Thursday, May 17, 2018

IDC: Inspur Leads China’s 2017 AI Infrastructure Growth as AI Market Continues Rise

IDC: 2017 saw 235% growth of AI Infrastructure Market in China, with Inspur taking No.1 position at 57% market share.

IDC just released the 2017 China AI Infrastructure Market Survey Report. The report shows that in 2017, AI servers in China delivered total revenue of US$568 million and total shipment of 27,863 units. In terms of the competitive landscape, Inspur’s share is clearly ahead of other manufacturers, ranking first in the market. With shipments reaching 14,438 units, and sales reaching 2.03 billion RMB, which accounted for 52.7% and 57% respectively, Inspur exceeded the sum of all other manufacturers combined.

AI still at the core of technology, GPU server share will reach 16.6% in the next 5 years

According to the IDC report, AI is “still in the middle of a typhoon”, with 2017 total AI market investment reaching 8.64 billion U.S. dollars, an increase of 240% YOY. CAGR in the next five years is estimated to exceed 50%. Total AI market includes three segments: hardware, software and services.

The hardware systems are crucial to the development of AI; they are the main component of the AI market. In 2017, AI hardware revenue increased by 235% YOY, accounting for 72.4% of the total AI market. It is expected that the total AI market revenue will reach 37.7 billion US dollars by 2022, with the GPU server revenue reaching US$34 billion. AI server accounts for 16.6% of the overall server market.

AI enters new era, Internet is the main driving force

CSPs are still the main force of AI development, with US$4.06 billion in revenue, accounting for 71.9% of the overall market.

2017 was the turning point for AI development. Previous AI infrastructure invested and built by major CSPs were primarily to support internal business operations such as behavior analysis, searching, and ad click-through forecast. Starting in 2017, almost all CSPs began to increase AI investment and accelerate AI infrastructure development, with some hyperscalers starting to provide AIaaS (AI as a Cloud Service).

According to IDC data, Inspur’s market advantage comes from the internet industry which has 68% market share. Over 90% of the AI servers used by China’s top tier CSPs including Baidu, Alibaba, and Tencent, are provided by Inspur. In addition, Tier2/3 CSPs and end users such as toutiao, iFlYTEK, NetEase, PINGAN have been purchasing in bulk various Inspur products and solutions of AI infrastructure systems.

4GPU and 8GPU become mainstream in shipment

The use of applications is driving the fast change of AI infrastructure. Take the GPU server as an example. Due to the GPU cross-node communications bottleneck, the number of GPUs supported by the server is increasing. According to IDC, in 2017, 2GPU servers dropped to 18.8% in mix, and 4GPU and 8GPU server have become the mainstream in shipment, at 49% and 26.8% in sales revenue respectively.

Inspur is one of the server vendors who has the most comprehensive GPU server product portfolio. In 2017, Inspur released new AI products such as SR-AI Rack Scale server, AGX-2 super-computing server, and GX4 emerging AI accelerated expansion box. The SR-AI is the world’s first AI solution using the PCIe Fabric interconnect architecture and can implement large scalable nodes supporting 16 GPUs in one box. The solution supports up to 64 GPUs and solves the problems of power supply and heat dissipation. The product complies with Scorpio 2.5 standard and is suitable for deployment of ultra-hyper-scale cloud data centers. AGX-2 is the world’s first high-speed interconnected 8-bit high-performance GPU accelerator server in 2U space. It is currently the highest performance density AI infrastructure solution and is suitable for various data center environments.

Traditional industries can’t be ignored

Although CSPs are currently the main force to drive AI development and applications, the traditional industries will provide even bigger opportunities for AI in the future, among which financials, retail, medical and healthcare should see faster growth in AI applications. The primary constraint for AI in traditional industries today comes from the technical capability gap with the end users, in particular the deployment and maintenance of full-stack AI solutions.

While sustaining the strength of Internet, Inspur has build a full stack of AI system platforms at four layers to deliver total solutions, including compute platform (e.g., GPU, FPGA AI servers), management kit (e.g., AIStation for deep learning and training cluster management), framework optimization (e.g., Teye for AI performance tuning tool), and application acceleration (e.g., Caffe-MPI for open source deep learning framework Cluster Edition).

Peter Peng, vice president of Inspur Group, stated that Inspur is building an AI ecosystem that will penetrate deeply in industry’s applications, and, by working with partners, Inspur will provide end-to-end AI total solutions to end users and promote AI to much broader usage in commercial applications.

New era shapes the new structure

Peter Peng said that the changes in cloud computing at the data center level and the AI changes at the application level will drive innovation and reconstruction in the infrastructure industry. The rapid growth of Inspur's business is driven by four forces: adoption of the innovative JDM model, smart manufacturing facility in production, promoting open compute technologies, and early readiness for the AI market. Going deep in AI will remain to be one of the core strategies for Inspur in 2018.

In the IFP18 (Inspur Partner Forum for Cloud Datacenter) that just ended, Inspur announced a new AI brand called TensorServer, and launched the T Plan at the same time. Inspur will leverage the new brand and the new plan to drive growth opportunities in the new era, achieving the goal of becoming the top computing solution provider worldwide.