AMD sees 35% revenue growth, targets AI chip surge by 2027

AMD expects its annual revenue to rise by about 35% over the next three to five years, driven by growing demand for AI data center chips.

CEO Lisa Su said the company’s AI data center business could see 80% annual growth and reach tens of billions in sales by 2027, helping AMD achieve double-digit market share in the segment.

The US chipmaker recently announced a multi-year deal to supply its Instinct AI chips to OpenAI starting in 2026, alongside partnerships with Oracle and Meta.

AMD reported US$5 billion in AI chip sales for fiscal 2024 and updated its forecast, estimating the total AI data center parts and systems market could reach US$1 trillion annually by 2030.

🔗 Source: CNBC

🧠 Food for thought

Implications, context, and why it matters.

AMD’s 2027 AI goal hinges on scarce advanced packaging

AMD targets “tens of billions” in AI accelerator sales by 2027. That needs Chip-on-Wafer-on-Substrate (CoWoS) capacity to integrate processors with High Bandwidth Memory (HBM), with Nvidia holding 60% of 2026’s 1 million wafer demand, or 595,000 units 1.
The company booked 105,000 CoWoS wafers for 2026, an 11% share, with 80,000 from Taiwan Semiconductor Manufacturing Company (TSMC) for Instinct MI355 and MI400 1. It needs a larger allocation to reach the goal.
TSMC plans 93,000 to 100,000 CoWoS wafers each month in 2026 2 3. Blackwell and Rubin orders absorb much of the growth 2, so AMD must win more from TSMC or Powertech Technology 4.

Software vendors should pursue ROCm migration as Microsoft explores CUDA-to-ROCm

Microsoft built Compute Unified Device Architecture (CUDA) to Radeon Open Compute (ROCm) converters 5. This opens a market for vendors offering migration, optimization, or managed inference tools, since CUDA code can run on AMD chips without full rewrites.
ROCm now supports PyTorch 2.6 to 2.8, TensorFlow 2.17 to 2.20, and Google’s JAX 6. That gives MLOps (machine learning operations) platforms a base for compatibility layers, along with developer-tool makers.
IT system integrators and cloud migration firms can address ROCm’s incomplete Application Programming Interface (API) coverage 5 with automated tuning. Focus on inference jobs where AMD pricing helps, since Microsoft’s toolkit hints that limited runtime translation can work.