This research contains complete information of the AI-related processors specifications and capabilities which were produced by the key market players and start-ups.
The comprehensive analysis can aid you in your technology acquisitions or investment decisions related to the fast-growing AI processors market which is predicted to grow from $ 6.9B in 2021 to $ 37.6B in 2026 and may become a new sector of the economy.
Key Highlights
- Most of the DL-related tasks are performed on GPUs and ASICs. The main training workflow will still be bound to the GPUs, but the increased adoption of AI in the consumer and edge segments will shift the ratio towards parity, against the current 80% of the market being dominated by the GPUs.
- The ASICs market has historically been much more varied than the CPU or GPU markets. Where there is a need which cannot be answered by other means - there is an ASIC for it. The market actors with large data centers try to optimize and scale-up their clouds while Edge players look to squeeze every TOP out of every watt. We expect the growth of the ASIC market to be much faster than the GPU’s, with FPGA taking an increased foothold in the area.
- The FPGAs used to be a somewhat exotic part, taking the niche segments of scientific and industrial sectors. The rise of the AI-related demand and market integration allowed for the quick progress in the area and dramatically expanded FPGA’s capabilities.
- We are poised to see a 34% average growth of the edge sector until 2025, as companies strive to reduce the data transfer related latencies between data acquisition devices and data processing centers. About 94% of the companies in the Industrial Internet Of Things (IIoT) and Robotic process automation (RPA) have already declared plans to integrate edge-AI or are already doing it. One of the growth factors in the edge market is the mobile processors. This sector is expected to almost double until 2025, from $13bn in 2020 to $22bn with an average annual growth of 10.7%.
- Neuromorphic chips are clearly in the research and development phase but the promise of ultra-low power consumption puts this kind of attempt in the center of the long- term growth for the industry.
AUTHORS
Pretiosum Ventures:
The fund providing seed and series A venture capital to companies in the fields of AI, Fintech and future of Work/Life
The Lydian Group:
The first fully integrated digital asset group with 10+ portfolio companies under management, spanning several sectors and present in the whole digital asset value chain.
Table of Contents
Samples
LOADING...
Executive Summary
Artificial Intelligence (AI) is almost as old as computers themselves and has a rich history of ups and downs. It carries a promise to liberate humans from their tedious and repetitive work, providing assistance in almost every aspect of human life. Bayesian statistics, the basis for most modern AI, was developed in the 18th and early 19th centuries by Thomas Bayes and Pierre-Simon Laplace. Despite the invention of the artificial neuron by Warren McCulloch and Walter Pitts in 1943, it has been mostly dominated by expert systems, which simply follow pre-set rules and various statistical procedures.
After the main breakthrough at the turn of the century AI started to incorporate more and more artificial neural networks, connected in an ever-growing number of layers, now known as Deep Learning (DL). They can compete and outperform classical ML techniques like clustering but are more flexible and can work with much more complex datasets, including images and audio.
As machine learning entered exponential growth, it expanded into areas usually dominated by high-performance computing - such as protein folding and many-particle interactions. At the same time, our lives become increasingly dependent on its availability and reliability. This poses a number of new technical challenges but at the same time opens a road to novel solutions and technologies, in a similar way as space exploration or fundamental physics does.
More so, the commercial success of AI-enabled systems (autopilots, image processing, speech recognition and translation, to name just a few) ensures that no shortage of funds could hinder this growth. It has clearly become a new industry, if not a sector of the economy, one that is gaining importance with every passing year.
As any industry, it depends on several factors to prosper. Rising consumer demand has led to the consensus of major forecasters on the rapid growth of the sector - around 40% yearly in the near future, so funds shortage is not an issue. Instead, we must concentrate on other requirements for the efficient functioning of the industry.
The three main components are the availability of processing tools, the abundance of raw materials, and the workforce. Raw materials in this case are represented by big data, and there is often more of it than our current systems can make sense of. The workforce also seems to grow sufficiently fast, as ML cements its place in the university curriculum. So the processing tools, as well as the available energy to run them are clear bottlenecks in the exponential growth.
The end of Moore’s extrapolation law due to quantum tunnelling and such, which become increasingly important with the reduction in transistor size, sets clear bounds on where we can go. To ensure long-term investments in the industry, a clear strategy must be developed to offset what will happen in 10 years
Companies Mentioned
- Alibaba Group
- Amazon Web Services
- AMD
- Apple
- ARM
- Baidu
- Bitmain Technologies
- BrainChip
- Cambricon Technologies
- Cerebras
- EdgeCortix
- Esperanto Technology
- Graphcore
- Groq
- HiSilicon Technologies
- Huawei
- IBM
- Intel
- Kneron
- Kunlun Technologies
- Lightelligence
- Lightmatter
- Lighton
- MediaTek
- Mobileye
- Mythic
- NVIDIA
- Optalysys
- Qualcomm
- Rebellions
- SambaNova Systems
- Samsung
- Sophon.AI
- SynSense
- Syntiant
- T-Head Semiconductor Co.
- Tachyum
- Tesla
- Unisoc
- WaveComputing
