00:01
Mhm. Welcome everyone to the discussion on building an AI future with Kerry. My name is Greg Chung. I'm a part of the pure storage alliances team focused on our analytics and AI partners. And I'm joined here with Justin Emerson from pure storage and Tony Pack a day from NVIDIA. Would you like to introduce yourselves, guys?
00:25
Justin, why don't you go first? Hi. My name is Justin Emerson. I'm a principal technology evangelist for pure storage covering the flash played product. And I'm 20 Pike. Glad to be here. I run product marketing for the dejected business unit at video. Thank you for that. So why don't we kick off this discussion with
00:44
the first question on what have we learned together by being the first mover in the AI market with Harry? Well, I think the biggest thing that we realised as area rolled out of the market is that one of the major challenges for customers when it comes to deploying enterprise and infrastructure is simplicity because the market moves so fast because technology is advancing
01:05
so quickly in this area, really not having to worry about the underlying infrastructure complexity is a huge win for our customers so simplifying that environment, making it easy to scale, making it seamless and really reducing That complexity has been one of the biggest win for customers that have adopted area in the Enterprise. You know, I couldn't agree more, Justin. And another factor that's really in play that
01:30
creates urgency around what you just said is the fact that so many organisations that are two companies deal with often times have this widening gap between data science, innovation, namely data science and AI practitioners who are crafting models and the I t teams who build infrastructure that needs to support that effort. Um, you know, we recently conducted a survey, uh, and found that too many over
01:59
half of our respondents enterprises around the globe saw the same disconnect and saw operational ising ai at scale a real challenge. So that's definitely, um, something that now are two companies I think are confronting with our solution with areas. That's really interesting. What effect do you think that's having on the evolution of AI for the enterprise?
02:22
Yeah, great question. You know, if I lean into that survey a little bit more, um, it was remarkable to see that almost 40% of respondents did not involve I t until like the bitter end of developing their AI models and then trying to turn them over to production. Uh, so imagine, you know, building an application and only at the 11th hour asking
02:48
your i t team to deploy when they've had no stake or or ownership or involvement in it. And, you know, we affectionately call this shadow ai or sometimes shadow I t. Um, and there's a lot of factors I play when this happens. Oftentimes, there's lack of a multi functional, multi disciplinary team that should include, obviously your I T stakeholders. Um, sometimes there's a lack of executive
03:11
leadership. And there's this mindset that data scientists in and of themselves are the ones who production allies AI applications versus it. Being a team sport that intimately depends on i t. I couldn't agree more, and that really sort of shows a failure of you know, these different teams that accompany to work together.
03:30
The reason that an AI team would would approach it this way is either one. They're not aware of the capabilities that I t. Maybe building or two. They don't have confidence in the capabilities that I t is building, and we see area as a way for it. Departments to show thought leadership within their organisations about providing the kinds of capabilities and services
03:54
that these data practitioners data scientists data engineers need. Um, and using a reference platform like Harry can help, uh, not only build that confidence, but also, uh, speed up time to value for customers that are looking to deliver these kinds of services to hopefully avoid these kinds of shadow ai environments popping up because they present a huge problem both in regards to wasted cost or
04:21
wasted, uh, investment. But also they can be major security problems. Uh, and and lastly, if if you're building these in the cloud and they start very small and they may be very cost effective, it's a great place to start. But let's say that that one application becomes your business. And now suddenly it's on an environment where, um, it may be very scalable,
04:43
but it can also significantly impact the, you know, profitability or the or the ability of that application to run in a cost effective way. Yeah, great points all around. You know, if I lean into the typical, uh, impacts that we see with those who tend to go cloud first or cloud only, um, there's a remarkable trend that we see that it was kind of,
05:06
uh, parent to us in survey results. And that is that those who were cloud first or cloud only um were often almost up to two X more likely to struggle with scaling A I successfully and conversely, those who succeeded in scaling ai uh easily. 90% of those embraced hybrid infrastructure. So complementing existing cloud resources, which were great for productive experimentation
05:38
with on Prem or complicated infrastructure that offered the deterministic performance they needed. And this really speaks to more and more successful. AI scholars are driving really two modes of infrastructure. As we see it. One is around purpose. Build infrastructure like dairy that delivers
05:57
the performance you need for large, monolithic workloads like natural language processing or recommend er systems and the like, um and or, uh, you know, ai centre of Excellence that essentially service an entire enterprise with multitudes of developers all subscribing to this infrastructure, tackling their individual products that projects that might be much smaller.
06:18
And so we see more and more enterprises embracing this instead of, um, looking at cloud as the end all and Bill and having this hybridised approach, we see it in financial services, automotive, retail, government, public sector. I mean, you name it, and I think that's increasingly becoming the trend. As certainly as these survey results board out,
06:40
those are really interesting points on sort of that evolution as well as challenges. Are there any special announcements that you'd like to make here today? Thanks. Great. So what we're announcing today is R E s, which is the next generation of our very solutions. Um, Harry s takes the best from NVIDIA and pure storage, The latest invidious DG X family,
07:01
the latest, uh, video networking technology and Flash Blade s from pure storage, which was announced earlier this morning, Um, and puts them together in a new new generation of reference architectures, Um, R e s is a platform that we're going to be innovating on further into the future. Um, but really, what I'm most excited about R E s is that it should unlock a new level of capability for our customers.
07:28
Um, flash Blade s provides new levels of performance and capability and density and scale. Um, but almost as important as efficiency. Um, so flash Blade s represents a new plateau, a new paradigm for how efficient these ai training storage environments can be. And that's really important, because when you're trying to deal with these massive
07:52
problems, power efficiency becomes a huge factor because the actual gating factor of how large you can go is how much power you can provide. And so, with Flash Blade s, we can get as low as 1.3 terabytes per watt at really incredible performance numbers. Yeah, I am really excited about, uh, the ability for airy to unpack so much innovation
08:14
and delivered to customers in a way they can consume. And from the video perspective, it becomes the ideal vehicle for the things we've been doing on the project side. Um, if you look at how we've been approaching a I, we understand that AI involves problems that are exceeding the bounds of what you can do on a single GPU or even a single system I mentioned recommend er systems
08:40
NLP. These are all really data centre sized problems. So essentially the computational math, the workload that we're talking about is data centre scale and enterprises are shifting their computing paradigm accordingly. Uh, so they need increasingly ai that infrastructure this purpose built and customised versus off the shelf. And they're coming to our two companies to
09:06
build this infrastructure. That can take a problem that used to be solvable in seven days and shrink it down to 20 hours or less using area. So we get there by continuing to innovate, not just at the accelerator level, but at the everything in between, like the interconnect within systems, between systems, interconnecting storage and just ultra large cluster design.
09:29
Which area benefits from, um, we are doing everything we can between our companies to eliminate latency and the data path across storage and the GPU and the network that interconnects them. And from a D GX perspective, we've taken all the things that we know in hardware and software and technologies like envy, link and envy switch and Magnum io and more and systemized them in a modular building
09:52
block that customers can use scaled to the size of the problem that they're trying to tackle now carried within areas. So, um, areas becomes this incredible foundation to similarly make these technologies and time to solution that much more accessible for customers. That's really great to hear. Tony, can you share more about why you're so excited about this partnership and why our
10:15
customers should be equally as excited? I'd love to. You know, I think back to the earliest days of when our two companies partnered, um, to to create this solution and credit to pure, pure defined the space. You know, we appeared in a lot of the same organisations a lot of the same accounts over and over again and realised that there were
10:36
incredible synergies between our two technologies. Very similar design philosophies, very similar strategies around efficiency. Like the stuff that Jensen mentioned. Uh, sorry that Justin mentioned Jensen's our CEO. You know, we took all of this and realise that our two companies can now embody this in a
10:57
common solution that every righty organisation and their stakeholders can benefit from. And why I'm so excited is that r E s now takes the best of all of that and continues that tradition of pushing the leading edge of AI infrastructure solutions, uh, delivering the best of our innovations so not just with the G s A 100 today, but you know DG X h 100 in the future and whatever comes next.
11:24
You know all of that embodied and solutionize, if you will within the area. Solution. Tony. Justin, Thank you so much for having this conversation in sharing your insights. We look forward to hearing more customers can find more information about areas at pure storage dot com slash cherry. Encourage all of you to visit the website.
11:45
Thank you so much. Thank you, Grace. Thank you.
