With batching he means entity rendering I think.

So for exemple gras, 2 different models, but 200 instances all send individually to the gpu (without batching). Matt has this in sphere 2.