AI is creating unprecedented demand for chips—but the biggest opportunity is only beginning.
According to a joint report by the Semiconductor Industry Association (SIA) and Deloitte, revenue from semiconductors deployed in AI data centers is projected to surpass $1.2 trillion by 2028, nearly ten times higher than five years earlier, as artificial intelligence adoption accelerates across cloud computing, enterprises and consumer applications.
The rapid expansion of AI workloads is driving demand for processors, memory and networking chips that power data centers. The market for AI data center logic chips, primarily AI accelerators, has already grown from $30 billion in 2022 to $70 billion in 2024 and is forecast by World Semiconductor Trade Statistics (WSTS) to reach $190 billion by 2026.
While AI model training has fueled semiconductor demand in recent years, the report says the next phase of growth will come from AI inference—the process of running trained models to generate responses, make predictions and perform real-world tasks. As generative AI adoption expands across businesses and consumers, inference workloads are expected to grow much faster than training, reshaping demand for AI hardware.
Unlike training, which requires massive computing power and memory bandwidth, inference prioritizes low latency, high throughput and energy efficiency. These workloads are increasingly being deployed not only in cloud data centers but also on edge devices such as AI PCs, smartphones, vehicles and industrial equipment, creating demand for specialized AI accelerators and custom silicon.
The report notes that data center operators are increasingly adopting application-specific integrated circuits (ASICs) optimized for AI workloads, while chipmakers are developing dedicated inference processors to improve performance and reduce power consumption. Over time, the industry is expected to see a clearer separation between chips designed for AI training and those built specifically for inference.
Looking ahead, inference is expected to become the dominant driver of AI semiconductor demand. Industry estimates suggest training-related revenue will grow at a 30% compound annual growth rate (CAGR) between 2023 and 2028, while inference revenue is projected to expand by 122% over the same period. By 2032, inference could account for 80% of total AI compute demand, up from around 20% in 2024.
The report also underscores AI’s growing importance to the broader semiconductor industry. McKinsey estimates the global semiconductor market will reach $1.6 trillion by 2030, with AI and data centers driving much of that growth. With the semiconductor industry expected to generate nearly $792 billion in revenue in 2025, generative AI alone could contribute more than 40% of the sector’s future expansion.
