Hark Pro picks Nebius as its inference partner

Tomislav Marinovic @DrTomsLensOriginally posted 2026.10.14

Open original on X

$NBIS is now the inference partner for Hark Pro, a new AI product from Hark, founded by Brett Adcock, the man behind Figure AI. This is a pretty interesting customer for Nebius, for several reasons. First, what is Hark Pro? Think of it as a personal AI assistant, similar to Meta’s Muse, that helps people get things done. Agentic applications like this can generate substantial recurring inference demand because, put very simply, even a single task can involve several steps with AI models basically talking to each other until they figure out what to do. So the more people use these products, the more compute is needed to run them. And this is where I think things get interesting for Nebius. Nebius doesn’t need to build the next great consumer or enterprise AI application itself. It can provide the infrastructure for hundreds or potentially thousands of them. And as those applications grow, Nebius can grow with them, without having to compete for the end user. And this is not just about Hark. Look at the strategic partnership with Palantir, where Nebius is bringing its inference infrastructure into Palantir’s enterprise perimeter as a preferred AI infrastructure partner. Palantir already has the applications, customer relationships, and operational workflows. Nebius now provides the infrastructure to run AI inference within that ecosystem. And Nebius has been building toward this for a while. The Token Factory buildout, acquisitions of Eigen AI and Inferize, and integration of Clarifai’s technology and engineering talent are all part of making inference faster, cheaper, and more reliable at scale. This also fits Nebius’ broader agent-native cloud strategy. Rather than just renting out GPUs, Nebius is building a vertically integrated AI cloud that supports customers from training and inference to deploying AI applications and agents. At the same time, partnerships and distribution channels like TD SYNNEX help bring that infrastructure and software to a much wider customer base. The idea is not just to wait for demand, but to make Nebius an increasingly natural choice for companies building and scaling AI products. Now, there’s another interesting angle here. Brett Adcock also founded Figure AI, one of the most interesting American pure plays in humanoid robotics. And Hark isn’t completely disconnected from Figure. Its models have reportedly been trained on Figure’s robots, so there’s already some technological overlap between the two businesses. Figure recently announced a massive compute partnership with Nscale. But we’re also seeing AI companies increasingly spread their compute across different providers, often using different infrastructure for training and inference. I’m not going to speculate about whether Figure might eventually do the same and move some of its workloads to Nebius. There’s no announced partnership between the two. But with Nebius now providing inference for another company founded by Adcock, it’s certainly something worth watching. And here’s the bigger picture. Whether it’s agentic AI, enterprise applications, physical AI and robotics, or AI-driven biology, all these fields are moving toward more sophisticated models, more autonomous workflows, and potentially enormous inference demand. So Nebius doesn’t need to predict which application, robot, or AI-powered scientific platform will win. It needs to build an infrastructure and software platform that many of them can use. And the more successful AI applications there are, the more opportunities Nebius has to serve them. In a way, much like Palantir is positioning itself for the explosion of AI applications, Nebius is positioning itself for the explosion of inference demand those applications will generate.

This post quotes another post: x.com/nebiusai/status/2110101424801796430