Chips for artificial intelligence

The main articles are:

• Artificial intelligence
• Processors

Neuromorphic processors

Main article: Neuromorphic processors

Chronicle

2024

Production of photonic integrated circuits for AI launched in Russia

Russian manufacturers manufactured photonic integrated circuits for the topology of 90 and 350 nanometers in December 2024. The development is designed to create optical quantum and neuromorphic processors necessary to accelerate artificial intelligence systems and computing devices. Read more here.

India unveils its first AI processors

In mid-August 2024, Ola Electric, one of India's largest manufacturers of electric two-wheeled vehicles, introduced the country's first processors for artificial intelligence tasks. Announced chips Bodhi 1, Sarv 1 and Ojas for applications of different classes. Read more here

Challenge for Nvidia and Qualcomm. Production of 6nm 12-core AI processors begins in China

At the end of July 2024, the Chinese company Cixin Technology announced the start of mass production of Cixin P1 processors for devices with support. artificial intelligence This product can be an alternative to some chips and. Nvidia Qualcomm More here

Introduced a super-powerful AI processor that is 20 times faster than Nvidia

On June 25, 2024, startup Etched announced a super-powerful accelerator called Sohu for artificial intelligence tasks. It is said to be tens of times faster than Nvidia's GPU-based solutions. Read more here

The world's first language processor is presented. It will revolutionize the AI market

At the end of February 2024 startup Groq , he introduced a specialized LPU (Language Processing Unit) processor designed to speed up the work of large language models (LLM). The product is expected to revolutionize the market. artificial intelligence More. here

Memory sharpened for AI has been released. It is 50% faster and denser than analogues

On February 27, 2024, Samsung announced high-bandwidth memory HBM3E a new generation that is designed for use in artificial intelligence systems. The products are said to be more than 50% superior to previously released similar products in terms of data transfer rate and capacity. Read more here.

2023

The global AI accelerator market grew 224% over the year

In 2023, the global market for GPU-based accelerators and specialized accelerators increased by 224% compared to 2022. The driver of the industry was the high demand for servers designed to handle loads associated with artificial intelligence and large language models (LLMs). This is stated in a study by Dell'Oro Group, the results of which TAdviser got acquainted with in early October 2024.

The accelerator market is growing steadily as hyperscalers deploy next-generation AI infrastructure for larger AI models and growing inference needs. The adoption of AI applications in enterprises is accelerating, which leads to the need for additional capacity both in public clouds and in private data centers, says Baron Fung, senior director of research at Dell'Oro Group.

𝓲

Freepik

It is noted that in 2023 Nvidia led in revenue in the segment of components for servers and storage systems. This is due to the increasing sales of GPU-based accelerators. Next in the ranking are Intel and Samsung. Revenues from accelerators in 2023 exceeded revenue from central processors (CPU) for the first time, reflecting a shift towards high-performance computing. GPU sales revenue is projected to continue to grow rapidly.

Revenues in the Smart NIC segment rose more than 50% in 2023 due to the widespread adoption of hyperscale solutions for both AI and traditional use cases. Smart NIC solutions accelerate certain network functionality, thereby reducing the load on server CPUs. This improves the efficiency of demanding tasks.^[1]

Sales of AI processors in the world for the year increased to $27.31 billion

In 2023, global sales of artificial intelligence-enabled processors reached $27.31 billion. This is a quarter more than the result for 2022, when the volume of sales was estimated at $21.8 billion. The key driver is the rapid introduction of machine learning tools and big data analysis tools in various fields. Industry trends are addressed in the Market Research Future survey published in early September 2024.

The authors of the study note that AI is used in a wide range of applications - from facial recognition and natural language processing to predictive analytics and the generation of all kinds of content. Against this background, specialized hardware solutions optimized for processing AI loads are becoming more and more in demand. In addition, there is a tendency to integrate AI chips at the peripheral level, which allows you to process data in real time without the need to transfer them to cloud platforms. This is particularly important in areas where there is a need for decisions with minimal delays, such as autonomous vehicles or industrial automation. As peripheral AI gains momentum, the demand for hardware products designed for such tasks increases.

The introduction of AI is also stimulated by government initiatives: authorities around the world are investing huge amounts of money in related research and development. This leads to the rapid development of new AI technologies and applications, which in turn generates additional demand for specialized processors.

The authors of the report divide the market into four segments: central processors (CPUs) with AI functions, graphics chips (GPUs), user-programmable gate arrays (FPGAs) and special purpose integrated circuits (ASICs). In 2023, CPUs held the largest share of the market under consideration, due to their versatility and economic efficiency. However, GPU-based products are rapidly gaining popularity due to their high performance in mass parallel computing. In terms of the use of AI chips, image processing and recognition, natural language processing, machine learning, deep learning and predictive analytics stand out. The image processing and recognition segment showed the largest result in 2023 - more than 35% of total revenue.

Significant players in the global AI processor industry are, Baidu, Cerebras SambaNova Systems, Hailo Technologies,, Apple Mythic,,,,, Groq,, Nvidia,, and Intel. with Microsoft Amazon AMD its Graphcore Google Qualcomm North America advanced technology infrastructure dominates the global market, with top tech companies based here investing heavily in AI-related R&D. In second place is where Europe there is a strong presence of startups in the field of AI. The Asia-Pacific region, led by China, is showing significant growth thanks to government initiatives and the presence of a large production base. South America, the Middle East and have Africa significant potential as governments recognize the benefits of AI and invest in its development.

Analysts believe that in the future the industry will develop steadily. The CAGR (compound percentage CAGR) is projected at 25.24%. As a result, by 2032, the global market for processors for AI tasks could reach $206.9 billion.^[2]

Global sales of AI chips for the year rose to $53.66 billion

At the end of 2023, global revenue from the supply of artificial intelligence chips reached $53.66 billion. In the future, this segment is expected to grow rapidly, as stated in the Gartner study, the results of which were released on May 29, 2024.

Analysts note that the growth of demand for accelerators with graphics processors(GPU) and specialized accelerators is facilitated by the rapid development of services. generative AI (Genia) Such hardware solutions are in demand in data centers () DPC and cloud sites. At the same time, there are more and more computers equipped with a neuroprocessor unit (NPU) to speed up operations related to AI.

𝓲

Gartner

According to Gartner, in 2024, global sales of AI chips will reach $71.25 billion, rising 33% compared to 2023. And in 2025, the costs in the segment under consideration can reach $91.96 billion. The study says that in 2024, AI accelerators for servers will provide approximately $21 billion in revenue. AI chips for computer electronics will account for $33.4 billion, or about 47% of the total market. Another approximately $7.1 billion will bring AI products for automotive systems, about $1.8 billion - solutions for consumer electronics.

As of 2024, the bulk of AI computing loads in data centers are performed by GPU-based accelerators. At the same time, all major hyperscalers, including AWS, Google, Meta (recognized as an extremist organization; activities in the Russian Federation are prohibited) and Microsoft are investing in the development of their own chips optimized for AI. Although such products are expensive to build, the use of specialized solutions can improve the efficiency of services, as well as reduce the cost of providing users with AI-based services.

Analysts at Market.us Scoop are also talking about the rapid expansion of the global AI chip market. According to their estimates, the volume of the industry in 2023 reached about $23 billion (the figures differ from the values ​ ​ of Gartner in connection with a different calculation method). In the future, the CAGR (compound percentage CAGR) is expected to be 31.2%. As a result, by 2033, spending could rise to $341 billion.

Among the key factors contributing to the growth in the supply of AI chips, Market.us Scoop specialists name the integration of AI technologies into mobile and household devices (smartphones, smart home equipment, etc.), the development of peripheral computing, achievements in the architecture of neural networks, the need to increase the energy efficiency of AI platforms, quantum computing, as well as expanding the scope of AI in general.

On the other hand, there are certain problems. A significant obstacle is the technological complexity of the design and production of AI chips. Another challenge is integrating such products into existing systems and infrastructures. In addition, developers have to solve the compatibility issues of various components, as well as create specialized software that can fully unleash the potential of AI chips.

In general, analysts emphasize, AI chips play an important role in stimulating innovation. They allow the development of new models and applications of artificial intelligence, contributing to progress in areas such as the creation of promising drugs, environmental monitoring, smart city technologies, etc. The introduction of AI chips also contributes to economic growth through the creation of new markets.^[3][4]

2022: Hindus create processor that is 100 times faster than GPU for data centers

On February 2, 2022, the Indian company Qpisemi announced sales of processors specifically for applications related to artificial intelligence - AI 2.0. They, according to the developers, are 100 times more powerful than GPUs used in data centers. It is assumed that this technology will contribute to the development of technology in the field of bioinformatics, drug discovery, machine intelligence modeling and production optimization. Read more here.

2020: Created the most powerful processor for artificial intelligence. It has a trillion transistors and 381 thousand cores

At the end of November 2020, Cerebras engineers presented the most powerful processor for artificial intelligence with 381 thousand cores. This solution has already begun to be used by the company's customers. Read more here.

2019: Intel unveils Nervana Neural Network Processors to train neural networks

On November 13, 2019, the company Intel introduced the Intel Nervana Neural Network Processors (NNP) for Training (neural networks NNP-T1000) and Logic Inference (NNP-I1000) accelerators. These first Intel Specialized Chips (ASICs) for complex tasks machine learning with good scalability and efficiency are designed for customers developing cloudy technology and. data centers In addition, Intel introduced the Intel Movidius Myriad generation of visual processors (Vision Processing Unit or VPU) for processing media data on peripherals, building stand-alone systems computer vision , and building logical leads. More. here

2018

Intel plans to squeeze Nvidia into the neural network chip market

As of 2018, almost all applications, one way or another related to neural networks, run on Nvidia servers, and if otherwise, still on Nvidia GPUs. But there is a serious chance that Nvidia's monopoly will be broken by Intel's efforts. A competitor capable of squeezing, or maybe even shifting, the GPU from the position of leader will be the new, unparalleled Intel Nervana Neural Network Processor (NNP) processors. They, as the name suggests, include intellectual property acquired by Intel with Nervana in 2016 (more).

Amazon develops AI chips

In February 2018, it became known about the development of Amazon's own chips. They focus on computational tasks related to artificial intelligence (AI). Read more here.

Notes