A prediction market on Manifold (source: asks whether DeepSeek lied about the GPU compute budget used in training v3.

A prediction market on Manifold (source: https://manifold.markets/AndrewPeterson960c/did-deepseek-lie-about-the-number-o) asks whether DeepSeek lied about the GPU compute budget used in training v3. As of 2026-05-26, the consensus probability is only 4.01% (YES), indicating the market considers the claim highly unlikely. While this is a single, low-confidence source with no corroborating evidence, the question itself touches on DeepSeek's efficiency narrative, which is relevant to AI infrastructure and semiconductor demand. If true, it would imply DeepSeek's reported efficiency gains might be overstated, potentially affecting competitive dynamics and GPU procurement patterns. However, given the low probability and lack of concrete evidence, the signal is weak. No spillover entities are identified.