FPGA: A Versatile Chip Igniting New Momentum and the Future of Domestic Substitution (2023)

FPGA chips offer high flexibility, low development cost, and short time‑to‑market, making them attractive across diverse downstream markets such as industrial control, networking, consumer electronics, data centers, automotive, and artificial intelligence.

According to Xilinx financial reports, the share of data‑center applications grew from 7% in 2019 to 10% in 2021, outpacing other sectors and indicating strong growth momentum.

In 2022, AMD and Intel saw their FPGA businesses boost data‑center revenue, with AMD’s acquisition of Xilinx driving revenue from $3.9 billion in 2021 to $10.6 billion in 2022, while Intel’s DCAI group achieved a 14% revenue increase.

Industry experts predict that CPU+FPGA heterogeneous computing will occupy one‑third of cloud data‑center workloads, and the global accelerator market (CPU, GPU, FPGA, ASIC) is expected to grow from $28.4 billion in 2018 to $211.9 billion in 2023, a 49% CAGR, with FPGA accelerator revenue rising from $10 billion to over $50 billion.

Key questions addressed include why data centers deploy FPGA, the challenges of chip‑level power limits and dark‑silicon, and the rising energy and cooling pressures in large‑scale data centers.

Major vendors are pursuing heterogeneous strategies: Nvidia’s GPU‑centric Superchip (Grace CPU + Hopper GPU) and its DGX GH200 AI engine; AMD’s CPU+GPU Instinct MI300 APU and upcoming CPU+GPU+Memory solutions; Intel’s CPU+FPGA integration within its Data Center and AI Group, planning to launch 15 new FPGA products by 2023.

Market research from Frost & Sullivan shows the global FPGA market grew from $4.34 billion in 2016 to $6.08 billion in 2020 (8.8% CAGR) and is projected to exceed $12 billion by 2025, with China alone expected to reach $33.22 billion in sales and ship 3.3 billion chips.

Long‑term forecasts from leading FPGA vendors anticipate a 16% YoY growth in 2023 and a sustained double‑digit CAGR over the next five years, driven by AI‑induced demand for high‑performance, low‑power compute in servers.