$NVDA · Mistral CTO praises Blackwell GB200/GB300 and NVFP4 for 2.5x training speedup on MoE models

episode briefing

extracted intelligence

NVIDIA

How Mistral Is Building Frontier AI for the Enterprise | NVIDIA AI Podcast Ep. 301

now viewing▶ 15:57

Mistral CTO praises Blackwell GB200/GB300 and NVFP4 for 2.5x training speedup on MoE models

2026-06-10 · 2 company · 9 thematic

sentiment

2 bull0 bear0 neu

speakers

tim lacroix

Co-founded Mistral AI two and a half years ago with Guillaume Lample and Arthur Mensch after research roles at Meta; leads technology strategy including open-weight models, enterprise platform (Forge), and owned infrastructure (Mistral Compute).

quote

“Yeah, I mean definitely the GB200 which we've been using since June of 2025, I believe, we quickly saw a 2.5x improvement, at least, like, out of the box when training large, sparse mixture-of-experts models especially... we run a lot of o…”

—tim lacroix

now playing · $NVDA

$MISTRALbullish· high· postim lacroix

Mistral CTO outlines open-source frontier strategy with enterprise platform and own compute

Mistral is executing a full-stack strategy: releasing open-weight frontier models to accelerate community innovation, while building an enterprise platform (Forge) for model custo…

$NVDA···bullish· hightim lacroix

Mistral CTO praises Blackwell GB200/GB300 and NVFP4 for 2.5x training speedup on MoE models

NVIDIA's Blackwell architecture delivers a 2.5x out-of-the-box improvement for training large sparse mixture-of-experts models, and NVFP4 quantization enables efficient inference…