The market for AI processors at the "edge" was worth US$10.6 billion in 2018 and is expected to increase by about 74 percent in 2019 to $18.4 billion. Thereafter it will grow with a compound annual growth rate (CAGR) of 31 percent over the period 2019 to 2024, the market research firm said, taking it to a market worth about $71 billion in 2024.
In its second report ABI Research puts the size of the market for cloud AI processors at $3.5 billion in 2018 with a compound annual growth rate over the period 2018 to 2024 of 33 percent. This will take the market to $4.6 billion in 2019 and $19.1 billion in 2024.
Edge environments include end devices such as smartphones and IoT nodes, gateway products such as routers and on-premises servers. As a result, an edge AI processor is generally designed for inferencing using neural networks although some level of AI training may also be supported.
ABI points out that as well as there being scores of proprietary vendor-specific IP cores and processor architectures there also open-source chipset startups developing AI extensions based on RISC-V.
At present the cloud AI processor market is served by non-captive manufacturers such as Nvidia, supplying GPUs adapted for AI training loads and Intel CPUs. But cloud service provides such as Google, Facebook, Huawei, Alibaba, Baidu, Amazon are active in this area and apparently feel they can produce custom silicon that will address their specific needs and trade-offs between performance and power consumption.
ABI Research reckons the captive AI cloud processors captured just 2.3 percent of the market in 2018 but will grow as a market to be responsible for 9.4 percent in 2024. ABI mentions that Cerebras Systems, Graphcore, Habana Labs, and Wave Computing as having announced chipsets.
Related links and articles: