HBM (High Bandwidth Memory)
A specialized memory technology that stacks memory chips vertically and connects them with wide data buses, providing the massive bandwidth needed by AI accelerator chips.
HBM is essential for AI performance — the NVIDIA H100 uses 80GB of HBM3 memory with 3.35 TB/s bandwidth, while the B200 uses HBM3e with over 8 TB/s. SK Hynix and Samsung dominate HBM production, with a combined market share exceeding 90%. The HBM market is projected to exceed $20 billion by 2025, growing at over 80% annually. HBM supply constraints have been a bottleneck for AI chip production. Memory bandwidth, not compute power, is increasingly the limiting factor for AI model performance, making HBM a critical strategic technology.
Explore the Data
Related Terms
AI Compute
The computational resources — primarily GPU and TPU processing power — required to train and run AI models, typically measured in FLOP (floating-point operations) or GPU-hours.
Capex (Capital Expenditure)
Long-term investment spending by companies on physical assets like data centers, GPU clusters, and networking infrastructure — the backbone of AI deployment at scale.
Data Center
A facility housing computer systems and infrastructure used to process, store, and distribute data — increasingly built specifically for AI training and inference workloads.
Fine-Tuning
The process of further training a pre-trained AI model on a specific, smaller dataset to specialize it for a particular task or domain, requiring far less compute than training from scratch.
Foundation Model
A large AI model trained on broad data that can be adapted to a wide range of downstream tasks — examples include GPT-4, Claude, Gemini, and Llama.
Frontier Model
The most capable and advanced AI models at any given time, typically trained with the largest compute budgets and achieving state-of-the-art performance on benchmarks.
AI Economy Pulse
Every Friday: the 3 AI data points that actually matter this week. Free, forever.
Latest: “AI Investment Hits $42B in Q1 2026 — Here's Where It Went”
No spam, ever. Unsubscribe anytime.