Training Cost
Training cost is the total cost — GPU/TPU compute, energy, data acquisition, and labor — required to train a machine-learning model from scratch or to fine-tune a pretrained one to a target capability or domain.
Full definition
Stanford's AI Index 2024 estimated GPT-4 training cost at $78M and Gemini Ultra at $191M, with frontier costs roughly 2-3x annually. Most enterprises will never train a frontier model; they will fine-tune (typically $1K-$200K) or rely on adapters like LoRA ($10-$10K). The training-cost decision is build-vs-buy: pretrained-API is cheapest until it isn't, fine-tuning crosses over at scale, and full pretraining almost never crosses over outside hyperscalers.
Why it matters
Boards routinely conflate "AI investment" with "training cost," when 80%+ of enterprise AI spend is inference, integration, and people. Properly framing training cost prevents both panic ("we can't compete with $200M models") and overconfidence ("we'll just train our own").
Example
A retailer evaluates training a custom catalog model: estimated $3.2M one-time + $400K/year vs. $180K/year for fine-tuning a frontier API. Build-vs-buy math kills the build case at the first finance review.
Related terms
- Inference CostInference cost is the dollar cost of running a trained AI model in production — per request, per user, or per business outcome — and the operating expense that determines whether an AI feature has positive unit economics at scale.
- Fine-tuningFine-tuning is the process of further training a pretrained model on a smaller, task-specific dataset so it specializes in a particular style, domain, or behavior — without retraining from scratch.
- LoRA (Low-Rank Adaptation)LoRA (Low-Rank Adaptation) is a parameter-efficient fine-tuning method that freezes a pretrained model's weights and injects small trainable low-rank matrices into selected layers — reducing fine-tuning cost by 100-1000x with negligible quality loss on most tasks.
- Build vs Buy (AI)Build vs buy in AI is the strategic decision between developing an AI capability internally — model, platform, data layer — and procuring it from a vendor, hyperscaler, or open-source ecosystem, weighed against differentiation, cost, time-to-value, and lock-in.
Source & further reading
Primary source: Stanford HAI — AI Index Report 2024 (2024).
Citation policy: this entry is part of the AIDOLS AI Implementation Glossary and may be quoted for research, journalism, and education with attribution to aidolsgroup.com/nl/glossary/training-cost/.