Semiconductor 2Q26: Entering into the second half marathon

• The semiconductor sector’s forward return profile is likely to be characterised by still-strong but moderating earnings growth, a gradual normalisation of valuations from AI-driven peaks, and elevated volatility driven by geopolitical event risk.
  
  
• Our base case is therefore not a collapse in spending, but a normalisation in growth. Spending levels should remain high, but another leg of explosive growth from here is less likely without a meaningful technological or commercial breakthrough.
  
  
• Among SMH, we like foundry and memory. We maintain a Neutral rating on SMH with a target price of USD 490, implying 25.3% upside potential from the current price of USD 392. Alternatively, investors can seek Asian Semiconductor (HKEX:3119) for more foundry and memory exposure. 

“Entering into the Second Half Marathon” reflects a shift from the adrenaline-fueled sprint of the past year to a phase where endurance and execution take centre stage. While we remain structurally bullish, backed by superb earnings and the undeniable momentum of AI, we have to acknowledge that the easy miles of this rally are likely behind us. After such an aggressive re-rating, the market's path forward is expected to be more measured and demanding, making a neutral stance the most prudent approach.

SMH performance for the past 1 year

As of 2 April 2026, the  VanEck Semiconductor ETF (NASDAQ:SMH) closed at USD 392, delivering approximately 9% YTD returns. The share price rallied strongly at the start of the year and peaked at USD 426, before retracing in March amid the US-Iran conflict.

We have maintained a constructive stance on semiconductors since 2022, and that call has largely played out as expected. The figure below summarises our stance over the past year: we reaffirmed our positive view earlier in the year, while gradually tempering optimism as upside potential narrowed despite continued earnings strength.

That positioning has held up well. With the latest price action and earnings revisions, we are now revising our target price upward while maintaining a Neutral view on SMH.

Figure 1: SMH performance

Notable performance divergence across the holdings

The semiconductor sector delivered an extraordinary 1Q26 earnings season: Nvidia’s USD 68.1bn revenue, Broadcom’s 106% y/y AI growth, TSMC’s 62% gross margin, Micron’s contracted HBM supply, and ASML’s record €13.2bn bookings collectively painted a picture of a sector in the midst of a generational growth cycle.

The AI infrastructure buildout is real, multi-year, and accelerating. Yet the paradox remains: the market penalised the stocks that delivered these strong earnings.

The message is clear: the era of multiple expansions on AI hype is over.

Performance was heavily bifurcated. Foundry and memory sub-sectors led (TSM +19%, MU +30%, INTC +20%), while the dominant fabless heavyweights lagged or declined (NVDA -2%, AVGO +5%, AMD +5%). Equipment stocks delivered mixed-to-positive returns (ASML +15%, AMAT +10%), supported by record bookings and >20% WFE growth guidance for 2026.

The central paradox of 1Q26 was that the best-reporting companies received the worst share price reactions. Nvidia delivered record quarterly revenue (+73% YoY) and guided USD 78bn for Q1 27, yet shares fell -5.5% post-earnings and remain stuck in a six-month trading range. Broadcom posted 106% AI revenue growth and guided Q2 at USD 22b (+47% y/y), yet the stock barely moved.

Table 1: Summary of components in SMH

Sub-sector	Key Holdings	% of SMH	1Q26 YTD	Key Theme
Fabless / GPU	NVDA, AVGO, AMD, QCOM, MRVL	~40%	Mixed (-2% to +5%)	AI compute; priced for perfection
Foundry / IDM	TSM, INTC	~16%	+19% to +20%	Best visibility; value re-rating
Memory	MU	~5%	30%	HBM supercycle; contracted supply
Equipment	ASML, AMAT, LRCX, KLAC	~20%	+10% to +15%	Record backlog; WFE US D135b
Analog / Mixed	ADI, TXN, NXPI, ON, MCHP	~15%	Flat to +5%	Cyclical trough; recovery 2H26
EDA / IP	ARM (if held)	~4%	Varies	AI chip design complexity

Source: Bloomberg Finance L.P, iFAST Compilation, Data as of 31 March 2026.

The fundamental problem with an overweight position in SMH is structural: the index is ~33% concentrated in fabless heavyweights (NVDA, AVGO, AMD), which have entered a regime where delivering strong earnings is necessary but no longer sufficient for sustained share price appreciation. The market now demands evidence of durable, accelerating growth, not just earnings beats, and has shown a willingness to sell into strength when expectations are merely met rather than massively exceeded.

In other words, the hurdle rate has moved higher. Continuously beating elevated expectations is becoming increasingly difficult.

Structural outlook remains intact and why investors still need exposure

The semiconductor industry is entering what may be the most consequential structural inflection since the advent of the smartphone era.

At the core of this transformation are several secular demand vectors: AI-driven data centre buildout, the proliferation of edge computing and AI-enabled devices, and the electrification and digitalisation of the automotive fleet.

The AI data centre cycle alone is expected to sustain hyperscaler capex above USD 600bn annually through at least 2028, underpinning demand for leading-edge logic (TSMC 3nm/2nm), high-bandwidth memory (HBM4 and beyond), advanced packaging (CoWoS, hybrid bonding), and the lithography and deposition equipment that enables these technologies.

That said, the structural outlook is not without tension. The industry’s capital intensity is rising sharply: TSMC’s USD 52–56b 2026 capex, Intel’s multi-year fab buildout, and Micron’s USD 20bn annual investment collectively represent the highest sustained level of semiconductor capital expenditure in history.

While demand visibility is strong today, the semiconductor industry remains inherently cyclical, and the lag between capacity investment and production creates a window for potential oversupply, particularly in memory and mature-node logic. The industry’s supply chain has also become more geopolitically fragmented, with US-China tech decoupling, export controls on advanced chips and equipment, and the weaponisation of rare earth and energy supply chains all adding structural friction to what was once a globally optimised manufacturing ecosystem.

For investors, the implication is clear: the semiconductor sector’s forward return profile is likely to be characterised by still-strong but moderating earnings growth, a gradual normalisation of valuations from AI-driven peaks, and elevated volatility driven by geopolitical event risk.

Can CAPEX growth be sustained?

Table 2: Hyperscaler 2026 CAPEX guidance

Hyperscaler	2026 Capex Guidance	Capex/Revenue
Amazon	$200B	~25%
Alphabet	$175–185B	~46%
Microsoft	$120B	~47%
Meta	$115–135B	~54%
Oracle	$50B	~86%
Total	$660B

Source: Bloomberg Finance, iFAST Compilation, Data as of 31 March 2026

The major hyperscalers are projected to spend a combined ~USD 660b in capex during 2026, with the bulk directed at AI data centre infrastructure. This figure has risen from roughly USD 200b in 2024, representing a threefold increase in two years. Additionally, neo-cloud players (CoreWeave, Nebius) and sovereign AI initiatives are adding incremental demand.

This CAPEX wave is the primary demand driver for the semiconductor ecosystem: NVIDIA and AMD GPUs, TSMC foundry capacity, Micron HBM, Broadcom custom ASICs, and ASML/AMAT/LRCX equipment.

The critical question is durability: is this a three-year buildout that peaks around 2027,  or a decade-long structural shift?

Figure 2: Big Tech consensus CAPEX in 2030e

Source: Bloomberg Finance, iFAST Compilation, Data as of 31 March 2026

We lean towards the latter, though we acknowledge that growth rates will decelerate.

Why?

The projected deceleration in hyperscaler CAPEX growth is driven primarily by the law of large numbers and a shift from a land-grab phase to a focus on Return on Investment (ROI).

Maintaining high double-digit growth on a massive USD 660bn base is mathematically difficult, as it would eventually consume an unrealistic share of corporate revenue. Furthermore, physical bottlenecks, specifically power grid constraints, land availability, and the engineering complexity required for liquid cooling, create a natural ceiling on the pace of physical deployment, regardless of capital availability.

The AI monetisation debate is also likely to intensify. Investor focus is shifting from rewarding AI potential to demanding AI profitability. This introduces a digestion phase, where hyperscalers must prove they can monetise their installed infrastructure before committing to another leg of exponential spending.

Our base case is therefore not a collapse in spending, but a normalisation in growth. Spending levels should remain high, but another leg of explosive growth from here is less likely without a meaningful technological or commercial breakthrough.

Key technological advancement

Stepping away from the financials, the underlying technology remains highly supportive.

On the compute front, NVIDIA’s Blackwell architecture has reached full production scale, with networking revenue tripling year-over-year as the platform gains broad adoption. The Vera Rubin next-generation platform was previewed at CES 2026 and is targeting H2 26 production, compressing Nvidia’s product cadence to an annual cycle that keeps hyperscalers locked into continuous upgrade commitments. Meanwhile, AMD’s MI500 series is in active development as the successor to the MI300 family, with MI455X and Helios rack-scale systems already announced through OEM partnerships with HPE and Lenovo.

At the foundry level, TSMC’s 2nm node is sold out for its initial production ramp, with the N2P performance-enhanced variant now introduced and the A16 node, featuring backside power delivery, scheduled for H2 2026. CoWoS advanced packaging capacity, the most binding bottleneck in the AI supply chain, continues to be aggressively expanded. Intel’s 18A process is tracking 7–8% monthly yield improvement, with volume production underway and 14A customer engagements expected to drive supplier decisions by late 2026 or early 2027. The Nvidia equity investment and co-development partnership has materially bolstered Intel Foundry’s credibility.

In memory, Micron’s HBM4 with pin speeds exceeding 11 Gbps is on track to ramp with high yields in Q2 26, with the entire calendar year supply already contracted under price and volume agreements. The HBM total addressable market is now projected at USD 100bn by 2028, pulled forward by two years from prior estimates.

On the custom silicon front, Broadcom projects USD100bn in AI chip revenue by 2027 from five hyperscaler customers, while Marvell is also scaling its custom ASIC business, a trend that represents both a growth driver for the broader ecosystem and a competitive risk to Nvidia’s merchant GPU franchise.

Underpinning all of this is the largest equipment spending cycle in semiconductor history: ASML’s record €13.2bn Q4 bookings (the majority for 2027 delivery), Applied Materials’ guidance for >20% 2026 equipment growth, and total WFE projected at ~USD135b collectively signal sustained multi-year infrastructure investment.

Sub-sector analysis

Memory (Micron): Overweight

Micron is the only US-listed pure-play memory company with direct HBM exposure, and the investment case rests on an unusually attractive combination of demand visibility and valuation dislocation. The company has completed price and volume agreements for its entire 2026 HBM supply, including its industry-leading HBM4 product, a level of revenue certainty that is virtually unusual in the memory industry. HBM4, with pin speeds exceeding 11 Gbps, is on track to ramp with high yields in Q2 26, aligning with customer product plans from NVIDIA and other accelerator designers.

The structural dynamics remain compelling. The HBM total addressable market is now projected to reach USD100b by 2028, representing a ~40% CAGR and pulled forward by two full years from prior estimates. Critically, HBM’s 3:1 wafer trade ratio versus DDR5 means that every incremental HBM wafer displaces three DDR5-equivalent wafers, creating a structural DRAM shortage that supports pricing power across the broader memory stack. This dynamic is already visible in Micron’s Q1 26 results: gross margins expanded to 57% from 39.5% a year ago, and free cash flow hit a company record of USD 3.91b.

The primary risks are memory cyclicality, potential NAND oversupply, and the execution demands of a USD 20bn annual capex programme. However, contracted HBM volumes provide an unprecedented revenue floor that fundamentally improves the risk profile compared to prior memory cycles. The stock is up ~30% YTD, but we believe further re-rating is warranted as the market increasingly recognises that this is not a typical memory cycle, it is a structural shift in the economics of the memory industry.

Related articles: Micron Q2 26 Earnings: Stronger Than Ever, Yet Questioned

Foundry (TSMC): Overweight

TSMC remains the irreplaceable foundation of the AI supply chain, commanding ~70% foundry market share and manufacturing chips for virtually every major fabless designer, including Nvidia, AMD, Apple, Broadcom, and Qualcomm.

This structural dominance is reinforced by the company’s unmatched process technology leadership: 2nm capacity is sold out for initial ramp, CoWoS advanced packaging expansion is addressing the most binding bottleneck in AI infrastructure, and the 2026 capex commitment of USD 52–56bn signals management’s conviction in multi-year demand sustainability.

The financial profile also supports the thesis. Revenue is expected to grow ~30% in USD in 2026, with gross margins expanding to 63–65% in Q1 26, representing arguably the best operating leverage profile in the semiconductor sector. AI accelerator revenue is projected to compound at a mid-to-high 30s percent CAGR through 2029, providing rare long-duration visibility for what the market still tends to treat as a cyclical business. At ~21x forward P/E with ~30% earnings growth, TSMC trades at a PEG of ~0.7x, a meaningful discount to NVIDIA’s 27x despite comparable quality and arguably superior structural positioning. The primary risk remains Taiwan’s energy supply vulnerability in the context of the Iran war and Middle Eastern LNG disruption, though the Arizona fab ramp provides partial geographic diversification over the medium term.

Fabless Heavyweights (NVDA, AVGO, AMD): Neutral

These remain outstanding businesses with clear secular tailwinds, but at current levels, they are still priced to perfection. The risk/reward is therefore more balanced: strong earnings will sustain prices, but meaningful upside requires new catalysts, whether through new product cycles, further margin expansion, or broader evidence of AI monetisation. NVDA’s PEG of 0.5x is mathematically attractive, but a market capitalisation above USD 4tn naturally introduces fund-flow and expectation constraints. We therefore recommend maintaining existing positions, but would be more selective on adding fresh exposure at current levels.

Related articles: 

Nvidia Q426: The secret everyone knows

Broadcom 1Q26: Better earnings, flatter share price 

AMD Q425: Beat and Dip

Equipment (ASML, AMAT, LRCX, KLAC): Neutral to Slight Overweight

Record backlogs, >20% WFE growth guidance for 2026, and multi-year EUV demand provide solid visibility. However, margin mix headwinds (3600-series EUV tools), China revenue declines, and regulatory risk (AMAT’s BIS settlement) create near-term uncertainty. Within the group, ASML’s backlog visibility through 2027 still offers the most attractive risk/reward.

Related articles:

 ASML: Tariff uncertainty clouds near-term outlook but secular growth story remains

Analog/Industrial (ADI, TXN, NXPI, ON, MCHP): Watch for 2H2026 entry

The cyclical trough in analog and industrial semiconductor appears near, with restocking likely to begin in 2H 2026. Texas Instruments’ large-scale 300mm capacity investment positions it well for the eventual recovery, though near-term earnings momentum remains limited. NXP and ON Semiconductor are more directly levered to an automotive recovery. We recommend building watchlists and closely monitoring leading indicators (auto production, industrial PMIs) to confirm the turn.

Maintain Neutral, TP at USD 490 with 25.3% upside potential

Combining the above, we hold a constructive view of the semiconductor sector's long-term technological advancement and earnings trajectory. Accordingly, we have revised forward EPS upward to express our improved earnings assumptions, while also acknowledging that geopolitical tensions, supply chain bottlenecks and elevated expectations could continue to cap multiple expansions.

As such, we maintain a Neutral 2.5 stars rating on SMH with a target price of USD 490, implying 25.3% upside potential from the current price of USD 392.

This may appear generous for a Neutral call, but the rating reflects our view that upside is still dependent on execution into FY2027, rather than offering an obviously asymmetric entry point today.

Table 3: SMH’s valuation

	FY25	FY26	FY27
EPS (Index)	446.8	666.8	833.6
Growth y/y		49.3%	25.0%
Current level (index)	16008
Current price (SMH ETF)	392
PE	35.83	24.01	19.20
Fair PE			24
Upside potential			25.3%
Target Price			490

Source: Bloomberg Finance, iFAST Compilation, Data as of 2 April 2026.

Key Takeaway

SMH is still worth holding. The AI infrastructure buildout remains the most significant capex cycle since the internet era. Forward EPS growth of ~70% is exceptional, and a current P/E of 22–24x is justifiable on a PEG basis. However, at the index level, the risk/reward is not balanced, but rather outright compelling. The upside requires delivery of FY2027 EPS, which introduces execution risk.

As such, we maintain a Neutral rating on SMH with a target price of USD 490, implying 25.3% upside potential from the current price of USD 392. Alternatively, investors can seek Asian Semiconductor (HKEX:3119) for more foundry and memory exposure. 

For investors who are already positioned, we would maintain current holdings while directing new capital more selectively, including allocations to lower-correlation regions like Asia, to improve portfolio diversification.

For first-time investors, a phased entry approach via a Regular Savings Plan (RSP) remains a more disciplined way to build exposure as the sector transitions into a more mature phase of the cycle.

Disclaimers:

For specific disclosure, at the time of publication of this report, IFPL (via its connected and associated entities) holds a NIL position in the abovementioned securities. The analyst who produced this report holds a position in Global X Asia Semiconductor ETF (3119.HK).

This research report was prepared with the assistance of artificial intelligence (AI) tools. iFAST Financial Pte Ltd does not rely exclusively on AI for content generation; the content of this report – including all investment theses, ratings, price targets and conclusions – has been independently reviewed and verified by the research analyst(s) to ensure accuracy and professional integrity.