Market and Technology Trends
Computing and AI for Data Centers 2022
By Yole Intelligence —
Data center processor diversification is ongoing with AI co-processing and DPU acceleration. Myriad players want a slice of the expected $106B 2027 market.
Achronix, Advanced Micro Devices (AMD), Alibaba Group, Ampere Computing, Apple, Akamai, Arm, Atos, Amazon Web Services (AWS), Baidu, Biren Technology, Cambricon Technologies, Cerebras, Cisco, Compal, Corigine, Corerain Technologies, Cray, Dayu, Dell, Deezer, Digital Reality, Dell, Denglin, Electronic Arts, Enflame, Equinix, Esperanto Technologies, Fujitsu, Fungible, Global Foxcom, GlobalFoundries (GF), Google, Graphcore, Groq, HiSilicon, Hitachi, Hewlett Packard Enterprise (HPE), Hjmicro, Huawei, Hygon, IBM, Intel, Inventec, Inspur, Jaguar, Kalray, Knuedge, Lattice Semiconductor, Lenovo, Lightelligence, Lightmatter, LiSuan Tech, Loongson Technology, Luminous computing, Lynxi, Marvell technology, Mellanox, Meta, Microchip, Micron, Microsoft Azure, Moffett AI, NebulaMatrix, NEC, NetApp, Netflix, Neuroblade, NTT Communications, Nuvia, Nvidia, Oracle, Optalysys, Penguin Computing, Pensando, Pezy Computing, Phytium, Qilingxin, Quanta Cloud Technology (QCT), Quality Technology Services (QTS), Quanta Computer, Rackspace, Resnics, SambaNova, Samsung, Sabey Data Centers, SAP, Semiconductor Manufacturing International Corporation (SMIC), SiPearl, SK Hynix, Socionext, SocNoc, Sony, Spotify, Sugon, SuperMicro, Tachyum, T-Head, Taiwan Semiconductor Manufacturing Company (TSMC), Tencent, Tenstorrent, Tesla, Tower Semiconductor, Twitter, United Microelectronics Corporation (UMC), Upmem, Untether AI, Xilinx, Wave Computing, Weibo, Yunsilicon, Yusur, and more
AI accelerators and SmartNICs/DPUs are pushing data center transformation
The growth of artificial intelligence (AI) is having a strong impact on server processor technology. First, extensions enabling accelerated AI inferences are being added to Central Processing Units (CPUs). Then, the growth of AI is strongly pushing the attachment rate of AI co-processors. The first devices to benefit from this push are Graphics Processing Units (GPUs), which are massively used to train AI technologies. AI Application Specific Integrated Circuit (ASIC) co-processors are also growing, both for inferences and training. In parallel, the rise of networking acceleration with SmartNIC and DPU is happening, mainly in hyperscale datacenters. With the growth of virtualization, ever more CPU cores are dedicated to networking and data management tasks. Adding a networking co-processor such as Smart Network Interface Cards (SmartNICs) and Data Processing Units (DPUs) offloads heavy networking tasks from CPUs. This is the start of a trend called datacenter disaggregation, which is putting these new DPU processors at the heart of data center networks to handle networking tasks, as well as storage and infrastructure management tasks. This push for a new type of processing is joined by a need for ever more servers to support the many growing applications. Together, these trends strongly drive revenue from processors for data centers, which will reach around $106B in 2027, at a Compound Annual Growth Rate (CAGR21-27) of 22%.
Arm-based processors, advanced packaging and memory innovation are driving the development of new processors
If x86-based server CPUs are still largely dominating the market, the diversity of other products available is growing. Ever more Arm-based solutions are on the market, with products from companies such as Nvidia, Ampere Computing, Amazon Web Services, and Alibaba. Meanwhile, with the decreasing pace of Moore’s law, advanced packaging and new technologies are essential to increase processor performance. This is leading to many innovations, such as Through Silicon Vias (TSVs), chiplets, and 2.5D and 3D packaging. They are used to bring memory closer to the processor with the stacking of High Bandwidth Memory (HBM) on the interposer. In parallel to this phenomenon, many innovations are trying to gather computing and memory in another way. They bring computing inside memory units, with computational storage and processing in-memory solutions for example.
A dynamic ecosystem: large fabless companies, cloud giants and AI start-ups
Three main types of players are developing processors for data centers. Large fabless companies are the first category, with Intel, Nvidia, and AMD owning most of the CPU and GPU market today. Each of these companies is currently consolidating its solution portfolio through internal development and multi-billion acquisition to offer a comprehensive data center offer. Then the cloud companies are playing a growing role, from big processor customers to processor designers. Many of them have started to design their own processors for their internal needs. The third category is startups dedicated to creating AI accelerators. In the last decade a high number have been created and they received significant funding. The Chinese ecosystem is particularly dynamic, pushed by several Chinese policies that aim at stimulating the development of the semiconductor industry, with a specific focus on AI.
Available on our Yole Group All-Inclusive Computing Package
Three-page summary
Executive summary
Context
Market forecasts
Market trends
- Cloud computing
- Video streaming
- Social media and the metaverse
- AI and Heavy computing
- Market trends conclusion
Market shares & supply chain
- Mainland China focus
Technology trends
- CPU
- GPU
- FPGAs and accelerators
- AI acceleration
- Networking acceleration: SmartNIC and DPU
- Advanced packaging
- Memory
- Technology Trends Conclusion
Conclusion
Appendix – Processor definition
Appendix – Definitions
Appendix – Artificial Intelligence
Appendix Netflix case study
Appendix Yole Group presentation
Key Features
- Data center application trends and impact on processor market
- Market forecast for processor by unit, revenue, and ASP
- Technology breakdown: CPU, GPU, AI accelerators and smartNIC/DPU
- Market breakdown: Hyperscale, enterprise, HPC
- Technology segmentation and associated trends
- Technology trends analysis for the main class of processors
- Arm-based CPU evolution
- GPU and AI accelerator landscape and technology trends
- Networking acceleration with smartNIC and DPU, in the context of datacenter disaggregation
- Market trends and new application analysis
- Ecosystem description from foundries to OEMs
- M&A, strategy analysis and competition landscape of key players
What's new
- Networking acceleration with SmartNIC and DPU
- Breakdown by market segment for hyperscale, enterprise and HPC
- Focus on advanced packaging innovation
- Focus on memory technologies
- Focus on the Chinese ecosystem
Product objectives
Provide a scenario for computing within the dynamics of the datacenter market, and present an understanding of AI’s impact on the semiconductor industry:
- Processor - revenue forecast, volume forecast, technology breakdown, market segment breakdown
- Focus on artificial intelligence acceleration
- Focus on networking acceleration
Deliver an in-depth understanding of the ecosystem and its players:
- Who are the competitors? What are the relationships inside this ecosystem?
- Who are the key suppliers to watch, and what technologies do they provide?
- Start-up funding and M&A
- What is happening in China?
Offer key technical insights and analyses into future technology trends and challenges:
- Key technology choices and technology dynamics
- Advanced packaging and memory technology innovation
- Emerging technologies and roadmaps
Available on our Yole Group All-Inclusive Computing Package