The AI sector shows no signs of cooling off, but the technology market in 2026 is unlikely to experience smooth sailing throughout the year. This annual outlook from Morgan Stanley divides the trajectory into two distinct phases: the first half will continue the momentum driven by AI capital expenditure and rising commodity prices; the real challenge arrives in the second half, as increasing costs begin to test demand, with "price elasticity" potentially pushing some end-users away.
According to analysis, Morgan Stanley research analyst Shawn Kim writes in a recent report that the first half of 2026 will extend the dominant theme of AI infrastructure investment from 2025, with memory prices continuing their strong performance. In the second half, the pass-through of rising costs from wafer foundry, packaging and testing, and memory will compress profit margins for consumer electronics and IC design companies. The adoption of edge AI in smartphones and PCs is likely to be delayed due to soaring Bill of Materials (BOM) costs.
The report presents an aggressive forecast for semiconductors: global semiconductor revenue is projected to surge to $1.6 trillion in 2026, with an expected year-on-year growth rate of approximately 96%. If this trend holds, capital is expected to favor segments that can "outpace consensus" growth expectations while simultaneously occupying bottleneck positions—memory, advanced foundry, front-end equipment, packaging & testing, and key materials are all poised for repricing.
A secondary, underlying theme is Agentic AI. As AI evolves from "generation" to "autonomous action," systemic bottlenecks will shift from simply accumulating GPUs towards the coordinated performance of a longer chain involving CPU orchestration, memory, and packaging/substrates. Stock selection, therefore, is less about betting solely on the "hottest AI" names and more akin to a barbell strategy: one end focuses on assets with pricing power and those situated at bottlenecks, while the other allocates space to overlooked companies with strong cash flows and reasonable valuations, serving as a hedge against potential volatility.
**$1.6 Trillion Semiconductor Market: Expansion Beyond GPUs to Memory, Foundry, and Equipment** Leadership in AI-related stocks is broadening from logic chips and commodity memory to a wider segment of the supply chain. The underlying logic is that when the AI investment cycle is perceived as "longer and more structural," market pullbacks can become windows to reassess entry points—provided investments are made in genuine bottlenecks, not just areas of high sentiment. The report maintains an overall neutral stance: valuations in the AI chain are high, necessitating more stringent selection criteria; in contrast, non-AI segments may offer greater potential for valuation repair—but this is contingent on AI leaders not possessing "undiscovered new growth drivers."
**Memory Returns to Center Stage: Record Prices and Supply Constraints, HBM Remains the Tightest Bottleneck** In the memory sector, DRAM prices are anticipated to surpass previous historical highs, and "record-high prices" often correlate with "record-high stock prices," this time backed by more substantial earnings support. The report also highlights an unusual phenomenon: a significant divergence between contract and spot prices, indicating exceptional supply-demand tension. HBM represents a more direct bottleneck. Based on supply chain calculations, the HBM market size is projected to grow from approximately $3 billion in 2023 to about $51 billion in 2026 and $72 billion in 2027. Furthermore, under assumptions regarding capacity, yield, and utilization rates, the "sufficiency rate" of HBM supply is expected to be compressed to an extremely low level of around 2% in 2026. The conclusion is clear: as long as AI inference demand persists, the memory sector's "supply-driven cycle" will be forced to accelerate, leading to more aggressive pricing and capital expenditure. The outlook extends further: if these tight constraints continue, "unprecedented" capital expenditure could emerge around 2028, potentially pushing the next bottleneck further upstream to segments like EUV lithography, which would again benefit the semiconductor equipment chain.
**Bottlenecks Shift Forward: Advanced Foundry, Front-End Equipment, and Packaging & Testing to Capture the "Second Wave"** As AI volumes continue to expand and constraints in memory and packaging intensify, the benefits are expected to evolve from "single-point explosions" to a "rotation of bottlenecks." The report focuses on several key areas within the semiconductor sector:
* **Advanced Process Foundry:** AI demand is expected to support TSMC in maintaining a revenue compound annual growth rate of approximately 20% over the next five years, with potential incremental growth from opportunities in manufacturing AI GPUs for Chinese companies. * **Front-End Equipment (SPE):** While strength in back-end equipment related to AI continues, front-end equipment is anticipated to re-accelerate in the second half of 2026, driven by demand for advanced logic and DRAM. Preference is for companies with high exposure to DRAM and those benefiting from expansion at advanced nodes. * **Packaging & Testing (OSAT):** AI demand is further tightening packaging and testing capacity. While Taiwanese backend suppliers benefit, this also implies that "supply bottlenecks" will become more commonplace.
**Agentic AI: CPUs Become the New Bottleneck, Rewriting the "Hardware War" as a "Collaboration War"** Agentic AI represents a shift from "generation" to "autonomous action." The critical factor becomes not just more powerful computing, but how systems coordinate and allocate resources. The report uses architectural diagrams to emphasize the orchestrating role of CPUs and provides estimates: by 2030, Agentic AI could create an incremental CPU opportunity of $32.5 billion to $60 billion, alongside generating 15-45 Exabytes of new DRAM demand. This will change how the market monitors performance: GPUs remain crucial, but the limiting factor could be any短板 within the "collaborative chain" of CPUs, memory, substrates, equipment, and foundry—and these短板 will rotate. The supply-demand inflection point for ABF substrates is imminent—a state of sustained shortage is projected from 2027 onwards, with tight supply for T-type glass substrates potentially persisting into 2028. The share of AI GPU/ASIC and networking boards within total ABF demand is forecast to rise from approximately 18% in 2020 to over 75% by 2030.
The logic for MLCCs is more direct: the MLCC content per AI server is more than 20 times that of a standard server—approximately $230 for an HGX Hopper versus a staggering $4,635 per rack for a GB200 NVL72 system. Current MLCC inventory sits at historically low levels while demand accelerates. In thermal management, the second quarter of 2026 is expected to set a quarterly shipment record for thermal management companies, with liquid cooling penetration rates continuing to rise; AVC is highlighted as a preferred player.
**The 2026 Watershed Occurs in the Second Half: Cost Inflation Risks "Demand Destruction"** The report directly attributes second-half risks to "tech inflation." Rising costs for wafers, packaging & testing (OSAT), and memory will transmit pressure downstream to OEMs and chip design companies: end-devices may not be able to fully pass on these price increases, leading to squeezed demand and more difficult margin protection. The outlook for "edge AI" is more cautious: computational upgrades for Edge AI in smartphones and PCs could be significantly delayed due to higher input costs. For many segments reliant on consumer electronics shipments and possessing weak pricing power, the latter half of 2026 presents a more challenging environment. The bear case is also clearly outlined: the initial phase often sees over-investment and over-consumption, followed by a more rational approach; "spending without a clear destination" will lead to repeated questioning of returns; meanwhile, proximity to the "power wall" and capital constraints will slow deployment pace. Coupled with the tension between "pricing power from price hikes" and "squeezed downstream profits," industry divergence is expected to become more pronounced. On the hardware side, for AI server demand, shipments of NVIDIA GPU server racks are projected to increase to approximately 75,000 in 2026, significantly higher than the roughly 29,000 estimated for 2025. Correspondingly, the thesis of increasing "content value" for components like ABF substrates, MLCCs, and thermal solutions remains valid, with the report suggesting ABF substrates could re-enter a tight supply state after 2027. However, the keyword for smartphones and PCs becomes "cost." The smartphone industry in 2026 will be burdened by rising component costs, pressuring profitability; PC OEMs/ODMs may also experience multiple quarters of profit margin compression due to memory price increases. In other words, while both fall under "hardware," the beneficiaries are diverging: segments closer to the data center are faring better, while those tied to consumer electronics face greater challenges.
Comments