According to the latest Global I/O Technology Hardware Research Report released by international financial giant UBS, as cutting-edge AI agent workflows such as Muse and Astra significantly expand the range of business tasks that can be completed automatically, global AI computing demand continues to surge, substantially extending the memory chip price increase cycle and making the "memory supercycle" thesis increasingly solid. However, UBS analysts stated that consumer electronics budget growth cannot keep pace, leaving the PC industry facing a "ruptured adverse situation" of rising prices, declining shipments, and severely diverging profitability.

UBS essentially maintained its forecast of approximately 241 million global PC shipments in 2026, a year-over-year decline of 11%, but revised its 2027 shipment growth forecast from approximately 2% growth to approximately 4% decline, expecting that price increases can only partially offset volume losses. Under the broader trend of agentic AI proxy workflows becoming widespread, driven by the world's most advanced AI agents and AI large models such as Meta Muse and OpenAI Astra, incremental computing demand will continue to spread to servers, networking, memory, and supporting components, while whether consumer electronics companies can benefit depends on product appeal, procurement conditions, and cost pass-through capability.

UBS calculation data shows that since mid-2025, the costs of dynamic random access memory (DRAM) and NAND flash have surged by 766% and 471% respectively. These cost increases are driving PC prices higher. UBS analysts emphasized that the sustained large-scale price increases in memory chips mainly suppress demand in PCs and smartphones and other consumer electronics, while critically important components such as AI GPU-equipped server clusters, data center CPUs, high-performance Ethernet network infrastructure, and data center optical interconnect systems are benefiting from unprecedented AI infrastructure boom catalyzed by the global popularity of Muse and Astra, generating new orders and unit product value increases. The hardware supply chain is immune to the "cost of memory price increases" and therefore presents different profitability trajectories.

Compared to top-tier global consumer electronics product line suppliers like Apple, UBS therefore prefers the AI server and high-performance network infrastructure supply chains, as well as manufacturers of core AI infrastructure-related components such as MLCC, copper foil, and electronic cloth that can increase unit product value, while taking a positive view on the prospects of traditional large PC manufacturers like Dell shifting their business growth focus to AI computing infrastructure. For hardware manufacturers such as Hon Hai Precision, Quanta Computer, Wiwynn, Wistron, and Delta that focus on both AI server manufacturing chains and consumer electronics product lines, UBS firmly maintains its most bullish "Buy" rating, with the bullish focus leaning more toward the massive expansion of AI server cluster demand under the AI computing frenzy. Below are the AI hardware supply chain leaders that UBS favors as beneficiaries of the AI infrastructure megatrend.

From Price Hikes and Spec Cuts to Investment Divergence: UBS Dissects Six Impacts of the Memory Supercycle on the Consumer Electronics Supply Chain

First, the PC market is digesting the pull-forward from advance purchases, with cost increases further suppressing subsequent sales. UBS's latest forecast data shows that global PC shipments in 2026 are expected to be 241.0 million units, a year-over-year decline of 11.0%; 2027 is revised down from the previous 246.6 million units to 231.1 million units, a 6.3% reduction in the forecast, with year-over-year growth turning from 2.1% growth to 4.1% decline. Average selling price is expected to rise from $837 in 2026 to $888 in 2027, a 6.0% increase, allowing 2027 industry revenue to still grow 1.7% to approximately $205 billion, but below the previous forecast of approximately $210 billion. By market segment, 2027 consumer PC shipments are expected to decline 6.3%, commercial PCs to decline 3.0%, and Chromebooks to decline 14.6%, reflecting more pronounced pressure on price-sensitive groups. Order-side evidence also confirms this: the five Taiwan-based notebook ODMs tracked by UBS saw their combined third-quarter shipment forecast revised down to approximately 25.08 million units, a 17% quarter-over-quarter decline and 26% year-over-year decline; full-year shipments of approximately 109.7 million units represent a 15% year-over-year decline. These changes are further compounded by advance replacement triggered by factors such as the end of Windows 10 support and price increase expectations.

Second, the duration of elevated memory prices is being further extended, and continued CPU price increases are adding to overall system cost pressure. UBS expects the DRAM upcycle to continue into the second quarter of 2028 and has pushed the NAND price peak from the fourth quarter of 2027 to the first quarter of 2028. Its model forecasts that blended DDR contract prices will rise 22% and 9% quarter-over-quarter in the third and fourth quarters of 2026, respectively, while NAND will rise 20% and 8%; in 2027, DRAM and NAND per-Gb prices will reach approximately $2.25 and $0.34 respectively, representing year-over-year increases of 39% and 34.6%; UBS's calculation model also expects cumulative price increases of approximately 766% and 476% respectively for the two from 2025 to 2027. Meanwhile, the UBS research report notes that CPU manufacturers are also pushing for double-digit percentage price increases, meaning PC brands face simultaneous price increases across multiple key components while consumer budgets can typically only increase modestly.

Third, consumers still have the willingness to replace devices and purchase AI PCs, but actual purchasing behavior is increasingly constrained by price. UBS Evidence Lab surveyed 1,500 PC users in the United States and China in August 2026, finding that the average replacement cycle shortened from 3.03 years to 2.86 years, and the proportion planning to purchase within the next six months rose to 32%; China's average purchase budget is approximately 7,954 RMB, an increase of about 3% from the previous round, while the U.S. average is approximately $825, essentially flat. However, among respondents who slowed their purchases, 50% cited memory price increases as the most critical reason, up from 37% in the previous round; when facing more expensive configurations, 19% chose to wait for prices to fall, 18% reduced memory capacity, 14% lowered other specifications, and 48% were willing to keep the configuration and accept higher prices. UBS noted that demand structure still has bright spots: 72% are interested in gaming PCs and 69% are interested in AI PCs; among those interested in AI PCs, 79% indicated they might purchase or upgrade, with AI PC budget premiums of approximately 18% and 16% in the U.S. and China respectively. Therefore, what UBS sees is that replacement willingness remains, but consumers are constraining spending through delays, spec reductions, and brand adjustments.

Fourth, the impact of memory price increases on overall system costs has been sufficient to change product configurations and pricing approaches. Using a reference computer configured with 32GB DDR5 and 1TB SSD as an example, UBS estimates that memory costs rose from $160 to $770, and total component costs rose from $506 to $1,116; if the configuration and absolute profits at each stage are maintained, the system retail price would need to increase from $640 to $1,250, a rise of approximately 95%. In the same model, ODM per-unit profit is maintained at $10, but the cost base has expanded, causing the model profit margin to drop from 2.0% to 0.9%, so a margin decline does not necessarily equate to a simultaneous decline in per-unit profit. UBS emphasized that fixed-budget consumers will increasingly face choices of reducing memory, shrinking SSD capacity, or lowering other specifications.

Fifth, brand competition is shifting toward pricing power, product mix, and procurement execution, with Apple and Lenovo demonstrating different response paths. UBS noted that Lenovo and ASUS maintained relatively stable sales and profit performance through price increases, increasing the proportion of high-end products, digesting previously procured low-cost inventory, and capturing advance purchase demand; UBS stated that Apple expanded share through Mac product updates and sales performance of the lower-priced Mac Neo: UBS's brand table shows that Apple PC shipments grew 20% year-over-year in the second quarter of 2026, with share rising to 11.3%, up approximately 2.3 percentage points year-over-year. Additionally, long-term industry concentration is also increasing, with smaller brands' supply assurance and cost absorption capabilities more easily tested. UBS's review of historical cycles further illustrates that in the early stage of memory price increases, price hikes may support brand revenue, but margin performance diverges; if volume adjustments deepen later, earnings pressure tends to become more apparent and may drive a batch of small brands toward extinction.

Sixth, global hardware-related sector investment opportunities are mainly concentrated in AI computing cluster business increments and AI infrastructure component suppliers such as MLCC, copper foil, and electronic cloth whose unit product value continues to rise, with valuations already showing clear divergence. The hardware sector covered by UBS trades at approximately 18 times forward P/E overall, ODMs at approximately 13 times, brand companies at approximately 10 times, and component companies at approximately 33 times, indicating that capital has already been more actively pricing the growth potential of AI components. Therefore, UBS's latest selection criteria for the global hardware technology chain include exposure to AI server volume ramp, value increases from specification upgrades, and cost pass-through capability, corresponding to companies such as Arista, Cisco, Hon Hai, Quanta, Wistron, Wiwynn, as well as Delta, Unimicron, Chaozer, and BizLink. The UBS research report also notes that notebook display-related components may still be dragged down by excessive shipments and channel inventory in the first half of 2026, with pressure extending into the second half of 2026 and 2027. UBS stated that within the same global hardware supply chain, new orders obtained by AI servers, network infrastructure, and key AI infrastructure-related components, versus the demand adjustments borne by ordinary PC supply chains, are forming different profitability trajectories.

Muse and Astra Further Expand Computing Boundaries: Why the More Pervasive AI Becomes, the More Consumer Electronics Costs Come Under Pressure

The key change brought by Muse and Astra is that a single user instruction can initiate continuous, multi-step computational work. Meta has explicitly disclosed that Muse runs on a dedicated cloud virtual machine equipped with a browser, saves data needed for tasks, and can continue working after the user closes the application; OpenAI positions Astra as a model capable of completing computer operations, browser tasks, software development, and multi-step professional workflows. This means that beyond model inference, it is also necessary to run browsers, code sandboxes, task scheduling, databases, and file services: GPUs, TPUs, and other accelerators handle neural network computation, while AMD, Intel, and Arm architecture CPUs handle tool execution and system services; long context and concurrent requests increase KV cache and HBM demand, virtual machines increase server DRAM demand, persistent memory and file processing increase storage access, and cross-node switching further drives networking, optical interconnect, and power supply. OpenAI disclosed that the Habitat online storage platform has already processed over 70 million requests per second and serves over 500PB of data, providing an engineering example for this transmission chain of "model capability expansion—actual task increase—complete computing system scaling." Total resource demand depends on user numbers, task frequency, and per-task consumption; after efficiency improvements lower task costs, if application scale expands faster, total infrastructure demand will still grow. This growth does indeed constitute a severe cost shock for price-sensitive, lower-margin consumer electronics products; its essence is the divergence in payment capacity and return-on-investment expectations among different customers under limited supply.

From the manufacturing side, HBM and ordinary DDR both belong to the DRAM system and compete for front-end wafer capacity and investment resources; Micron has pointed out that at the same node and equivalent bit output, HBM3E requires approximately three times the wafer resources of DDR5, and different products' processes, packaging, and certification mean capacity cannot be switched instantly. NAND is affected along the path of enterprise SSD demand growth and manufacturer capacity allocation. From an economic logic perspective, cloud providers can evaluate higher procurement costs based on computing power leasing and AI service revenue, while household PC purchase budgets have relatively limited room for expansion, making price increases more likely to translate into delayed replacement or reduced configurations. The divergence between servers and clients already has financial evidence: Intel's server product sales grew 9% year-over-year in the second quarter of 2026, with average selling prices rising 48%, where the price increase mainly came from a higher proportion of high-end products, reflecting demand for high-performance computing and product mix upgrades. This is also why UBS emphasizes that a more accurate judgment for consumer electronics is cost squeeze and profit redistribution: companies with stronger supply assurance, brand premium, and AI product capabilities can still compete for share, while manufacturers lacking these capabilities are more likely to simultaneously bear cost, volume, and profit pressure. This is precisely the underlying logic behind UBS's downward revision of PC forecasts while emphasizing continued preference for hardware suppliers associated with the massive expansion of AI computing cluster demand.