NVIDIA DGX Spark: AI supercomputer in the office

Small box, big deal

Sometimes, a lot of the future fits into a small box. At our last DevTrain, our founder Tibor Csizmadia brought exactly that, and we unboxed it together. Inside: the NVIDIA DGX Spark. It might sound unassuming at first, but it is actually an AI supercomputer of the kind that was, until recently, reserved for large data centers. Only this one is sized to fit comfortably on a desk.

The moment the whole team held the Spark up together was more than just a nice photo op. It was a statement: We are serious about local AI, and we are getting started right now.

‍

‍

What makes the DGX Spark special

Powered by the NVIDIA GB10 Grace Blackwell Superchip, the DGX Spark is a complete platform for local AI development. What sets the device apart are performance metrics that, until recently, would have sounded like they belonged in a server room:

• 1 PetaFLOP of AI performance—that is, one quadrillion computing operations per second (at FP4 precision).
• 128 GB of unified system memory—shared between the CPU and GPU.
• Models with up to 200 billion parameters—run locally, directly from your desktop.
• NVIDIA AI software stack—with frameworks and pre-trained models, pre-installed and ready to go.

And it gets even bigger: Two DGX Spark units can be linked via NVIDIA’s ConnectX networking to run models with up to 405 billion parameters. The upgrade path is already built in.

‍

The real leap becomes clear in comparison

To put it into perspective: In the past, you would have needed an entire server room for comparable AI computing power, including cooling, cabling, and ongoing operating costs. Today, that same power sits quietly and compactly on a desk, without anyone needing to build a dedicated air conditioning system for it.

This is exactly why the DGX Spark is currently getting so much attention. This class of AI computing power is no longer just available to large corporations, but to small and medium-sized enterprises as well.

‍

Why local? The benefits for our customers

For us and the companies we work with, the decisive factor is not just the computing power, but the location. Running AI directly within your own infrastructure noticeably changes the landscape:

• Data stays in-house.Sensitive information does not need to leave your own network. This makes a huge difference for data protection and compliance.
• No cloud dependency.No ongoing token counters, no surprises in operating costs, and no outsourcing of value creation.
• No vendor lock-in.Open models and an open software stack mean freedom of choice instead of being tied to a single provider.
• Full control.From prototype to production, the solution stays where it belongs: with the company itself.

For many medium-sized companies, this combination of powerful, secure, and independent in-house AI was simply not realistic until now. The DGX Spark brings it within reach. We see it as our mission to make this path viable for our customers: from the initial idea and proof of concept to the final solution.

‍

‍

What we are planning to do with it

For us at Devware, the DGX Spark is more than just an exciting acquisition. It is the starting signal for our AI Lab. The plan is clear: test thoroughly first, then see where the journey leads. In concrete terms, this means learning, building, and testing: creating our own AI agents and models, developed and fine-tuned within our own infrastructure.

This environment will serve as the foundation for an R&D space where AI solutions for our customers are born. Local. Secure. Under conditions that were not possible just a few years ago.

The possibilities this opens up are enormous, and we are only at the beginning. In one of our next posts, we will take a closer look at the business perspective: what local AI means for our customers in concrete terms.

Until then: We think ahead. We develop. We support.