2026), forecasting net sales of 2,390,000 million yen and an operating profit of 1,890,000 million yen.

Situational Awareness is down around 67% so far in July. Including July's losses, the fund remains up about 80% on the year.

Impressive...

Nomura: Optical

Market Overview & Growth Projections

> Revenue Expansion: The global datacenter optical transceiver market is projected to reach USD 47.7bn in 2026F and surge to USD 144.2bn by 2028F, representing a rapid CAGR of 74% from 2026 to 2028F.

> Primary Growth Catalyst: Market expansion is primarily driven by robust, buoyant demand coming from large language model (LLM) training and inference workloads.

High-End Transceiver Shipment Forecasts

High-end speeds (800G and 1.6T) are experiencing stronger momentum due to accelerated technology upgrades within AI networking segments:

> 800G Transceivers: Projected shipments stand at 33.8mn units in 2026F, scaling to 55.0mn units in 2027F, and 78.0mn units in 2028F.

> 1.6T Transceivers: Projected shipments stand at 26.1mn units in 2026F, increasing to 71.5mn units in 2027F, and reaching 126.0mn units by 2028F.

> Next-Generation Speeds (2.4T / 3.2T): Initial shipments are expected to commence in 2027F and 2028F, ultimately reaching volumes of 5mn units and 2mn units respectively by 2028F.

Architectural Evolution & Supply Chain Dynamics

> Co-Packaged Roadmap: As the industry migrates toward co-packaged optics, deliveries of NPO (Near-Package Optics) and CPO (Co-Packaged Optics) products are expected to start rolling out between 2027F and 2028F, serving as fresh demand multipliers for high-power optical lasers.

> Supply Chain Constraints: Unprecedented demand cycles continue to pressure optical chip suppliers' fulfillment capabilities, giving IDM players like Yuanjie a competitive edge in a supply-constrained market environment.

$Coherent Corp.(COHR.US) $Lumentum(LITE.US) $Applied Optoelectronics(AAOI.US) $Macom Tech(MTSI.US) $Broadcom(AVGO.US)

BofA: AI LLMs

Frontier Leadership & Benchmarks

> Closed Models Still Lead: Proprietary (closed) models currently retain top slots on the hardest AI evaluations and benchmarks, including HLE, GPQA Diamond, and SWE-bench.

> Performance Gap: The performance gap between closed and open models widens most in critical enterprise value pools, specifically agentic tool-use, coding, and complex reasoning.

> Benchmark Examples: Top closed models like Claude Opus 5, GPT-5.6 Sol, and Gemini 3.1 Pro outperform open alternatives like GLM-52, DeepSeek V4, and Kimi K3 across major indices.

Adoption Curve & Standards

> Industry Standards: Closed model APIs serve as the reference implementation for the industry (defining chat/completions shapes, function-calling schemas, and evaluation norms) which open providers frequently mimic.

> Enterprise Preference: Closed models dominate enterprise adoption because they offer zero infrastructure burden, managed compliance, predictable SLAs, and single-vendor accountability. Conversely, self-hosted open models place GPU provisioning and compliance risks directly on the buyer.

The Nature of Open Models

> Distillation Concerns: Unlike Linux (which was built via clean-room development), leading open models are widely believed to be derivatives of closed models created via distillation. For example, US officials and labs have alleged connections between certain overseas open models and frontier US closed models.

> Risks: Open-weight models—many of which originate from China—carry ongoing intellectual property (IP) and national-security risks that could be heavily regulated or altered by governments or closed-lab developers.

Coexistence & Market Outlook

> Coexistence Over Displacement: The long-term market outcome is expected to be a coexistence model similar to Windows and Linux, rather than open-weights entirely replacing closed systems.

> Cost vs. Capability: Open models generally win on cost and efficiency (running 5–10x cheaper per token), while closed models win on technological leadership, adoption, and enterprise trust.

> Implications for Semiconductors: Semiconductor demand can benefit regardless of the winner:

If closed models lead, frontier-scale training and runs will likely remain highly capital-intensive.

If open models proliferate, cheaper tokens will drive broader usage and surge overall inference volumes (Jevons paradox).

Goldman Sachs: Memory

Market & Pricing Outlook

> DRAM Pricing Growth: Conventional DRAM pricing is expected to see solid double-digit sequential growth in both the third and fourth quarters of 2026, driven by an ongoing supply shortage.

> HBM Upside Next Year: High Bandwidth Memory (HBM) pricing could potentially double next year due to rising conventional DRAM pricing. Goldman Sachs expects SEC's HBM pricing to rise 87% year-over-year, outpacing the Bloomberg sell-side consensus estimate of 52%.

> Capacity vs. Bit Growth: While capacity additions this year are outpacing historical levels, overall bit growth is expected to remain below historical averages because of a high HBM trade ratio.

Industry Dynamics & Competition

> Long-Term Agreements (LTAs): LTAs are providing strong binding power—covering over half of server DRAM—backed by terms such as sizable prepayments, take-or-pay clauses, and cancellation penalties. Coverage ratios are expected to rise.

> Chinese Suppliers: Chinese memory competitors pose a limited threat in the near-to-mid term due to sizable gaps in production yields, technology levels, and product reliability, despite active capacity expansion.

> Hybrid Bonding: Adoption is expected to be gradual. Although stacking more DRAM dies makes existing bonding tech harder, achieving mass-production yields for hybrid bonding will take considerable time, leading companies to explore options like fluxless bonding in the interim.

Meta EPS miss is an overreaction.

"The second quarter 2026 general and administrative expenses include $2.40 billion of charges related to legal proceedings."

Without this, they would've likely beat EPS.

$Meta Platforms(META.US)

$Meta Platforms(META.US)

Revenue: $60.8B vs $60.24B expected

EPS: $6.18 vs $7.14 expected

Full-year capex: $130B to $145B vs $125B to $145B previously

SK Hynix $SK Hynix(SKHY.US) on HBM price negotiations for 2027:

"Discussion is underway for 2027 HBM supply volume and pricing with our key customers, which is progressing smoothly, supported by solid customer demand. But of course, we cannot disclose the contractual terms or pricing details for individual customers. With conventional DRAM prices rising sharply in recent months, such market environment may also have some influence on our HBM pricing discussions. Having said that, of course, HBM pricing is not determined solely by conventional DRAM prices."

BofA: Intel

Investment Rating & Price Objective

> Rating & Price Objective: Maintain BUY with a Price Objective (PO) of $160.00 USD (against a stock price of $91.67 USD).

> Valuation Basis: Based on 31x CY30E EPS power of $6+, discounted back two years to account for long-term server CPU and external foundry wafer/packaging opportunities.

Server & Market Positioning

> Supply-Driven Share: Server market share through 2026–2028 is viewed as a function of supply rather than design. Coral Rapids (slated for 18A-P in 2027) will further help close the performance gap against competitors.

> ASP Strength: Q2 server Average Selling Price (ASP) jumped +48% YoY, driven by higher core-count Granite Rapids adoption.

> Competitive Edge in AI: Intel argues that agentic AI workloads vary in requirements and may not always favor ARM or AMD. Intel’s NVLink design-in establishes a level playing field for system-level integration, while x86's security lead provides an advantage in enterprise AI.

> PC TAM: The PC Total Addressable Market (TAM) is tracking for a 10–12% YoY decline in 2026, though Intel sees limited inventory risk due to clear sell-in vs. sell-through visibility.

Intel Foundry Progress

> Yields & Margins: Under CEO Lip-Bu Tan, 18A/4/3 nodes are showing upside on yield, cycle times, and unit costs. 18A yields are expected to approach industry standards by the end of 2026, with a path to operating margin (OpM) breakeven by 2027 (excluding external customers).

> External Nodes (18A-P & 14A):

18A remains internal-only, while 18A-P targets external customers (risk production underway, 2027 volume committed).

14A High Volume Manufacturing (HVM) has been committed for 2028, featuring external customer engagements from the outset.

> Long-Term Financial Goals: Long-term gross margin targets are set at mid-40%+ to 50%+. Multiple financial levers—including prepayments, non-core asset sales, and equity—are available to support rising capital expenditure intensity.

Advanced Packaging (EMIB)

> Backlog & Ramps: The EMIB-T (Through-Silicon Via variant) backlog is actively building for a 2027 ramp-begin and full 2028 ramp.

> Revenue Potential: Each packaging engagement is projected to be worth multiple billions of dollars per year.

> Capacity & Constraints: Assembly and packaging (A/P) capital intensity is lower than wafers (roughly 1 to 5), and Intel already has capacity to handle multiple engagements in 2027. Current supply constraints are centered on external substrates, which are being managed via supplier prepayments.

$Intel(INTC.US)

KLA with macro assumptions on the semiconductor industry:

> CY25-CY30E semiconductor industry CAGR of ~11% 

> Wafer Equipment grows ~1 pt. faster than Semi to $215B +/- $20B 

> ~60% foundry / logic, ~40% memory 

> Process Control market grows > WFE

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

BofA: United Microelectronics

Rating & Price Objective Changes

> Price Objective (PO): Raised from NT$63 to NT$120 (and ADR PO raised from US$9.96 to US$18.97), based on a 17x target 2027E P/E multiple (up from 15x, sitting in the upper half of its historical 2x–25x range).

Earnings & Financial Forecasts

> EPS Estimates: Adjusted to NT$5.11 for 2026E, NT$7.09 for 2027E, and NT$9.60 for 2028E, driven by improving capacity utilization and wafer pricing.

> Revenue & Margins: Gross margins are projected to lift to 31.9% in 2026 and 36.9% in 2027 (based on 85% and 90% utilization rates, respectively), though analysts note these sit below overly aggressive market expectations of 40–45% GMs at full capacity.

Core Investment Thesis & Skepticism

> Overhyped Thematic Optimism: While UMC's foundational mature-node business is recovering, BofA believes market expectations are overly inflated regarding:

1. Silicon Photonics: Projected to represent only a small fraction (3% in '27 and 6% in '28) of UMC's total sales. Specialty Memory Production: Viewed merely as a cyclical fab filler.

2. Intel Collaboration (12nm): Limited customer traction is anticipated.

3. Traditional Tech Exposure: Over 80% of UMC's sales remain tied to traditional tech products, presenting downside risks if inventory builds moderate and customer pushback on pricing intensifies.

Industry & Mature Node Fundamentals

> Supply & Demand Dynamics: Mature 12-inch industry utilization (ex-China) is projected to improve from the low-80% range in 2026 to 85–90% in 2027. 8-inch node fundamentals look even more encouraging, moving from high-80% utilization in 2026 to 90–95% in 2027.

> Capacity Adjustments: Capacity output is influenced by decoupling trends, TSMC retiring older 6", 8", and mature 12" equipment (-130k WPM through 2025–2027), and Samsung cutting back 8" and selective mature 12" operations.

Recent Performance & Near-Term Outlook

> 2Q26 Tracking Stronger: Sales grew 13% QoQ, aligning with resilient consumer demand reported by major fabless players.

> Caution Ahead: While 3Q26 could see resilient early builds (+10%) as customers try to get ahead of cost hikes, analysts advise caution regarding year-end inventory adjustments and pricing pushback.

$United Microelectronics(UMC.US)