Artificial intelligence (AI) applications are growing in everyday life, and deep-learning chipsets that help to enable AI are on track to grow in shipments, to 2.9 billion in units and US$72.6 billion in value by 2025, according to a Tractica report.
“Hardware is the key to solving AI problems, and chipsets are at the heart of that. There has been tremendous progress in terms of compute and AI acceleration in the past year, and the future continues to look as promising,” says the report.
Smartphones and the edge market will be the primary drivers for the overall market. The sheer volume of mobile phones contributes to the overall deep learning chipset market. Edge markets such as drones and consumer are starting to ramp up on shipments. Other prominent edge markets include automotive,
smart cameras and
robotics.
“Given the widespread applicability of AI, it is almost certain that every chip in the future will have some sort of AI engine embedded in it. The engine could take a wide variety of form factors, including a simple AI library running on a CPU or complicated custom hardware,” Tractica says.
While the deep learning chipset market has experienced a period of evolution in the past two years, chip companies are somewhat behind in their delivery schedules.
“Smaller chips aimed at the edge market are shipping, but larger chips aimed at the enterprise market are seeing delays. The shipping date of enterprise chipsets has been postponed to 2019,” Tractica says.
Meanwhile, market validation has already begun for the
edge market and should begin for the enterprise market in 2019.
Tractica expects that 2019 and 2020 will be the years when shipments of deep learning chips ramp up in volume and when “winners will begin to emerge.”
Chip providers on the market
“In less than a year, more than 70 companies of all sizes have announced some sort of deep learning chipset or intellectual property,” says the report. “Every prominent name in the technology industry has acknowledged the need for hardware acceleration of AI algorithms, and the semiconductor industry has responded by offering a wide range of solutions.”
ARM said that it sees AI as being a key differentiator for future generations of smartphones. Every prominent cloud company, such as Amazon, has announced its intention to design AI chipsets.
A second generation of startups is entering the deep learning chipset market. These companies are using innovative architectures and targeting their chipsets toward solving specific problems or pain points.
The shortage of expertise is still a problem in the AI field. The challenge with chipset design is that it requires understanding of AI, hardware and software design, making it hard to find the needed talent.
“The demand for expertise in
computer vision or AI continues to increase, while the number of PhDs has remained more or less the same,” Tractica says.
The software support required for chipsets is still unclear and offers limited opportunities for monetization. Some big names, such as Tesla and Google, are making chipsets a part of their complete solution and designing the chipset simultaneously with middleware, IP and application software.
Overall, Tractica remains bullish on the prospects for the deep learning chipset market. “Interesting applications will likely continue to emerge as the chipset technology starts to mature and starts to aid more real-life applications,” Tractica says, adding that 2019 to 2020 will serve as redemption years for the industry.