Data center AI Compute Infrastructure: Markets, technology, products and market forecast

The race for making perfect hardware to accelerate artificial intelligence (AI) applications has begun and many companies are jumping in with their products and solutions. Of the three key parts of hardware infrastructure – compute, storage, and networking – compute has made significant progress in the last couple of years wit many chipsets coming to market. This report focuses on the IT infrastructure for AI from a market and technology point of view. The reports analyzes technology and business trends for three key AI compute acceleration products: accelerator cards, workstations. There are four parts to this report: market, technology, products and forecast.

The market part focuses on use cases, applications and vertical markets. A summary of 25 enterprise vertical markets and their ramp up for AI usage and infrastructure purchase is analysed. The report also looks at hyperscalers and cloud as a market and analyzes differences between the buyers. A preview of the market drivers, challenges and business models is also provided.

The technology part focuses on wide range of aspects such as acceleration needs, architecture, bus interface, AI application needs, scaling, alternate architectures and utilization. It also identifies how different chipset architectures such as CPU, GPU, FPGA and ASICs are being used for AI acceleration. The technology focuses on aspects such as bus interface, AI application needs, scaling and utilization. Details of storage and networking aspects and how that pertains to AI acceleration is also provided. Technology section also discusses aspects such as training versus inference and how the compute infrastructure is different for the two.

The product section focuses on different vendors, their products, differentiators and our assessment. Products for both training and inference are considered along with pricing and business model aspect. A list of buyers and users is provided for each vendor based on publicly available information. We have also evaluated start-ups and how they could pose threat to dominant players. Finally, profile of key companies along with comprehensive listing of vendors is provided.

Market forecast is also provided for the products segmented by world region. The forecast includes data for servers, workstation, and accelerator cards. Data includes unit shipments, average selling prices, and revenue. Split by geography and training versus inference is also included.

Key Questions Addressed

• How has the AI hardware market evolved in the past couple of years?
• What are some of the new technologies being used in compute acceleration?
• How does network and storage relate to AI acceleration?
• What are the key technology issues surrounding Big Data in AI applications, and why is it so hard to solve them?
• What are prominent use cases and applications for AI in the enterprise market?
• What are the acceleration needs of different use cases?
• What are some of the popular data types that need AI acceleration?
• Who are the key players in the AI hardware infrastructure market, and what are their relative strengths and weaknesses?
• What is the size of the market opportunity in different hardware categories, and why will some grow faster than others over the next 5 years?