QumulusAI, a Georgia-based provider of graphics processing unit-powered cloud infrastructure for artificial intelligence, is making moves to claim a leadership position in the emerging neocloud market today by naming Michael Maniscalco new chief executive officer.

Maniscalco (pictured), former chief technology officer of Applied Digital Corp., will bring his technical skills to an already deep team to bring enterprise-grade AI infrastructure to market.

What is a neocloud?

If you’re not familiar with the term, neoclouds are an emerging subset of cloud providers that offer bare-metal access to GPUs by focusing on AI and high-performance computing. The value-add of neoclouds is providing organizations with a GPU-centric infrastructure that delivers fast availability, reduced costs and greater network optimization for such demanding AI-related tasks as model training and inference.

Maniscalco has been hired to expand the company’s differentiated approach of owning the full stack – from energy and data centers to GPU-accelerated cloud services. I spoke to Maniscalco before the company’s announcement to get his thoughts on the neocloud market and the role he envisions QumulusAI playing in it.

The growing role of AI in the stack

As he considered joining QumulusAI, Maniscalco said the one thing he kept coming back to is how far up the stack AI is going. “When you get to the application layers, you get to the specific business workflows and what people are doing with those workflows, and how that is applicable to real-world business use cases,” he said. “I am bullish at the opportunity and the pace at which the AI tools are improving to enable that.”

After speaking with several companies in this market, he said, QumulusAI stood out because “they had most of the pieces. It was just a matter of getting the last few in place.”

Providing compute for ‘a big part of the world’

Maniscalco said hyperscalers are building out compute for the major AI labs. “The overwhelming majority of that is spoken for by the OpenAIs of the world, which leaves a big piece of the world trying to find compute and either having a hard time accessing the right architectures for the use cases and then balancing that with pricing,” he said.

According to Maniscalco, the demand for compute is outstripping supply and “growing faster than we can build and keep up.” But, he said, “If you can find a way to bring that supply to the demand faster, you’re in a fantastic position.”

He conceded that’s not a long-term strategy, but he has ideas for how QumulusAI can leverage the situation. “We’re going to look at quickly deploying chips to market through smaller pockets of compute, rather than trying to build gigawatt-scale data center campuses. Not only is it easier to find power in those smaller pockets, but it’s a lot faster to deploy if you can put it into a semi-modular type form factor, and the focus is just speed and cost,” Maniscalco said.

“Over time,’ he added, “we’ll start to look at how we balance that, so regardless of what you need from a quality of service or quality of output or latency or cost perspective or security, we can offer you, based on that geographic distribution, the right solution for your needs.”

The rise of neoclouds

QumulusAI is not the only company devising a plan to bring the differentiated capabilities of a neocloud provider to more organizations. Hyperscalers are good for the top 100 or 200 enterprises, but there are a couple of million other businesses with more bespoke requirements. The key to success for QumulusAI or any neocloud company is to bring a differentiated solution to that large group of potential customers.

Maniscalco said the starting point is to “get compute in the hands of the customers who want it as quickly as we can.” And the way to do it, he said, is to “layer on better user experiences and interfaces to meet their needs. Our team is spending a lot of time on that.”

Since all serious competitors in this space will have the infrastructure to get compute in the hands of customers quickly, the key is to enable customers to leverage resources quickly but in a way that’s consumable for developers and other that don’t speak data center jargon.

“Our team thinks it’s cost-effective compute,” he said. “That’s where the team is currently focused, and it involves a lot of capital and execution. Those two things are where we’re focused. Over time, we want to move into the next category, and that is finding various partners in the ecosystem that deliver user experiences tailored to the types of customers we want to go after that may not want to host the compute or deal with managing this infrastructure, especially at scale.”

Maniscalco added that “GPUs are very different from managing CPU workloads, and a lot of people just don’t want to get into that game. So we think we can go a long way through partnerships and service the customers, and grow as we like. And I think there’s room for that.”

How to grow the customer base

Maniscalco said QumulusAI is working with a lot of consulting businesses to position its solution as something they can scale up. And he said there are other verticals he and the company are excited about.

“One that’s hard to ignore is coding solutions and what we may be able to do there, because a lot of the improvements that are coming from that world are with deeper and longer reasoning,” he said. “The more you go and automate test cases to run against the code you just developed to verify it and then bug fix it, and kind of this back and forth, as you would in a pair programming or QA environment. The more you do that, the better quality code you seem to get out of the solution.

“So I think if you were to make an argument that these AI code development solutions are sooner than later, going to be as good as a software engineer, that’s a massive market opportunity that just requires tokens and more compute. That’s a pretty comfortable protocol to talk about.”

Not long ago, developers feared that AI was going to take their jobs. But in reality, companies can’t hire developers fast enough. If a lot of that initial AI development, or coding development, can be done by an AI, that can help companies actually keep up with the demand for developers and close that gap. Over the last year or so, a lot of that fear has gone away. That mindset opens the door for QumulusAI and other neoclouds to take advantage.

“The potential of these technologies to take away mundane and repetitive tasks, that’s what it’s about,” said Maniscalco. “To use a very specific developer example, I don’t know many developers who really love UX and testing. This is getting that out of the way. It allows them to build faster and have better, more enjoyable careers and daily task lists. So that’s a positive.”

Neocloud differentiation

There’s no shortage of neoclouds and alternatives offered by big cloud providers for enterprises to choose from. When enterprises go through the process of deciding which vendors to use, they should be focusing on specific issues so they wind up with the right partner before they commit a major investment. There’s a lot more to consider than just access to GPUs. What advice does Maniscalco have for those decision-makers?

“It’s about some level of flexibility, because you don’t often get that with some of the larger providers who invested a lot in infrastructure a few years ago and pigeonholed themselves into a specific architecture or specific stack,” he said. “As a smaller, emerging neocloud, we have some level of flexibility you’re not going to find, and I think that’s big for enterprises.”

Another area of differentiation he sees is trust. “Trust plays into privacy, security and politics of the enterprise,” he said. “Who do I trust with my data? Who may be a friendly competitor because they have cloud services, but they also compete with us in these other spaces?”

Security also plays a role here, as well as cost. “Cost is certainly a big one,” he said. “Not everybody needs a top-tier, fully redundant data center in Metro New York, to use an extreme example, that’s very expensive and scarce. Some people do, and I think there are opportunities for us, as we grow, to be that provider.”

Where does Maniscalco see his company going?

“Grow or die” may be a business cliché, but there’s a lot of truth in that, especially as technologies evolve at hyperspeed and companies that are on top today can find themselves also-rans in the blink of an eye. With that in mind, I asked Maniscalco about his growth plans for QumulusAI.

“We want to grow aggressively,” he said. “We’ve got the capital and the momentum behind us to do that, and the timing is perfect to try and get this in the hands of our consumers as quickly as possible. The more capital that enables us to do that, the better we’re going to do. What I like here is the fact that the market has shifted to more of a small-scale training or inference-type world.”

Growth outside the U.S. is also on the to-do list for the company. “We’ve got several conversations going with international partners,” he said. “For us as an emerging company in a high-growth space, it’s just a matter of finding the right time and the right people to do it with.”

Another pillar of QumulusAI’s current and future growth plan is its relationship with the leading AI chip maker, Nvidia Corp. The company already provides a wide range of Nvidia offerings, which Maniscalco doesn’t believe is going to change.

“What I do think it’s going to change — and what I hope we can help with — is I don’t believe the average AI developer or data scientist really cares all that much about GPU type or location,” he said. “What they care about is: Is it reliable? How quickly can I get a response and that information back? It comes down to that and maybe a few other vectors. So I’d love to find ways to extract away the compute and make it more developer-friendly and then talk in terms developers care about.”

What paths will QumulusAI and other neocloud companies ultimately take? Time will tell, but with new leadership from Maniscalco and the recent additions of Ryan DiRocco, formerly of Performive LLC, as chief technology officer, and Stephen Hunton, who was head of global social and content experience at IBM Corp., as chief marketing officer, QumulusAI has a strong team in place. Now it’s time to go execute.

Zeus Kerravala is a principal analyst at ZK Research, a division of Kerravala Consulting. He wrote this article for SiliconANGLE.

Photo: QumulusAI