Discovering Disruptions in Tech with Sanjay Poonen of Cohesity

Thank you, good to see you and thanks for having me on your show.

Yes, I'm based in Silicon Valley since the early 90s when I finished my undergrad and came out to work at Apple. I've been there, but I come here for re:Invent. It's like a big show. I was here for the first re:Invent. I was actually a keynote speaker when I was at SAP. We announced our platform on ERP and HANA on SAP in 2012, I believe.

Yeah, I mean, listen, I was 20 years ago in this space at a company called Veritas that next week we complete the acquisition of... acquiring them, so it's kind of full circle around. I think at the time, looking back 20 years ago, this was sort of a storage-related topic that you abstracted away compute from hardware storage platforms and you were able to do an engine called backup and you neutralize what storage platform they had it on because people wanted to put that on originally tape and then later on disk and whatever have you. I think what changed, from my vantage point watching this space over the last 10, 15 years at VMware, was one: the cloud architecture brought a very different way of looking at web-scale ways of doing this, which the movement towards hyper-convergence, and our founders were able to take advantage of. I think the birth of Cohesity was driven on hyper-convergence as the way the cloud works. EC2, S3 were sitting at re:Invent, and Mohit and the team building on kind of the Google file system ideas were able to make that like principle number one. I think there was a watershed moment 10, 11 years ago where you built things in a web-scale architecture that was very different from pre that. And I think then it was validated by every company after Cohesity emulating the Cohesity architecture. So I think that's one shift that made things just faster, better, and cheaper in a cloud web-scale architecture.

I think the second thing that started to happen over the last five years ago was every ransomware attack started happening on backup data because the bad guys realized, 'Why would I want to go after your primary copy if I could go after all of your data that's in the backup? So you okay, you like to pay me. I get all of your data. The likelihood of you paying goes up exponentially when you can't recover.' And backup is basically a time series index of everything you have, right, in a compressed fashion. So every attack on data started to become on secondary data, the backup data, vaulted, archived, snapshotted data. And I again, I'm actually not a file system or data management expert. I'm a security guy most of my life at VMware. And I began to see this happen. I said, 'Pretty smart that this is now becoming not a storage and infrastructure topic but a joint topic with security.' And you know, most of my interest as the board called me to come and be CEO of this company was to significantly invest behind a security vector to protect the world's data from attacks. And you know, today 200 attacks every second, not all of them successful. And the bad guys, nation-state actors, they're typically just trying a lot. It's like these drones that Iran's sending. You just have to get one of them to land someplace and then everyone worries. I mean, so they're cheap to build and there's a lot of people who do stupid stuff in terms of cyber hygiene. So multi-factor authentication, we're surprised how many people don't do that. Encryption of data, segmentation of your applications. In fact, I was at VMware, I wrote an article about the five rules of cyber hygiene that everyone should learn. You'd be surprised how many companies don't do. Now the cyber defenses at major companies that spend a lot on security, like banks, the walls are pretty high and it's harder. You don't see cyber attacks against the JP Morgans. They spend a lot, not successful ones. But the mid-tier banks have some vectors that might be hitting them and some of them have gotten exposed. But hospitals, it's merciless. I mean, people are going after hospital records, schools, I mean, Unified School District. It's just there's so many of them and you kind of have a little bit of a heart of pity because these guys don't have the budgets of the bigger banks and so on and so forth. But they are... and it's not like they've got a lot of... I mean, in the case of hospitals, medical records, but it's not a lot of financial data. Sure, it's kids' school grades, but they're just looking for an embarrassing article in New York Times or other times that they could say, 'We attacked the United States and look who fell.' Sure, it's also like in those cases, it's also networks that are highly democratized. Exactly right. Hospitals have an enormous amount of mixed user use cases everything and haven't invested in correct security infrastructure. They are not able to pay their people probably as much as the banks could. Their IT budgets are smaller, so they're vulnerable. And it's like a bully picking on a vulnerable. But they're just looking for claim and fame. Or the attacks on the hospitality sector here in Las Vegas, consumer goods, retail, manufacturing, all places that are on the lower scale of technology investment. And MGM goes down, they're out for like 30 days. Sure, because they refuse to pay the ransom. So well, also like if you remember reading that all their recovery procedures were also encrypted. They didn't have printed recovery procedures anywhere and they said that added weeks of recovery because all those procedures were also encrypted and they had no way to get those back. So they kind of had to manually do everything. And there in lies the opportunity we felt for how you should think about this as a guaranteed faster recovery. So one of the things we've really focused on is making sure that when we simulate an attack with our customers, we can guarantee them the fastest recovery of anybody in our space and really just perfect the technology on speed of cyber recovery for every workload possible: on-premise virtual machines, databases, NAS files, in the cloud AWS or Azure, Google, SaaS apps. And that's the thing that I keep emphasizing on engineering: process, cycle coverage.

And then I think the third trend right now is that once you've got your data hunkered down, it's safe, it's pointless just leaving it in a bunker. Why not use these AI tools to get insights out of there? Because it really is your gold. So that had not been operated on by many of these companies because they weren't thinking like database companies or like companies that like Databricks or Snowflake. But you know, a backup is just basically a highly compressed version of a whole bunch of files or databases. And if it's safe, you don't want to just keep it in a bunker somewhere. You want to use it with AI tools to get insights in it. So any of our customers have 10, maybe 100 million PDFs, contract documents. Wouldn't it be nice if you could use a co-pilot or a ChatGPT type of query that says, 'Summarize all the contract terms that I did with my vendors the last 10 years and give me a nice summary table.' That's GPT-4 has the ability to do that on a single document easily, right? Sure. Why can't you do that on 100 million documents that are sitting in your backup? So we cracked that problem with NVIDIA. So these are things that I think have been very exciting about this space. One is transforming it to a cloud architecture. Number two, making it really ready for the modern threat vector, security. And then taking advantage of some of these AI tools. And you know, if you're a tech person like you and me, you see these forces coming together: cloud, security, and AI. And you ask yourself, how do you build a modern company that could be a Snowflake or a Databricks or a VMware one day? And that's what we're trying to do.

Yeah, absolutely. I think you hit the nail on the head. And you know, I think one is if that data is safe and it's got compute resources close to it, you can run many of those AI algorithms without affecting your primary workloads that have to be always on. So think about if your Snowflake or your Databricks or your other modern stores of data have to be... latency is going to be really low to get a query back. The backup of that data could be sitting on another farm of servers. It doesn't have to be hit all the time. For sure, you can run that infrastructure overnight with compute on it and you don't have to run as much compute. So the GPU cycles and the GPU requirements are lower and you have more time to be able to do it and you can do it safely. And I think that's another reason why. And most often people were using compute to do the backup and then computers are sitting idle after, right? Right, so you have a lot of possibilities. There wasn't as much use of GPUs in that structure, need CPUs for backup. But I think this whole idea of search and summarization of backup data opened the opportunity from NVIDIA. And that's actually one of the reasons NVIDIA has got a very interesting story. They chose to invest in us because they saw the opportunity to attach their GPUs and their software stack adjacent to backup, right? Because effectively we're now moving into a concept that I don't think ever took off the way it should have: data gravity. Yes, right. How do I move? It's very, very expensive to move data. Yes, right. So since I've already moved the data and now I have compute sitting there unused, there's no other workload that runs other than backup on backup right now. How do I really leverage that? Going, okay, the data is here, let's move compute close to that. I already have spare compute, so now I can add on some GPUs. It seems like a really logical kind of transition.