Model Training
The computationally intensive process of teaching an AI model by feeding it data and adjusting its parameters to minimize errors, often requiring thousands of GPUs running for weeks or months.
Training compute for frontier models doubles roughly every 6 months. Training GPT-4 required an estimated $78–191 million in compute costs and consumed 3,600 MWh of electricity — enough to power 120 American homes for a year. The Stargate project, a $500 billion AI infrastructure initiative, is largely driven by the need for more training compute. Training costs are a key barrier to entry, concentrating frontier AI development among a handful of well-funded companies.
Live Data
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.
ChatGPT
OpenAI's conversational AI assistant, launched in November 2022, which catalyzed the current generative AI boom by demonstrating the capabilities of large language models to a mainstream audience.
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.
AI Economy Pulse
Weekly AI economy data in your inbox. Free forever.
Latest: “AI Investment Hits $42B in Q1 2026 — Here's Where It Went”
No spam, ever. Unsubscribe anytime.