Semiconductor 2H25: Look Beyond

1H25 Market Review 

The first half of 2025 has been a volatile yet defining period for the semiconductor industry. While growth remains robust, the sector is still recovering from previous challenges, from unexpected technological breakthroughs in China to escalating trade wars.  

Year-to-date, the MVIS US Listed Semiconductor 25 Index recorded a positive performance of 18.2%, despite a -11.7% and -16% peak-to-trough decline following the DeepSeek and Liberation Day sagas. We have expressed our optimism in the semiconductor industry despite the sudden pullback due to the unprecedented events, as shown in the table below. 

Table 1: VanEck Semiconductor ETF YTD performance 

	Peak-to-Trough	Trough-to-Today
DeepSeek	-11.7% (1/22 - 1/27)	22.7%
US Liberation Day	-16.0% (4/2 - 4/8)	60.8%
YTD performance	18.2%

Source: Bloomberg Finance L.P., iFAST Compilation, Data as of 16 July 2025.

As highlighted by the title “Look Beyond”, we want to remind investors: ignore market noises and look beyond. The semiconductor industry continues to demonstrate resilience, with firmer optimism in long-term technological demand. 

Figure 1: VanEck Semiconductor ETF YTD performance 

DeepSeek's impact on semiconductor and AI 

In January, no one could miss the catalyst for the AI stock sell-off: Chinese startup DeepSeek launched a new AI model, challenging investors’ prevailing assumptions about the resources needed to develop high-performing AI models (e.g. capital, leading-edge semiconductor chips and power). In other words, DeepSeek could accomplish quite a lot, and at a much lower cost than its competitors. In the wake of the DeepSeek news, investors questioned whether hyperscalers (providers of cloud computing and operators of data centres) would continue their aggressive capex, and whether and when these investments would pay off. 

However, the phenomenon of the Jevons Paradox should also be taken into consideration. If the cost of training AI models becomes significantly cheaper, it could lead to an accelerated demand for inference—the process of running trained AI models in real-world applications. This increased demand for inference could, in turn, push up prices for high-end chips, particularly for popular components like AI accelerators from Nvidia or AMD, even if individual training costs decrease. In short, greater efficiency leads to greater consumption.

Figure 2: AI model, cost per million tokens, $

Source: One Useful Thing, JPMorgan, Data as of April 2025. 

Read more: DeepSeek Ignited China’s Chip Stock Boom - But How Long Can the Rally Last?

A divergence in the chip industry: AI and others 

The Q1 2025 earnings reports collectively reveal a distinct bifurcation in the semiconductor industry's performance. Companies at the forefront of AI chip design and advanced manufacturing are experiencing robust growth, which validates the narrative of AI as a powerful secular growth driver. Their strong year-over-year revenue increases, particularly in data centre segments, underscore the direct benefit they are reaping from the continued expansion of AI infrastructure.

In contrast, companies with less direct exposure to cutting-edge AI — such as Intel, which expects a general market contraction, and Samsung's memory division, which reported a profit decline driven by HBM speculation and falling average selling prices — are encountering significant headwinds. This polarisation suggests that the market may heavily favour firms with strong, direct linkages to AI infrastructure buildouts, while traditional segments may continue to experience volatility and competitive pressures in 2H25.

Foundry firms such as TSMC, Samsung Foundry, and GlobalFoundries outperformed on the back of robust demand for AI accelerators and specialised chip production. In addition, memory players including Micron and SK Hynix rebounded with improved pricing in DRAM and NAND, especially for high-bandwidth memory (HBM) used in AI training chips.

Despite the advent of AI PCs, the turnover cycle for consumer chips has been slow. Qualcomm and MediaTek face headwinds from subdued consumer sentiment in emerging markets, reflected in the slower adoption of new PCs and mobile gadgets.  Automotive and industrial chips remain in the early stages of recovery, and the pace of pickup has been slow due to the gloomy macroeconomic outlook.

Table 2: Semiconductor companies' latest earnings update  

Top 10 Holdings in SMH	Latest quarter revenue y/y growth (%)
Nvidia Corp	69.2
Taiwan Semiconductor Manufacturing Corp Ltd	41.6
Broadcom Inc	20.2
Advanced Micro Devices Inc	35.9
ASML Holding	46.3
Micron Technology Inc	36.6
Applied Materials Inc	6.8
Lam Research Corp	24.4
Texas Instruments Inc	11.1
Analog Devices Inc	22.3

Source: Bloomberg Finance L.P., iFAST Compilation, Data as of 2 July 2025.

AI secular growth remains intact, from cloud to edge 

We believe agentic AI, which refers to programs that can operate with a human-like level of intelligence and independence, will likely dominate the next wave of software innovation. Market leaders will deliver both a better consumer experience and productivity gains for enterprises across sectors and geographies. 

AI remains the undisputed growth driver for the semiconductor industry, with demand consistently outpacing supply. This is particularly evident for high-performance computing silicon, which is essential for LLM training and the extensive AI infrastructure buildouts currently underway globally. 

The influence of AI extends far beyond hyperscale data centres, permeating into diverse applications. Edge AI, encompassing smart sensors and autonomous devices, as well as industrial automation, is increasingly driving demand for compact, energy-efficient chips capable of real-time processing. 

Figure 3: AI Scaling Laws Drive Exponential Demand for Compute

Source: Nvidia, iFAST Compilation, Data as of 2 July 2025.

The AI race among countries 

Geopolitical tensions have emerged as a dominant factor shaping the semiconductor industry, introducing significant supply chain risks and accelerating strategic shifts. A global Sovereign AI Race is underway, with nations and economic blocs initiating ambitious strategies and making significant investments to build domestic AI infrastructure and semiconductor capabilities.

The confluence of US-China trade disputes, stringent export controls, and the heightened military activity in the Taiwan Strait is not merely causing temporary disruptions; it is fundamentally reshaping global semiconductor supply chains. This pressure is accelerating a trend towards regionalisation and "de-risking," where countries and economic blocs (such as the US, EU, Japan, India, and Middle Eastern nations) are investing heavily in domestic manufacturing and R&D capabilities. This strategic imperative is driven not just by economic competitiveness but by critical national security and resilience concerns.   

While the gap with the US in the AI race remains substantial, both China and the EU are making accelerating progress in select areas, particularly in application-specific domains and domestic infrastructure investments. For the near term, the United States is poised to maintain its undeniable leadership in the AI domain. This dominance is particularly evident in foundational models, advanced chip design, and software innovation, underpinned by its robust ecosystem of tech giants, venture capital, and world-class research institutions.

However, it would be a critical oversight to underestimate the immense potential emerging from other regions. China, for example, with its vast data resources and strong government backing, is rapidly advancing in specific AI applications. Areas like industrial AI and autonomous vehicles are prime examples where China's large-scale data control and application-specific development can yield significant competitive advantages.

We foresee geopolitical volatility as the new normal, yet it is not completely doom and gloom. Given this evolving environment, a balanced portfolio approach is advisable. This strategy should include a strategic overweighting in US semiconductor companies, allowing investors to capitalise on the current leadership and innovation driving the AI trend. Concurrently, maintaining exposure to the burgeoning opportunities and resilience factors emerging from other key regions will provide a more robust and adaptable investment posture.

The ultimate winners of the AI race may not yet exist

The AI secular growth story is unequivocally intact. 

Demand continues to outstrip supply for high-performance computing silicon, fueling robust growth for key players. While the ultimate winners in the broader AI application space may not yet be revealed, and innovation from companies like DeepSeek could shift the landscape, the foundational AI infrastructure players are indispensable at this nascent stage of AI development.

For investors, this juncture presents a clear strategy: focus on AI infrastructure providers. These companies, supplying the essential compute power (GPUs, custom ASICs, HBM), manufacturing capabilities (foundries, equipment), and underlying components, are direct beneficiaries of every dollar invested in AI R&D, data centre buildouts, and sovereign AI initiatives. Their demand is more immediate and quantifiable than that of nascent AI application layers. 

As the AI ecosystem matures, the winners among application layers will become clearer, but for now, the surest bet lies with the enablers of the AI revolution.

Table 3: Semiconductor players that serve as AI infrastructure proxy 

AI Infrastructure segment	Examples
GPU	Nvidia AMD
ASICs	Broadcom
Memory	SK Hynix Micron	Tariffs turn into opportunity: Potential upside for the South Korean market near 60%
Foundry	TSMC
Equipment Suppliers	ASML	ASML: Tariff uncertainty clouds near-term outlook but secular growth story remains
IDM	Intel	Intel FY25: More road to go

Source: iFAST Compilation, Data as of 11 July 2025

Tech giants are spurring CAPEX in AI race  

While the long-term ROI of these massive AI investments is a subject of debate, the competitive landscape often prioritises keeping pace with rivals over immediate profitability, signalling a strong belief in the future growth of AI.

Beyond physical infrastructure, software companies are heavily investing in developing and refining AI software, platforms, and services, including the development of advanced LLM such as ChatGPT from OpenAI, Gemini from Google and LLAMA from Meta. 

Companies are also actively integrating AI into their products to achieve enhanced productivity. Microsoft, for example, is integrating Copilot across its Microsoft 365 applications for productivity and automation, aiming to embed AI directly into enterprise software for direct monetisation and widespread business adoption.

Table 4: Latest development from Technology companies in AI 

Company	2025 AI CAPEX Estimate
Microsoft	$80 billion
Amazon	Over $100 billion
Alphabet	$75 billion
Meta	$64–72 billion
Apple	$500 billion (over 4 years)
IBM	$150 billion (over 5 years)
OpenAI	$40 billion (raised)
Anthropic	$3.5 billion (raised)
Alibaba	$53 billion (over 3 years)

Source: Various Media, Company Reports, iFAST Compilation, Data as of 2 July 2025.

SMH TP: $360, 23.7% upside potential 

In our view, current levels of AI-related capital expenditure and stock valuations do not suggest the presence of an AI bubble. Stock prices have risen in tandem with earnings, and capital expenditures have been funded largely from internal cash reserves rather than through equity or debt issuance.  Nonetheless, we are closely monitoring payback on AI projects and pockets of froth (in valuations and/or capex levels) in sectors such as data centres. 

Combining the catalysts and risks mentioned above, we maintain a rating of 3.5 stars “Attractive” for the semiconductor industry. Our target price for the VanEck Semiconductor ETF is USD 360, representing an upside potential of 23.7% as of 16 July 2025.

Table 5: SMH Valuation 

	2024	2025E	2026E	2027E
Earnings Per Share (EPS)	287.10	411.3	509.0	607.4
Earnings Growth YoY		43.3%	23.7%	19.3%
PE Ratio (X)	41.1	28.7	23.2	19.4
Upside Potential (based on fair PE Ratio of 24X)				23.7%
Target Price				360
Source: Bloomberg Finance L.P., iFAST Compilations. Data as of 16 Jul 2025

Key takeaway 

Focus on AI pure-plays and enablers in 2H25. 

We opine that the current AI-driven trend in the semiconductor industry will persist into 2H25. We recommend prioritising investments with intense, direct exposure to AI infrastructure that are best positioned to benefit from the robust and sustained AI tailwind. 

While we acknowledge that macro uncertainties may contribute to higher volatility in 2H25, we encourage investors to take a longer-term view — particularly on the transformative potential of AI as a game-changer.

Buying the dip at any unprecedented sharp drawdown due to externalities could offer an appealing upside in the long term. 

Declaration:

For specific disclosure, at the time of publication of this report, IFPL (via its connected and associated entities) and the analyst who produced this report hold a NIL position in the abovementioned securities.