Synthetic intelligence (AI) continues to be probably the most transformative know-how of our period, with semiconductor firms main the cost and powering groundbreaking developments. After a strong restoration in 2024 pushed by demand for logic and reminiscence chips, the worldwide semiconductor market is forecast to develop by 15% this 12 months, reaching a complete worth of $728 billion, with the Americas and Asia Pacific anticipated to steer the expansion.1 The growth of information facilities continues to drive vital development, particularly for firms specializing in AI and semiconductor improvements. International gross sales in June 2025 have been US$60 billion, representing a year-over-year enhance of 20%.2
chart1
Supply: Semiconductor Business Affiliation. As of August 4, 2025.
As AI-driven monetization alternatives start to take form, hyperscaler capital spending stays on the rise regardless of mounting tariff and financial headwinds. International knowledge middle capex soared by 53% year-over-year in Q1 2025, marking the sixth straight quarter of double-digit annual growth.3 Microsoft, Amazon and Google reported that demand persistently exceeds out there infrastructure capability for AI workloads, with projections indicating that extra capability will proceed to broaden all year long. Amazon is about to take a position not less than US$20 billion in Pennsylvania4 and US$13 billion in Australia5 to broaden its knowledge middle infrastructure for AI and cloud providers. Meta’s capital spending might enhance additional in 2026, as it’s constructing a number of multi-gigawatt knowledge middle clusters to gas its AI ambitions, with the primary facility slated to go reside subsequent 12 months. The corporate has made AI central to its promoting technique and plans to allow manufacturers to completely design and goal campaigns utilizing AI instruments by the tip of subsequent 12 months. Based mostly on the shopper’s funds, these new instruments would generate the complete commercial, together with photos, movies and textual content, and ship it to the focused viewers.6
Up to now, AI demand has primarily centered on coaching workloads, notably for frontier fashions. Whereas main tech firms proceed pouring assets into constructing ever-larger AI fashions, they’re additionally reallocating extra funding towards inference. Inference is the stage the place skilled AI fashions course of new knowledge to generate insights, make predictions or assist decision-making. Whereas coaching a mannequin is basically a one-time expense, prompting a mannequin (inference) produces tokens, every of which carries a price. Through the Google I/O 2025 keynote, Alphabet CEO Sundar Pichai shared that the agency processed 480 trillion tokens throughout its merchandise and APIs in April 2025, 50 instances greater than the identical month a 12 months earlier.7 The speedy surge in token quantity displays rising utilization and adoption of AI fashions, signaling a larger want for computing energy and driving larger demand for chips.
The shift in funding towards inference has additionally gained momentum with the launch of latest reasoning fashions. Whereas conventional AI fashions reply swiftly and excel at sample recognition, they typically fail to grasp broader contexts and wrestle with advanced reasoning. Reasoning fashions are constructed to deconstruct advanced issues into smaller, manageable steps and resolve them via express logical reasoning. They’re particularly skilled to indicate their work and comply with a extra structured thought course of, which leads to longer computation instances for person queries. These fashions demand considerably extra compute throughout inference to cause via intricate issues. This evolution from fundamental sample recognition to structured reasoning is pivotal to AI, unlocking its potential to deal with advanced real-world challenges successfully. As AI adoption quickly expands, demand for inference will correspondingly intensify.
AI brokers are positioned to revolutionize how organizations operate, delivering breakthroughs in productiveness and operational effectivity. They’re clever programs designed to execute duties independently by comprehending goals, formulating choices and taking actions to attain predetermined targets. Whereas people outline the specified outcomes, AI brokers autonomously choose optimum actions required to perform these targets. These brokers boast a broad spectrum of makes use of, from supporting tutorial analysis and streamlining on-line purchases to planning leisurely holidays. Customer support, gross sales and advertising, and IT and cybersecurity are the three enterprise capabilities the place AI brokers are most steadily deployed or deliberate for implementation within the subsequent six months.8 As enterprises progressively combine AI brokers throughout numerous operational purposes, demand for computational infrastructure is escalating dramatically.
Hyperscalers are more and more centered on ASIC (application-specific built-in circuits) infrastructure to satisfy surging AI demand. ASICs are custom-built for particular workloads and may execute these duties way more effectively and at a considerably decrease price than high-performance GPUs. Though the preliminary funding to develop ASIC infrastructure is appreciable, the long-term price of working GenAI workloads on these chips is predicted to be decrease as soon as the upfront expense is absorbed. For instance, in April 2025, Google unveiled Ironwood, its seventh-generation Tensor Processing Unit (TPU), particularly designed for inference workloads.9 Whereas Google’s in-house TPUs have been as soon as restricted to inner use, the corporate is increasing exterior entry to drive sooner development of its cloud enterprise. Marvell Expertise initiatives the {custom} computing system market will surge to US$55.4 billion by 2028, greater than eight instances its dimension in 2023.10
HBM represents a cutting-edge reminiscence know-how engineered to ship sooner knowledge entry whereas lowering vitality consumption, which is crucial for the efficiency of AI processing. HBM’s market share within the dynamic random entry reminiscence (DRAM) section is forecast to leap from 18% in 2024 to greater than 50% by 2030.11 Beginning with the next-generation HBM4, the bottom die can be produced utilizing logic processes, enabling decrease energy consumption and customizable options tailor-made to shopper necessities. Pushed by escalating computational calls for from AI coaching and inference workloads, HBM’s market outlook stays robust. As the first HBM provider for Nvidia, holding a 62% share of world HBM shipments in Q2 202512, SK Hynix initiatives the worldwide HBM market to broaden by 30% yearly via 2030.13
Masking the 30 largest US-listed shares and ADRs of firms primarily concerned within the design, distribution, manufacture and sale of semiconductors, Nasdaq’s PHLX Semiconductor™ Index (SOX) delivered a complete return of 96% over the previous three years, outperforming the NYSE Semiconductor Index by 12 proportion factors and virtually doubling the return of the S&P Semiconductors Choose Business Index.
semi3
Supply: Nasdaq International Indexes, Bloomberg. As of July 31, 2025.
SOX is a modified market capitalization-weighted index, with the highest three constituents by market capitalization capped at 12%, 10% and eight%, respectively, and the remainder capped at 4% throughout quarterly rebalancing. For the complete index methodology, please go to our web site.
As of the tip of July 2025, the ten largest constituents accounted for 61.5% of the index weight. 79.8% of the index weight is within the Semiconductor Subsector, with the remainder within the Manufacturing Expertise Tools Subsector, based on the Business Classification Benchmark (ICB) classification system.
semi4
Supply: Nasdaq International Indexes, FactSet. As of July 31, 2025.
All seven largest holdings posted constructive whole returns over the previous 12 months. On common, the highest 10 corporations achieved a one-year whole return of 21%. Inside the identical {industry}, inventory performances assorted considerably. The distinction in one-year whole returns between the best- and worst-performing shares among the many high 10 constituents (Broadcom and NXP Semiconductors) was a staggering 102 proportion factors. This demonstrates the significance of diversification, even when investing in a single sector or theme.
semi5
Supply: Nasdaq International Indexes, FactSet. As of July 31, 2025.
As the most important constituent in SOX, Nvidia was the third-best performer amongst all constituents, gaining 52% over the previous 12 months. It grew to become the primary publicly traded firm in historical past to attain a US$4 trillion market valuation in July 2025, after becoming a member of the trillion-dollar membership in Might 2023.14 Nvidia’s technological edge stays pronounced, as Blackwell shipments speed up, pushed by hovering AI reasoning demand and realized economies of scale. Past AI, the corporate identifies robotics as its most substantial addressable development alternative, with autonomous autos representing the primary main industrial deployment. The agency not too long ago unveiled new NVIDIA Omniverse libraries and NVIDIA Cosmos world basis fashions that empower builders to construct next-generation robots and autonomous autos by integrating AI reasoning with scalable, bodily correct simulations.15
Because the second-largest constituent in SOX, Broadcom emerged as one of the best performer amongst all index elements, producing a one-year whole return of 85% via July 2025. The corporate continues to dominate the AI ASIC and AI networking semiconductor markets. The chipmaker is engaged with the highest seven hyperscalers for {custom} silicon merchandise, together with accelerators for Google, Meta and ByteDance. Furthermore, Broadcom has vital publicity to enterprise software program following its acquisition of VMware in 2023. Its infrastructure software program division posted a 76% working margin final quarter, up from 60% a 12 months in the past.16
Because the world’s largest contract semiconductor producer, TSMC ranks because the third-largest SOX constituent, delivering a 47% whole return over the trailing twelve months.17 With web income from its high-performance computing section increasing to 60% from 52% a 12 months in the past, AI-driven demand stays the first development catalyst for TSMC, whereas its market dominance offers it with pricing energy. Superior chips with sizes 7nm or smaller accounted for 74% of the corporate’s whole wafer income in Q2 2025. Regardless of the US president’s current announcement of tariffs on semiconductor imports, Taiwan confirmed that TSMC secured an exemption from the newest levies given its substantial American manufacturing investments.18
Past the escalating demand for superior coaching capabilities to assist more and more advanced AI mannequin architectures, AI inference has surfaced as a pivotal development driver. The speedy enhance in token quantity signifies increasing utilization and adoption of AI fashions. The proliferation of AI brokers is poised to rework varied industries and considerably enhance compute demand. This dynamic panorama highlights the pivotal position of the semiconductor sector in driving the subsequent wave of AI innovation.
Nasdaq’s PHLX Semiconductor Index (SOX) delivered a complete return of 96% over the three years via July 2025. Funds monitoring SOX embody the Invesco PHLX Semiconductor ETF (Nasdaq: SOXQ), the Mirae Asset TIGER US PHLX Semiconductor Sector Nasdaq ETF (South Korea: 381180), the Cathay PHLX Semiconductor ETF (Taiwan: 00830), the International X Semiconductor ETF (Japan: 2243) and the Yurie PHLX Semiconductor Index Fund (South Korea: 7D01596). The Mirae Asset TIGER Synth-US PHLX Semiconductor Sector Leverage ETF (South Korea: 423920) tracks SOX with two instances leverage.
Written by David Tsoi, Head of Index Analysis, APAC
Nasdaq®, PHLX Semiconductor™, SOX™ and Nasdaq-100® are registered logos of Nasdaq, Inc. The data contained above is supplied for informational and academic functions solely, and nothing contained herein ought to be construed as funding recommendation, both on behalf of a specific safety or an general funding technique. Neither Nasdaq, Inc. nor any of its associates makes any advice to purchase or promote any safety or any illustration in regards to the monetary situation of any firm. Statements relating to Nasdaq-listed firms or Nasdaq proprietary indexes should not ensures of future efficiency. Precise outcomes could differ materially from these expressed or implied. Previous efficiency isn’t indicative of future outcomes. Buyers ought to undertake their very own due diligence and thoroughly consider firms earlier than investing. ADVICE FROM A SECURITIES PROFESSIONAL IS STRONGLY ADVISED.
