Closed-source AI

Definition

AI models whose weights, training data, and architecture are proprietary and accessible only through APIs. OpenAI, Anthropic, and Google run closed-source models, monetizing via usage-based pricing.

Why it matters

The open-versus-closed debate is the most consequential strategic question in AI. Closed-source vendors offer higher capability ceilings and managed infrastructure, but create vendor lock-in and data exposure risk. Open-weight alternatives offer control, privacy, and cost predictability, but require engineering investment. For enterprises, the choice depends on your risk tolerance, regulatory environment, and engineering capacity. Most mature organizations run a hybrid strategy: closed-source models for frontier capability tasks and open-weight models for high-volume, cost-sensitive, or privacy-critical workloads. Anyone going all-in on a single closed-source provider is taking a bet that rarely pays off long-term.

In practice

OpenAI's GPT-4 and Anthropic's Claude remain closed-source, with access only through APIs or managed platforms. Google keeps Gemini Ultra closed while releasing smaller Gemma models as open-weight. The competitive dynamic intensified when Meta's Llama 3 405B achieved near-GPT-4 performance as an open-weight model, pressuring closed-source vendors on pricing. Enterprise surveys show roughly 60% of companies use both closed and open models, with closed models for complex reasoning and open models for high-volume tasks like classification and extraction.