ROBOCAP

$NBIS MANAGED AGENTS PLATFORM: The transition to Agents-as-a-Service Lately, @romanchernin has teased a new Agentic AI software offering on Nebius Cloud. Both on the earnings call and etn. interview (video in the tweet below). This platform would sit on top of Token Factory, and would capture even more margins. But first, what is Managed Agents? And what would this AaaS platform look like? A managed agents platform runs your AI agents for you, handling the models, GPUs, tools, memory, and guardrails so you only write the prompt, and the platform takes care of the rest. For example, an enterprise customer like Revolut could prompt: "Run a New Year's Eve campaign for our top customers." That one sentence kicks off a multi-step agent job. The agent figures out who qualifies, generates unique gift codes, drafts localized messages, creates the graphics, runs everything past compliance, and ships the campaign. It runs all of these steps on Token Factory inference. AWS, Google Cloud and Claude already offer Managed Agents, if Nebius moves quickly they can be among the first platforms to offer this service. Nebius is already partnered with DataRobot on this layer (datarobot.com/solutions/part…) but Nebius only provides tokens while DataRobot earns higher margins by selling agent runs to enterprises. (Image 1) That's why I believe Nebius will develop (or acquire) its own Managed Agents layer, to earn superior margins and move up the stack. Because right now Nebius only sells tokens, but selling agent runs is much more profitable. Possibility A (more likely): Nebius can build its own Managed Agents platform and will ship it in 2026/27. The talent pool at Nebius is more than enough to tackle this task (world-class engineers from EigenAI and Clarifai + people already at Nebius + maybe other surgical small acquisitions) Possibility B: Nebius can purchase DataRobot's Agentic AI software IP and hire its core engineers (like they did with Clarifai). This would fit $NBIS like a glove. Nebius runs the NVIDIA hardware, generates tokens, optimizes models and squeezes efficiency through EigenAI and Clarifai, runs agentic search on Tavily, human validation on Toloka Tandem, and sells Agentic runs through DataRobot's enterprise-grade, already proven orchestration platform. Latest funding round implies 6B$ valuation in peak 2021 bubble. Secondary private markets today imply around 3.5B$. 1.8-2.2B$ can be a fair ballpark estimate for IP carve-out and core talent. I don't expect a full acquisition because of the bloated 800+ headcount and declining legacy ML products. However, it is an expensive acquisition. A 70% stock 30% cash deal (similar to EigenAI) can be dilutive and pressure cash reserves. Bottom Line: $NBIS wants to move up the stack to achieve better margins and go directly to enterprises rather than giving up margins to middle-men. They achieved this with launching Token Factory, and a new Agentic AI platform is the next step. The AI infrastructure industry is fast and evolving. Lower layers (raw GPU rentals) are commoditizing and margins are decreasing. That's why it is critical to move up the stack, where you are closer to the end-customer and value creation is rewarded with superior margins. (remember Jensen's 5 layer AI cake, Image 2) $IREN's Mirantis explained this dynamic here: xitter.gq/MirantisIT/status/2055… By building the Agentic AI platform, Nebius can thrive at all 5 layers: Energy - Nebius owns 75% of their capacity in 2027 Chips - Nebius has NVIDIA Cloud Partner status, which enables earlier access to latest chips Infrastructure - Nebius builds its own server racks, saves 10-15% margins Models/Inference - Token Factory Applications - New Agentic AI Enterprise Platform & $NBIS's already stacked Physical AI software layers ( )