Covering Disruptive Technology Powering Business in The Digital Age

Home > DTA news > News > 2020 Will be a Banner Year for AI Custom Chipsets and Heterogenous Computing
2020 Will be a Banner Year for AI Custom Chipsets and Heterogenous Computing
December 24, 2019 News


2020 will be an exciting year for the Artificial Intelligence (AI) chipset market. The global tech market advisory firm, ABI Research states that in 2020, more than 1.4 million cloud AI chipsets and 330 million edge AI chipsets are forecasted to be shipped, which generate a total revenue of US $9 billion.

According to the new whitepaper, 54 Technology Trends to Watch in 2020, ABI Research’s analysts have identified 35 trends that will shape the technology view and 19 others that, although it is attracting huge amounts of speculation and commentary, it is likely to happen over the next twelve months. Stuart Carlaw, Chief Research Officer at ABI Research said, “After a tumultuous 2019 that was beset by many challenges, both integral to technology markets and derived from global market dynamics, 2020 looks set to be equally challenging,”

What will happen in 2020:

More custom AI chipsets will be launched:
“We’ve already seen the launch of new custom AI chipsets by both major vendors and new startups alike. From Cerebras Systems’ world’s largest chipset to Alibaba’s custom cloud AI inference chipset, the AI chipset industry has been hugely impacted by the desire to reduce energy consumption, achieve higher performance, and, in the case of China, minimize the influence of Western suppliers in their supply chain,” says Lian Jye Su, AI & Machine Learning Principal Analyst at ABI Research. “2020 will be an exciting year for AI chipsets. Several stealth startups are likely to launch programmable chipsets for data centers, while the emergence of new AI applications in edge devices will give rise to more Application Specific Integrated Circuits (ASICs) dedicated for edge AI inference workloads.”

Heterogeneous computing will emerge as the key to supporting future AI Networks:
Existing Artificial Intelligence (AI) applications and networks are currently serviced by different processing architectures, either that be Field Programmable Gate Array (FPGA), Graphical Processing Units (GPUs), CPUs, Digital Signal Processors (DSPs), or hardware accelerators, each used to its strength depending on the use case addressed. “However, the next generation and AI and Machine Learning (ML) frameworks will be multimodal by their nature and may require heterogeneous computing resources for their operations. The leading players, including Intel, NVIDIA, Xilinx, and Qualcomm will introduce new chipset types topped by hardware accelerators to address the new use cases,” says Su. “Vendors of these chips will move away from offering proprietary software stacks and will start to adopt open Software Development Kits (SDKs) and Application Programming Interface (API) approaches to their tools in order to simplify the technology complexity for their developers and help them focus on building efficient algorithms for the new AI and ML applications.”

What won’t happen in 2020:

Quantum computing:
“Despite claims from Google in achieving quantum supremacy, the tech industry is still far away from the democratization of quantum computing technology,” Su says. “Existing vendors, such as IBM and D-Wave, will continue to enhance its existing quantum computing systems, but the developer community remains small and the benefits brought by these systems will still be limited to selected industries, such as military, national laboratories, and aerospace agencies. Like other nascent processing technologies, such as photonic and neuromorphic chipset, quantum computing systems in their current form still require very stringent operating environment, a lot of maintenance, and custom adjustment, and are definitely not even remotely ready for large-scale commercial deployments,” Su concludes.