So John, as you said, right, the biggest problem right now in AI clusters is how do you train bigger and bigger models, right? 

As you know that the model sizes have been increasing, we are now getting up to 20 to 50 trillion parameter models in the next three to five years. 

To train those models, to run inference on those models, you really need much bigger GPU clusters than we have today. 

And the only way to get to that is to basically go from copper because copper runs into distance limitations and that is the only solution is to run with photonics. 

And that's what we enable.