Neri announced in June that he had tested positive for coronavirus and went into a two-week quarantine. Now, in early August, he's fully recovered and ready to deal with what might be a greater challenge: transforming his company from a server vendor to a service provider.
HPE is betting heavily on a product called GreenLake to launch it into the cloud era, hoping that its customers, and other companies that have lagged behind the transition to cloud computing, will pay for a service that helps them manage computing and storage resources across their own data centers and public cloud providers. The "low-hanging fruit" of cloud business, as Neri put it, has already been plucked, and HPE is one of several companies, including VMware and IBM's Red Hat, vying for the rest of that bounty.
The problem is that revenue from HPE's core businesses is declining at an alarming clip, down 16% during its last fiscal quarter. A fair amount of the decline in that quarter can be chalked up to supply-chain disruptions and economic uncertainty, but HPE needs to start making more money from its modern businesses to offset those declines.
In an interview with Protocol earlier this week, Neri talked about his COVID-19 experience, why cloud computing is a little like the Eagles' "Hotel California," and the future of the hardware business.
This interview has been edited and condensed for clarity.
How are you feeling? I've talked about coronavirus and COVID-19 with dozens of CEOs over the past five months, and you're the first one I've talked to who has been directly affected by it.
I'm doing great. Thank you for asking. I guess I got a mild case.
I started on a Friday, started feeling a little bit tired, and then by Sunday had some fever. On Monday I tested positive and obviously went into quarantine. But the next Monday or Tuesday, a week [or] eight days after I tested positive, I was feeling good. You know, throughout the period, you're always tired. You want to sleep the whole time, you have a fever on and off. But I never had any respiratory [problems] or complications.
And from there on, there's been no problems for the last month or so.
Does your experience change the way that you've thought about managing the company during this period of time, in terms of working from home or the way offices will be structured at HPE?
I don't believe that 50% of employees will ever come back to the office to do their jobs, thinking about [it as] coming to a desk, dock your PC, and do your daily job. What we think we're going to have is a more collaborative, innovative set of centers, where we will have space for people to come, but not to do the actual daily job, more to collaborate and do meetings and planning and obviously to drive the innovation, which is very important in the context of the culture, the fabric of a company.
When it comes to GreenLake, is the idea behind that to basically transition current compute customers, as they digitally transform and need some help doing that?
What we see is a market-based transition from a capex to a consumption-based model. And the consumption-based model has several elements: a subscription base, where you subscribe to the software to deliver some sort of service behind it, or a utility-based model, which means you pay only for what you consume. The public cloud experience brought that kind of experience to the market, where it starts by renting infrastructure, and then paying only for what you consume, whether it's per virtual machine, or per container, or whatever for memory or for storage.
The problem is the low-hanging fruit of that, meaning the workloads and applications that were able to move to the public cloud, has already [been] taken. Now we're talking about production-scale workloads where the data aspect is pretty significant. Data has tremendous gravity. And when you look at the cost model of what you pay in a public cloud versus what you pay on prem, you realize that it is no longer the cost of compute … it is about the cost of egressing data back and forth.
Many applications need the data to be close enough because of latency, because of compliance, or because of the type of workload you're talking about: an example of this can be AI and machine learning, big-data analytics, simulation, modeling and so forth, where the data needs to be close to the compute. Otherwise, you're paying an enormous bill by egressing data back and forth. When you check in to the public cloud, it is like checking in to Hotel California: You check in and never check out, because that cost is ginormous.
To me, cloud is not a destination, it is an experience. With GreenLake, we bring that cloud experience for all your apps and data. If you have it in your data center, we can automate everything, and we can cloudify everything, and you can [also] pay in your data center for what you consume. So you don't have to invest in massive amounts of capex, where you buy computer storage, then you have your people putting it together, automating everything and then spend an enormous amount of opex running it.
We can do all of that through GreenLake, where we bring the compute and storage we have detected for that workload, so we can take cost out.
You said you had 1,000 customers who are signed up for it. What percentage of that 1,000 were previous HPE customers?
I don't have the exact number, but in general, obviously, many, many customers came through the installed base that we have. And a lot of this is a combination of both legacy workloads and new workloads. Legacy workloads, because it's hard to move them, but they still wanted that as-a-service experience. And then new workloads, an example of this is machine-learning ops as-a-service.
Earlier this year, I talked to Oxide Computer, a company working on next-generation server hardware. It's very, very early for them, but it got me thinking: Do you see the basic server, the compute part of your business, that I believe is still the largest part of your business, do you think there is innovation left in that segment?
There is no question that general-purpose compute has been commoditized by the Intels and AMDs of the world. I've been driving that roadmap for a number of years, particularly with Moore's law. And now, obviously, you run into issues with Moore's law, as we saw a couple of weeks ago, where you go in fabs from 14 [nanometers] to 10, to seven, maybe to four going forward. But that's not easy, and Intel is obviously having some challenges in the short term.
The problem is, you start to run into some issues, because the more you put closer to the CPU, the more you close that ecosystem. And there's less innovation [possible] around what the future could be.
If you think about it, we had a mainframe, then we went to the PC client/server, then the internet came along, and we connected the offices. And now we live in a mobile-first, cloud-first approach where your smartphone is your terminal, instead of that green terminal that some of us used in the past.
What I see now is the cloud moving closer to where the action is. To me, the action is where we live, where we work, in hospitals and manufacturing floors, in cars or autonomous vehicles.
It is easier to move the compute to the data, than the data to the compute. It's just the physics, the physics and cost. The next computer architectures will be much more data-driven than CPU-driven.