> we expect Maia 300 to deliver 40% more tokens per megawatt compared to leading GPU-based systems.

Apparently, not as performant as Cerebras or Groq?