“We don't see that going away anytime soon, and it applies to inferencing as well as well as training.”

In context

“lished all the time and for your AI models to be relevant they have to be up-to-date. So it's not that you train and then do nothing but inferencing thereafter. So you're training and your inferencing sit right next to each other. As long as we can do this stuff twice as fast as everybody else that's on the -- not just on the training side, that's also on the inferencing side, then we are going to be half the cost or better. So we think we are going to be very, very competitive across the board whether it's training or an inferencing. So we don't -- so we are pretty confident that we've got a cost-performance advantages. Again, if you run twice as fast in the cloud, you cost half as much because you pay by the hour. So the performance advantage is really an enormous cost advantage for us. We don't see that going away anytime soon, and it applies to inferencing as well as well as training. Now as far as GPUs, are GPUs a low-margin business? Not for a 100% automated cloud with very, very low cost. We think, in some cases, our prices for GPU training, which are very profitable by the way, for us, but are often lower -- our prices are lower than the cost of other hyperscalers doing the training. Mark Moerdler: Thank you. That's very helpful. Operator: We'll take our next question from Keith Weiss with Morgan Stanley. Keith Weiss: Excellent. Thank you, guys, and thank you for taking the question. I wanted to drill in on Cerner, basically, the one-year anniversary of that acquisition. And maybe from Larry, get an update on where we are with modernizing that solution and modernizing that product? And basically then whether Cerner has kind of lived up to your expectations thus far”