Diversity is the keyword people talk about the edge of artificial intelligence (AI) chipset industry. In 2019, the ai industry has witnessed a continuous migration of ai workloads, especially ai extrapolations, to edge devices, including premise servers, gateways, terminal devices, and sensors. Based on AI's growth in 17 vertical markets, ABI Research expects the edge AI chipset market to grow from $2.6 billion in 2019 to $7.6 billion in 2024, with no single vendor taking more than 40% of the market.

The market leader is NVIDIA, with 39% of revenue in the first half of 2019.GPU vendors have a strong presence in key ai verticals and are currently leading in AI deployments, such as cars, camera systems, robots, and intelligent manufacturing."Faced with different use cases, NVIDIA chose to release GPU chipsets with different computing and power budgets."Lian Jye Su, chief analyst at ABI Research, said, "with its large developer ecosystem and partnerships with academic and research institutions, the chipset supplier has established a strong foothold in the edge ai industry."

NVIDIA is facing stiff competition from Intel, which has a full portfolio of chipsets, from Xeon CPUs to Mobileye and Movidius Myriad. Meanwhile, FPGA vendors, such as Xilinx, QuickLogic, and Lattice semiconductor companies, are creating compelling solutions for industrial ai applications. One vertical missing from NVIDIA's broad footprint is consumer electronics, especially smartphones. In recent years, ai processing in smartphones has been driven by smartphone chipset makers and smartphone makers such as Qualcomm, Huawei and Apple. In smart home applications, MTK and AmLogic make their presence known through the widespread use of voice-controlled frontend and smart devices.

Looking ahead, AI chipset vendors are likely to adopt one of three strategies. The first is to create AI chipsets with the goal of supporting the AI premise server and gateway market. These servers and gateways support enterprise use cases, often with high processing power, and require a flexible AI chipset architecture to support changing AI reasoning and training workloads. This is an area well served by NVIDIA, Intel and Xilinx, but new entrants such as Huawei, Graphcore and Habana LABS could challenge the status quo.

The second strategy is to target smart edge devices and nodes, which to some extent benefit vendors active in consumer electronics.Chipset vendors like qualcomm and MTK have a natural advantage here.Manufacturers that design their own ai chipsets, such as apple, Huawei and Samsung, are also beginning to expand their ai-enabled product portfolios for consumer devices.

The final strategy is to target low-cost, battery-powered terminal devices with minimal computing power and long life.These devices are typically deployed in smart cities, smart buildings, smart transportation, and utilities, all of which rely on public Ethernet or low-power wide area networks (LPWAN) for connectivity.These "very marginal" devices require lighter AI implementations, an approach often referred to as tiny or thin AI.ABI Research predicts that shipments of these devices will increase from 900,000 units in 2019 to 5.7 million units in 2024, with a CAGR of 45.5%.Traditionally, these devices have relied heavily on more powerful resources, such as gateways, prerequisite servers, and public clouds for ai training.Recently, many chipset players have participated in this market by offering AI chipsets with extremely high energy efficiency and low price.These include GreenWave technologies, a start-up that USES open source RISC-v architecture to develop AI chipsets; Lattice semiconductor company, a supplier of FPGA chipsets; SynTIant, inc. is an ASIC supplier specializing in natural language processing.

The superior AI chipset market is a highly competitive market. Use cases are becoming more complex and diverse, and new players appear on the horizon almost every month. The major key players have a strong tradition of building global scale for their chips, even in very fragmented environments. Therefore, vendors, especially newcomers, must have a clear value proposition, a comprehensive software stack, and strong support from the partner ecosystem and developer community.