Market and Technology Trends
Data Center Semiconductor Trends 2026
Supported by strong AI-related demand, the data center semiconductor market is forecast to reach $1.39T by 2031.
YINTR26595- We updated our previous data center segmentation with the following changes:
- We added AI factory column that captures Neoclouds, dedicated hyperscaler AI campuses, sovereign AI sites
- We added colocation with Equinix / Digital Realty / NTT GDC …
- We added Edge / Micro-DC added for telco MEC, CDN POPs, regional inference.
- Most crypto miners have pivoted to AI hosting (CoreWeave, Core Scientific, IREN, Bit Digital, Galaxy, Cipher, Hive), so this category folds into AI factory.
- We have updated data center capex
- We present a ROI model for GenAI
- We have revamped revenues model, especially GPUs and HBMs, that have increased a lot.
- Specific analysis of the possible bottlenecks that are power supply, advance packaging capacities and global fab capacities and the level of scarcity of HBM and price increase and effect on the whole memory supply chain.
- We have updated the photonic market model.
- We added an analog forecast part.
Comprehensive analysis of the data center semiconductors industry:
- Examine the data center semiconductors market, including revenue forecasts, wafer starts (by device, node, substrate, size) 2021-2031.
- Focus on AI, GenAI, while also covering the non-AI data center market.
- Focus on the following semiconductor devices for data center:
- Compute & logic: GPUs, AI ASICs, CPUs, MCUs, FPGAs.
- Memory & storage: HBM, DRAM, NAND.
- Sensors, networking & optical: optical transceivers, optical I/O, co-packaged optics (CPO), PICs, sensors
- Power: power management ICs, GaN & SiC power devices, analog devices
Global understanding of the ecosystem and key players:
- Analysis of the data center supply chain.
- Examine the relationships and interdependencies within the ecosystem.
- Anticipate changes and shifts in the industry for the coming years.
- Devices’ makers supply chain.
- Main trends by geographical zone.
Key technical insights and future technology trends:
- Discuss key technology choices and the dynamics shaping the industry.
- Highlight the key enabling technologies and challenges in logic, memory, photonics, sensing, power.
Supported by strong AI-related demand, the data center semiconductor market is forecast to reach $1.39T by 2031
The semiconductor for data center market is growing rapidly, driven by the increasing demand for generative AI applications that require high-performance computing, HBMs, optical interconnects, networking and power devices.
Data center semiconductor revenue more than doubles in 2026, to ~$732B, and reaches ~$1.53T by 2031, growing faster than DC capex itself. Logic (GPU, AI ASIC, MCU, FPGA) remains the largest segment. GPUs keep ~72% of APU revenue in 2031 and AI ASICs grow fastest. In 2025, the logic market was $198B.
Memory is the 2026 swing factor and fastest-growing segment, jumping ~3.7x from 2025 to 2026 as server DRAM contract prices rise 60–70% in Q1 2026. In 2031, memory market will surpass half-trillion market value.
Power and photonics are critical technologies as well. Power and analog grows 4x in2031 on the shift to 800V HVDC for ~1MW racks. Opto and Sensors grows significantly driven by photonics as co-packaged optics ships in 2026. Interconnect becomes as critical to AI performance as compute itself.
AI is reshaping the entire data center supply chain, from chip design to rack architecture.
Nvidia continued to dominate the server GPU market in 2024 and 2025.
The AI ASIC market is divided between companies developing processors for their own use (Google, AWS, Microsoft, META, Baidu, etc.) and those offering their solutions for sale (Huawei, Intel Gaudi, etc.).
Samsung is first in DRAM and an HBM challenger. It got HBM4 validated by both Nvidia and AMD with mass production from February 2026.
The AI data center is a chess game characterized by:
- AI chip designers are challenging Nvidia and AMD. These companies are building their own AI chips to break up the Nvidia/AMD dominance.
- Big tech companies are co-developing their own AI ASICs. Companies like Google and AWS don't design chips alone, they partner with specialist firms (Broadcom, Marvell, Alchip) to co-build custom AI processors.
- Startups are exploring new designs, such as chips without HBM memory or wafer-scale architectures.
The whole AI infrastructure ecosystem is scaling fast. Cloud giants (AWS, Google, Microsoft, Oracle) are expanding at massive scale, while companies like Meta, Tesla, and Apple are building in-house silicon. AI labs like OpenAI, Anthropic, and xAI are investing in multi-gigawatt data center capacity.
Because power and photonics now gate data-centre growth, they will also redraw the supply chain, pulling value and bargaining power toward the power-device and optical suppliers : SiC/GaN, power-management ICs, transceivers and co-packaged optics able to scale fast enough to feed it.
AI moved the bottlenecks to the whole rack, and advanced packaging, memory, photonics and power set the direction
A big shift is that the whole rack is becoming the core element of data center. AI performance is now sold and measured at the full rack level, not chip by chip. Systems like Nvidia NVL576, AMD Helios, and Google TPU pod show this shift. Margins are moving from individual chips to complete integrated rack systems.
The compute level is shifting from the chip to the rack: scale-up fabrics (NVLink, UALink, SUE), HBM stacks and advanced packaging (CoWoS-L, SoIC, hybrid bonding) that determine system performance.
Generative and agentic AI workloads are reshaping data centers at every level. Demand is currently concentrated on a small number of hyperscalers (Microsoft, Google, Meta, Amazon, Oracle, Alibaba, Huawei, Baidu) who are simultaneously buying merchant GPUs and co-designing their own AI ASICs with players like Broadcom, Marvell, Alchip and others.
Fiber optics are replacing copper inside data centers: as bandwidth demands and distances grow, copper connections are being displaced by optical technologies: silicon photonics transceivers, optical circuit switching (OCS) fabrics, and co-packaged optics.
Power is now the glass ceiling with AI racks now approaching 1 MW each at the end of the decade, requiring advanced power delivery (800 V systems, SiC/GaN semiconductors) and liquid cooling. The real limit on deploying more AI infrastructure is not only chips, but also access to grid power.
3-page summary
- Semiconductors for data center – market forecast
- Semiconductors for data center – supply chain overview
- Semiconductors for data center – technology trends
Executive summary
Context
- Overview of the semiconductor device industry
- Definitions
- Data center types
- GenAI market segmentation
- Foundation Model APIs
- Cloud AI Platforms
- AI Copilots & Enterprise
- AI creative and media
- AI coding assistants
- AI search engine
- The AI demand is reshaping semiconductor allocation
Market forecasts
- Data center capex
- Data center capex
- Hyperscalers capex
- Hyperscalers AI capex
- AI revenue forecast
- GenAI ROI model
- Devices forecast in value
- Semiconductor device industry: 30-year perspective
- Semiconductors for data center forecast
- APU forecast
- Memory forecast
- Power & analog forecast
- Photonics & sensors forecast
- Semiconductors for data center
Devices forecast in units
- Wafers forecast
- Data center wafers forecast
- Data center wafers forecast by device
- Data center wafers forecast by node
- Data center wafers forecast by substrate
- Data center forecast by wafer size
- Take-aways
Market trends
Supply chain & market shares
- Data center supply chain
- Data center value chain
- Data center ecosystem
- Server and storage companies
- Companies offering AI solutions
- Hyperscaler vertical integration
Europe
- Europe
USA / Nvidia
- USA
- USA: NVIDIA
China
- Data center in China
- Mainland China's data center ecosystem
- Data center value chain in China
- Mainland China's supply chain
- China dependency
- Mainland China's ecosystem
- Key takeaways
Market shares
- Server manufacturers' market shares
- GPU market shares
- AI ASIC market shares
- CPU market shares
- MCU market shares
- FPGA market shares
- DRAM market shares
- Optical transceivers market shares
Technology trends
- Technology trends (computing, memory, photonics, power)
- Data center limitations & expansions
- Bottlenecks to data-center & AI expansion
- Critical issue 1: Advanced packaging
- Critical issue 2: Power
- Critical issue 3: Memory
Emerging trends: Quantum & Data Centers in Space
- Investments in quantum computing
- Logical qubits roadmap
- The different types of quantum computers
- Is the future of AI data centers in space?
Outlook
Related products
ABB, Accelink, Advanced Energy, Alibaba, Amazon, AMD, Amkor, Ampere Computing, Amphenol, Analog Devices, Anthropic, AOI, Apple, Arista, Arm, ASE, ASML, Astera Labs, Atos, Ayar Labs, Baidu, Broadcom, ByteDance, Cadence, Cambricon, Celestial AI, Celestica, Cerebras, Ciena, Cisco, Coherent, CoreWeave, CXMT, CyrusOne, d-Matrix, Dell, Delta Electronics, Digital Realty, Eaton, Enflame, Engie, Eoptolink, Equinix, Esera, Etched, Foxconn, Fuji Electric, Fujitsu, Gigabyte, GlobalFoundries, Google, Groq, Grundfos, GUC, H3C, Heijmans, Hikvision, HiSilicon, Hitachi, Honeywell, HPE, Hua Hong, Huawei, Hygon, IBM, Iluvatar, Infineon, InnoLight, Inspur, Intel, Interxion, Inventec, JCET, Juniper, Kalray, Kioxia, Lambda, Legrand, Lenovo, Ligent, Lightmatter, Lite-On, Lumentum, Luxshare, MacroSAN, Marvell, MediaTek, Mellanox, Meta, MetaX, Microchip, Micron, Microsoft, MiTAC, Mitsubishi Electric, Molex, Moore Threads, Murata, Navitas, Nebius, NEC, NetApp, Nexperia, Nokia, NTT, nVent, Nvidia, NXP, onsemi, OpenAI, Oracle, Pegatron, Penguin Solutions, Powerleader, Pure Storage, QTS, Quanta, Rackspace, Renesas, Rittal, ROHM, Sabey, SambaNova, Samsung, SanDisk, Schneider Electric, Securitas, Sensaphone, SiTime, SK hynix, SMIC, Solum, Source Photonics, SPIE, SPIL, STMicroelectronics, Stulz, Sugon, Sumitomo Electric, Supermicro, Synopsys, Teamsun, Tencent, Tenstorrent, Terahop, Tesla, Texas Instruments, TFME, Toshiba, Transphorm, TSMC, Vertiv, Winbond, Wistron, Wiwynn, Wolfspeed, xAI, ZTE, Zumtobel and much more.
