Marktechpost (@Marktechpost)T2·Infra
58 pts
Fireworks AI CEO Lin Qiao argues that zero data retention in AI infrastructure is widely misunderstood and actually depends on API design: customer-managed retention (upload, use, delete) versus ephemeral processing (send, compute, return). Batch APIs at OpenAI, Bedrock, Gemini and Fireworks follow the former, retaining data until deletion, while inference APIs like Chat Completions follow the latter. Customers retain control in both models, but ephemeral processing is preferable where workflows allow, avoiding forgotten datasets and missed delete calls.