If there is one takeaway from the Vast Forward 2026 conference (which I can only assume was named as a pun on fast forward), it is that the scale of Vast’s reach is, well, Vast.

Finding its origins as a storage solution provider when the company was founded in 2016, today Vast has evolved beyond storage alone, and through its AI OS (operating system) offering, is the foundational basis of many neocloud providers of the past few years.

At DCD we focus on the hardware and digital infrastructure side of things, and it is rare that we get a look in at the other layers that go into a cloud platform, high-performance computing cluster, or enterprise IT operations.

This week in Salt Lake City, we have somewhat deviated from the norm.

During CEO and cofounder Renen Hallak’s keynote, he took a look back at the company’s origins. At the time of its founding, Google had just acquired DeepMind. OpenAI was founded, but ChatGPT had not yet exploded.

Over the next few years, Vast continued expanding its offering, and by the time that the AI “boom” really had happened, and the neoclouds were swarming like flies, the company had an offering that could enable GPU infrastructure to be delivered as a cloud service, tightly integrating data, storage, and GPU compute.

Speaking during the prebriefing, Hallak noted that the company has continued to triple its revenue year-on-year, but that even he cannot deny that they have been “tremendously lucky” with how the chips have fallen and the choices Vast made in those early years to be ready to catch them. A choice metaphor was used to describe this, comparing Vast to a mosquito set free in a certain kind of “colony.”

Co-founder Jeff Denworth backed this up: “At that moment, I remember Renen saying, ‘Whatever we are doing, this is what matters most.’ ChatGPT had just come out, and we realized we had the only multi-tenant data platform in the market, so we told the whole go-to-market team to execute.”

At this point in time, Vast’s AI OS platform is being used by the likes of CoreWeave, Lambda, Nscale, Buzz HPC, Fluidstack, GMI Cloud, Firmus, Akamai Technologies, and Crusoe - and these are just a few of the names we know of. In total, the company has at least 700 customers, and many of these are also enterprises, and HPC centers such as TACC and those operated by academic institutions.

Of the many announcements made during the Vast event, two particularly stood out to me. Naturally, it was the solutions that danced closer to the hardware and physical infrastructure side of things.

The latest of many collaborations, Vast and Nvidia have been working together on a solution that sees the Vast Data platform directly integrated into Nvidia GPU-powered servers.

Dubbed the Vast CNode-X, the solution has been developed with Nvidia and will bring high-performance storage services to Nvidia GPU clusters, as well as make the OS directly available on the hardware. According to Vast, this will be optimized for AI pipelines, high-performance analytics, vector search, RAG systems, and agentic workloads. The servers are being developed by Cisco, HPE, Supermicro, and other OEMs.

This isn’t the first such effort; just earlier this year, the company announced it had made its OS available on Nvidia BlueField-4 DPUs, which, in this case, enabled AI context to remain confined to local GPU memory.

According to John Mao, VP of global business development at Vast, the CNode-X solution is mainly targeting enterprises. “It’s really an enterprise play, where they want something that’s more turnkey, where you can roll in a rack or two racks and get something running,” he tells DCD. Afterall, the cloud customers already have their own GPUs deployed - they don’t necessarily need to buy more with the platform integrated at a hardware level.

Mao added that the company ultimately plans to expand this beyond Nvidia GPUs, though did not specify which chips the company is targeting beyond the acknowledgement that Vast works with basically all of them (AMD, Arm, hyperscalers with proprietary chips, Cerebras, etc). “We are compute agnostic,” he explains, “in the sense that we can host services on any accelerator. But with CNode-X, we had to take that incrementally. Our operating system literally runs on the box like an appliance… and obviously it will take us some time to enable that capability on other GPUs, but going forward, there will be additional ‘flavors’ of that as well.”

Another development that tickled my fancy was the new platform, “Polaris,” which enables the provisioning, operating, and orchestration of distributed AI infrastructure regardless of whether it's in the public cloud, neocloud, or on-prem.

Building on the existing “DataSpace” offering that can connect clusters across significant distances, having previously demonstrated the capability as far as 10,000km (6,213 miles), linking one in the US with another in Japan, Vast states that Polaris “governs how those clusters are deployed and lifecycle-managed across cloud and hybrid environments.” This includes multi-site and multi-cluster deployments.

With data continuing to grow exponentially in the context of AI, it seems natural that the deployments will continue to be further distributed. We are also seeing this in the increasingly sprawling clusters and infrastructure being laid out by hyperscalers, including AWS with Project Rainier, Microsoft with Fairwater, and Oracle with its buildout for OpenAI.

With this in mind, Vast is clearly keen to ensure that its tendrils will creep ever further. After all, co-founder Denworth said (in jest): “The focus is on world domination.”