Keystone offers flexible consumption models that go beyond just how much capacity at such and such a speed, et cetera. We don't always know what the profile of that data will be. However, if we can quickly agree on terms that meet our needs and make NetApp still reasonably profitable, we can confidently deploy, see how things go and adjust. That kind of service delivery model, that customer service model has sped things up and made contract negotiations much easier. It frankly made the owners of the system more confident. We've experienced faster time to market. It's hard for us to find and retain infrastructure staff. We're in a business where the firm fixed price contract reigns supreme, and so we can't always just offer someone more money. However, if they can dedicate their time to learning one company's portfolio and learning it really well, but be useful in a bunch of different places, they will do well. We've seen that in a lot of different places. We've been able to hire younger people and retain them, moving them from program to program based on their understanding of the solution its skill set, and its portability. It's been useful for high-speed multi-protocol storage in places with ever-increasing density. We have limits on how much power and cooling and rack space we have, and yet they've delivered every time. We needed a storage company that had mastery of multi-protocol, and this solution stands out. They especially stand out as a secure provider. We require solutions that we can run ourselves, that we can air gap since so much of what we do is either classified or very sensitive or cannot live in a public ecosystem. For us, the issue consuming AI has been the trust of the models given to us by third parties. We can't necessarily trust their provenance, what fed them, what originally trained them, or what gave them their worldview, for lack of a better term. We can't simply just trust that at face value since we know nothing about where it came from or what inferences it might make. We must assume that some AI inferences were made deliberately to damage or hurt national security systems. So the models that we start with tend to be very, very primitive, crude, and not well trained, so we have to train them much longer and not always with the availability of cloud that has inexhaustible capacity. A partner who understands this and provides consistency at all scales is very important.
