According to IDTechEx's latest research report "2025-2035 Data Center and Cloud AI Chip Technology, Market and Forecast", by 2030, the widespread deployment of AI data centers, the commercial application of AI technology, and the continuous improvement of the performance requirements of large AI models will drive the AI chip market to exceed the $400 billion mark. However, the study also emphasizes that in order to meet the needs of efficient computing, cost reduction and efficiency increase, performance upgrade, system expansion, inference acceleration, and domain specialization, the underlying technology must continue to innovate.
Globally, cutting-edge AI continues to attract hundreds of billions of dollars in investment each year, with governments and hyperscalers vying to get a head start in drug discovery, autonomous driving infrastructure, and more.
Graphics processing units (GPUs) and other AI-specific chips play a key role in improving the performance of top-tier AI systems, powering deep learning in data centers and cloud infrastructure. However, with global data center capacity expected to exceed hundreds of gigawatts in the next few years, with hundreds of billions of dollars in investment, the energy efficiency and cost of current hardware solutions are a growing concern.
The largest AI systems today use hyperscale high-performance computing architectures, which rely heavily on GPU clusters and are mainly used in hyperscale AI data centers and supercomputers. Whether on-premise or distributed, these systems can provide exascale floating-point operations.
Image: AI chips and data center market size
Although high-performance GPUs are indispensable in the field of AI model training, they have many limitations: high total cost of ownership, vendor lock-in risk, low utilization of AI computing, and overperformance when handling specific inference tasks. To this end, hyperscalers are gradually turning to customized AI application-specific integrated circuits (ASICs) designed by manufacturers such as Broadcom and Marvell Electronics.
These custom AI chips have a core architecture optimized for AI workloads, are less expensive per computation, can be optimized for specific systems, and enable energy-efficient inference. At the same time, it enables cloud service providers to achieve full-stack autonomy, controllability, and differentiated competition without sacrificing performance.
Traditional chip giants and emerging AI chip start-ups have launched innovative solutions to outperform mainstream GPU technologies. These designs use similar or innovative chip architectures to create solutions that are better suited to AI workloads, reducing costs and increasing efficiency. AI accelerators developed by industry leaders such as Intel, Huawei, and Qualcomm use heterogeneous compute cell arrays (similar to GPU architectures) but are optimized specifically for AI computing, balancing performance, power efficiency, and application-specific flexibility.
Startups focused on AI chips are taking a different approach, adopting cutting-edge architectures and manufacturing processes such as data flow control processors, wafer-level packaging, spatial AI accelerators, in-memory computing (PIM), and coarse-grained reconfigurable arrays (CGRA).
There is a broad space for technological innovation in all links of the semiconductor industry chain. Driven by government policy support and huge capital investment, the development of cutting-edge AI technology will continue to break through boundaries, which will also generate massive demand for AI chips in data centers. IDTechEx predicts that the AI chip market will continue to expand at a compound annual growth rate of 14% from 2025 to 2030, and the market size is expected to reach $453 billion by 2030.
