AI infrastructure investment drives a 116% surge in data center semiconductor revenue: In-depth analysis of the industry chain and competitive landscape

Event Overview

On June 17, 2026, market research firm Dell'Oro Group released its "Data Center IT Semiconductors and Components Quarterly Report," showing that global data center IT semiconductor and component (server and storage system) revenue grew 116% year-over-year in the first quarter of 2026, hitting a record high. The report noted that the continued expansion of AI infrastructure and rising memory prices were the main drivers.

Why It Matters

This growth rate far exceeds traditional server market cycles, signaling a fundamental shift in semiconductor demand structure. AI training and inference not only drive demand for GPUs/accelerators but also trigger a synchronized surge in peripheral components such as HBM, high-speed networking chips, and large-capacity storage. For semiconductor value chain segments like wafer foundry, packaging, and equipment materials, this trend will reshape investment directions for the next five years.

In-Depth Analysis

Technology Impact

The core drivers of data center IT semiconductor market growth come from AI accelerators (including general-purpose NVIDIA GPUs and cloud providers' custom accelerators like Google TPU and Amazon Trainium) as well as high-bandwidth memory (HBM). In Q1 2026, HBM experienced rapid iteration across three generations (HBM2E, HBM3, HBM3E), with continuous increases in per-chip capacity and bandwidth, driving up average DRAM selling prices. Additionally, demand for high-speed back-end NICs surged due to inter-node communication in AI clusters. On the storage side, AI pre-training/post-training datasets require large volumes of NVMe SSDs, boosting NAND flash shipment volumes.

• Technology bottlenecks include:
• HBM is heterogeneously integrated with GPUs through TSV (through-silicon vias), making advanced packaging capabilities a key constraint.
• High-speed SerDes IP and optical interconnect technologies (e.g., Coherent Optics) determine the scaling efficiency of AI clusters.

Supply Chain Impact

• Upstream: DRAM and NAND prices were on an upward trajectory in Q1 2026, benefiting Samsung, SK Hynix, and Micron the most. HBM supply remains tight, and TSMC’s CoWoS capacity is running at full utilization.
• Midstream: Server ODMs (e.g., Quanta, Inventec) need to adjust production lines to accommodate higher-power GPUs and liquid cooling solutions. Networking chip vendors (Broadcom, Marvell) see surging orders for high-end switch chips and NICs.
• Downstream: Cloud service providers (AWS, Microsoft Azure, Google Cloud) continue to expand capital expenditures, increasing the proportion of self-developed chips, putting pressure on traditional x86 CPU suppliers (Intel, AMD).Risk points: Insufficient HBM and advanced packaging capacity may lead to supply gaps in the second half of 2026; geopolitical tensions could affect NAND production at Samsung and SK Hynix's factories in China.

Competitive Landscape

NVIDIA maintains its dominant position in the GPU market with its Hopper and Blackwell architectures, but the AMD MI300 series and Intel Gaudi series are catching up. In the accelerator space, the share of cloud vendors' self-developed chips (AWS Trainium2, Google TPU v5, Microsoft Maia) rose from less than 10% in 2025 to about 15% in Q1 2026. Custom chip design companies (such as Broadcom, Marvell) benefit from collaboration with cloud vendors.

Memory market: Samsung and SK Hynix maintain a stable duopoly in the HBM field; Micron, though lagging, is accelerating mass production of HBM3E. On the NAND front, Solidigm (a subsidiary of SK Hynix) has secured AI storage orders for QLC enterprise SSDs.

Regional Implications

• United States: NVIDIA, AMD, Broadcom, etc. are headquartered in the US, but manufacturing relies on TSMC and Samsung. The US CHIPS Act subsidies are accelerating the construction of domestic advanced packaging capacity.
• Taiwan, China: TSMC's CoWoS capacity expansion supports global AI chip shipments, but faces challenges of high power consumption and heat dissipation. Design companies like MediaTek and Realtek are also beginning to layout data center ASICs.
• South Korea: Samsung and SK Hynix are the main suppliers of HBM and NAND, and the high profit margins of HBM have driven a significant increase in South Korea's semiconductor exports.
• Mainland China: Domestic AI chips (Huawei Ascend, Cambricon) are still catching up due to restrictions on manufacturing below 7nm. Data center storage reliance on domestic NAND (YMTC) is increasing.
• Japan: Tokyo Electron and Disco benefit in the equipment segment; Rapidus is targeting 2nm, but mass production still needs time.
• Europe: Infineon and NXP are deploying in data center power devices (GaN/SiC), but their overall share is small.

Investment PerspectiveThe capital market continues to pursue the AI infrastructure theme. Dell'Oro expects full-year 2026 data center IT semiconductor revenue to maintain triple-digit growth. Investor focus points: - Whether NVIDIA’s data center business revenue growth is slowing (Q1 2026 earnings show it still exceeds 100%). - HBM price trends: If HBM4 enters mass production in 2026, prices may rise further; if there is overcapacity, gross margins will be under pressure. - Cloud providers' capital expenditure guidance: Microsoft, Google, and Amazon continue to raise AI investment in 2026, supporting demand.

Long-Term Outlook

• 3 years: AI accelerator market CAGR ~40%, GPU still dominates, but ASIC share will rise to 30%. Advanced packaging (CoWoS, 3D SoIC) capacity doubles.
• 5 years: HBM4 and HBM4E launched, memory bandwidth exceeds 2TB/s. Optical interconnects may partially replace electrical interconnects.
• 10 years: AI inference demand may surpass training, edge data center deployment drives demand for small and medium-sized chips. Early commercialization of quantum computing may change some computing workloads but will not impact the mainstream.

Industry Chain Analysis

Upstream: Semiconductor Equipment and Materials - Equipment vendors (ASML, Applied Materials, Lam Research) benefit from logic and memory fab expansion, especially HBM TSV etching equipment. - Material suppliers (Shin-Etsu, SUMCO, Dow) have stable silicon wafer shipments, but demand for photoresists and precursors is growing.

Midstream: Chip Design and Manufacturing - Fabless: NVIDIA, AMD, Broadcom, and cloud providers' in-house teams. AI chip design complexity increases, EDA tool (Synopsys, Cadence) revenue grows in tandem. - Foundry: TSMC's 3nm/5nm capacity is fully loaded, 7nm/6nm demand comes from AI inference chips. Samsung's 3nm GAA yield needs improvement. Intel Foundry seeks external customers.

Downstream: System Integration and Cloud Services - Server manufacturers: Dell, HPE, Lenovo, etc. ship AI servers, ODM direct sales proportion expands. - Cloud providers: dual strategy of self-developed chips + purchasing NVIDIA to reduce supply chain risk.

ConclusionThe 116% surge in data center IT semiconductor revenue in Q1 2026 is not a short-term phenomenon, but a sign that the wave of AI infrastructure investment is entering an acceleration phase. Various links in the industry chain will undergo structural adjustments: HBM and advanced packaging become bottleneck technologies, custom accelerators erode the market share of general-purpose GPUs, and cloud vendors delve into chip design, changing the traditional Fabless-IDM landscape. In the next five years, the data center semiconductor market will shift from a "single GPU engine" to a multi-engine drive of "GPU + custom + storage," placing higher demands on supply chain resilience and technological iteration.