5 min

You may have never heard of Vultr. Still, this company is one to keep an eye on. It has built an interesting range of stacks, services and technologies. The goal? To get the most out of the underlying technology at the lowest cost and get the whole world working with GenAI. We spoke to Kevin Cochrane, Vultr’s CMO, about these big ambitions in a challenging market.

Vultr has been around since 1999, but was a hosting company until 2014. That doesn’t sound very interesting, but that background has brought the company quite a bit. It hosted the online multiplayer game Call of Duty, which taught the people working there a thing or two about optimizing performance. The team behind Vultr built a control plane for this, allowing it to manage virtual machines and bare metal servers in nine data centers.

Vultr has done everything from the beginning in a cost-effective way, with as few people as possible, Cochrane points out. That is still the basis of Vultr’s philosophy. “We are a group of nerds who want to optimize everything,” he sums up. Having the lowest latency compared to competitors is a matter of pride, he cites as an example. The founder is still the sole owner and not a dime of investment from outsiders enters the company. That is quite extraordinary these days. It offers Vultr the opportunity to establish a completely unique business model that is or can be interesting for a large target group.

Vultr’s roots: the cloud-native stack

Vultr basically consists of two stacks, a cloud-native stack and an AI/ML stack. With the cloud-native stack, the company wants to enable things like running cloud workloads, databases and managed Kubernetes for customers. The latter in particular is very popular at the moment. It achieves all of this by sending pre-built servers around the world. It’s literally the whole world, too, we hear from Cochrane: “We operate on all six continents.” Globally, Vultr has availability in 32 regions.

On the servers it sends all over the world, which it builds in partnership with Dell Technologies, is the control plane that Vultr has developed over the past 25 years. As already indicated, it’s all about the optimal combination of performance and cost efficiency when it comes to this control plane. Cochrane talks about maximum performance at the lowest price.

This kind of statement always makes us a bit suspicious. After all, compromises must always be made, right? This will undoubtedly also be true of Vultr to some extent, but if you really want to, you can get very far in this regard. Compared to hyperscalers, Vultr is “up to 90 percent cheaper for core workloads,” Cochrane points out. He also immediately adds that Vultr still makes a good living out of what they do. This then also gives pause for thought about the premium customers pay at hyperscalers, but that aside. The operational model of a hyperscaler will no doubt be a lot less lean and it will place much more emphasis on making shareholders happy and other considerations like that.

AI/ML stack brings GenAI closer to the masses

In addition to the cloud-native stack, Vultr also has the AI/ML stack. As an Nvidia Elite partner (the highest status), the company has early access to that company’s new GPUs. Cochrane claims Vultr was one of the first allowed and able to launch the Grace Hopper GH200 superchip. “We are in the same pipeline as the hyperscalers,” he adds.

An important part of the AI/ML stack is the GPU stack that Vultr has built. This is a crucial part of the offering. GPUs are very complex, especially when they are brand new. Vultr has built a managed service around this. So customers theoretically don’t have to worry about the underlying stack and can build applications and have them leverage the capabilities of the GPUs. “They don’t have to worry about updates breaking a local cluster,” he gives as an example.

Vultr has also further extended its managed Kubernetes service from its cloud-native stack to the GPU stack. In doing so, the company is killing several birds with one stone. “Kubernetes on its own is challenging enough, but when combined with GPUs it becomes even more complex,” in Cochrane’s words.

Cloud-native GenAI around the world

We now know what Vultr does globally. If all the claims about cost and available hardware, software and features hold true, this is definitely a company to watch if your organization is looking into advancing its infrastructure. Especially if you don’t necessarily want to be squeezed for all you have by the hyperscalers. Cochrane states that Vultr is an excellent alternative for customers. “Organizations can run 90 percent of their workloads at Vultr,” he points out when we ask him about the role the company can play in the hybrid multicloud strategy that more and more organizations have.

The two separate stacks are quite an interesting story in themselves as far as we are concerned. There is more, however, we hear from Cochrane. Namely, Vultr’s goal is to “offer cloud-native generative AI around the world.” In other words, Vultr wants to combine the two stacks it has and offer them as a complete stack. It wants to become the platform on which organizations develop GenAI applications.

According to Cochrane, Vultr is the only one that can tie together a best-of-breed cloud-native stack and a best-of-breed AI/ML stack. “Even the hyperscalers can’t do this. We are the only one that can do this in 32 regions simultaneously,” Cochrane points out. That also has to do with the fact that this goal of Vultr is not necessarily the goal of the hyperscalers. Of course, if they wanted to, they could do it, but they have other priorities. But even then, it’s still interesting to learn that a relatively small company like Vultr can offer this.

Cloud inferencing

In summary, Vultr offers the 32 regions, the two stacks and the offering that results when they merge those. That is still not all, however. That is, Vultr recently announced that it is introducing its own cloud inferencing service. This new component will play a crucial role in Vultr’s ambition to enable inferencing anywhere in the world. Inferencing is what it’s all about in distributed, hybrid environments. Training models take place at central locations, but inferencing must take place close to the user and the application. So for that, Vultr has now come up with the cloud inferencing service.

Apart from the practicality of the cloud inferencing service, the way Vultr puts it together is also interesting. It does so by reusing older chips/GPUs in CDN servers. They may not be top of the line anymore, that doesn’t mean they’re useless. In this way, Vultr can create a global network where customers can have different types of environments with corresponding price tags, based on what the wishes and requirements are. That sounds very interesting as far as we are concerned and is also definitely something that will be needed once (Gen)AI is really deeply embedded in organizations. With their focus on performance and price, Vultr seems well positioned to play a significant role in this.

Also read: Vultr introduces global Content Delivery Network (CDN) platform