AMD CEO Lisa Su Reveals Ambitious AI Accelerator Market Projections: Inference Set to Surpass Training by 2026

AMD’s projections for the datacenter AI accelerator market have significantly evolved over the past few years, reflecting the rapid changes driven by advancements in AI technology and the ongoing shift towards generative AI (GenAI). In June 2023, CEO Lisa Su predicted that the market would grow from $30 billion in 2023 to over $150 billion by 2027, at a compound annual growth rate (CAGR) of over 50%. This forecast seemed reasonable, considering the overall datacenter systems market was valued at $236 billion in 2023, according to Gartner, and the broader IT market was estimated at $2.72 trillion. However, just six months later, at the launch of the third-generation "Antares" Instinct MI300 series of datacenter GPUs, Su revised her projections. She now expected AI accelerator sales to be $45 billion in 2023, growing at a CAGR of over 70% to surpass $400 billion by 2027. Another year passed, and in October 2024, Su extended her forecast to 2028, cooling the CAGR to over 60%, but still predicting the market to exceed $500 billion by 2028. The most recent projection, unveiled at the Advancing AI event in October 2024, provides a detailed breakdown of the market from 2023 to 2028. According to this data, the market for AI accelerators in the datacenter will reach approximately $600 billion by 2028, representing a 67.5% CAGR over the five-year period. Despite the rapid growth, Su noted that the market's expansion might slow in 2025 as customers wait for more advanced 2026 models before making significant investments. One of the key trends Su highlighted is the shift towards AI inference, which is expected to become the primary driver of AI compute demand. Inference, which involves running trained models to make predictions, is projected to grow at a CAGR of more than 80% over the next few years. By 2026, total inference accelerator spending is expected to surpass that of training accelerators. This shift is attributed to the emergence of new use cases and reasoning models, which require different types of systems with larger memory domains and tighter coupling. The implications of these projections are substantial. Using a multiple of 2.1 for full AI training systems and 1.8 for inference systems, the $180 billion in AI accelerator sales projected for 2025 could drive $351 billion in AI system sales, outpacing Gartner's estimate of $405.5 billion for total datacenter systems spending that year. By 2028, Gartner projects $589 billion in total datacenter systems spending, while AMD's data suggests AI system spending alone could reach around $1.13 trillion. These discrepancies highlight the uncertainty and potential for exponential growth in the AI market. Breaking down the AI market, it is estimated that one-third of the spending will be on distinct AI training systems, half on distinct inference systems, and the remainder on systems capable of handling both tasks. This distribution underscores the complex nature of AI architectures and the need for flexible, programmable compute resources. Looking ahead, AMD's potential market share gains are a critical aspect of these projections. Even with a pessimistic forecast for 2025, where AMD might secure only $6.1 billion in sales, the company's share could rise to 7.5% in 2026, 10% in 2027, and 15% in 2028. If these projections hold true, AMD's GPU sales from the datacenter could reach nearly $90 billion by 2028, a figure nearly four times the company's total revenue in 2024. Industry insiders commend AMD for its transparency in forecasting AI market trends, a practice Nvidia typically avoids. The accuracy of these projections, however, remains a subject of debate. While Gartner's projections offer a conservative view, AMD's outlook aligns with the explosive growth witnessed in the AI sector. The company's commitment to innovation and flexibility in its products positions it well to capture a significant share of this burgeoning market. Nevertheless, the rapid evolution of AI technologies and the competitive landscape mean that future outcomes are still uncertain, and continuous monitoring will be essential for staying ahead of the curve. AMD's ambitious forecasts and clear strategic direction suggest that the company is well-prepared to capitalize on the opportunities presented by the AI revolution.