The acquisition adds systems-engineering talent to Nebius’s inference team and follows earlier Token Factory additions from Eigen AI and Clarifai.
- Inferize’s technology is designed to reduce cold-start delays that can leave GPUs idle during AI inference.
- The technology will be integrated into Nebius Token Factory, its managed production-inference platform.
- Higher GPU utilization could help Nebius serve more AI workloads without maintaining as much unused capacity.
Nebius (NBIS) shares gained in early morning trade on Thursday after the company announced it was adding another piece to its AI infrastructure stack, acquiring inference optimization company Inferize to help customers launch and scale large AI models faster while getting more useful work from each GPU.
The technology and Inferize team will join Nebius Token Factory, the company’s managed inference platform for production AI. The deal terms were not disclosed.
NBIS stock gained as much as 1.9% in pre-market trade. On Stocktwits, retail sentiment around the Nvidia (NVDA)-backed company shifted to ‘bearish’ from ‘neutral’ territory over the past day.
Nebius Targets AI’s ‘Idle GPU Tax’
The company said its acquisition addressed a growing challenge in AI infrastructure, where GPUs sit idle while models load, new capacity comes online or model weights are updated.
Those delays, known as cold starts, become increasingly expensive when AI workloads operate at large scale. Cloud providers need to keep spare GPU capacity available to handle sudden demand, even when those GPUs are not actively processing requests.
Inferize’s technology is designed to make that capacity more elastic, allowing Nebius to bring additional inference capacity online faster and more closely match GPU availability with actual demand.
That could improve utilization and lower the infrastructure cost associated with serving each AI token.
Nebius Builds Out Token Factory
The acquisition also expands Nebius’s broader push to build a production-grade inference platform.
Nebius said Inferize will work across Token Factory, extending beyond the initial integration. The company previously added model, kernel and system-level optimization through Eigen AI, while Clarifai’s team and technology brought system-level inference and compute orchestration.
Inferize CEO Guy Bortnikov said the company’s technology is designed to reduce the need to keep spare GPUs running purely to remain ready for demand. Inferize was founded in January 2026 and built a working prototype within three months. Its engineers will now join Nebius’s inference team.
Read also: HPE Stock Price Target Hiked By Barclays On Networking Growth Outlook Jump – Analyst Sees 23% Upside Potential
For updates and corrections, email newsroom[at]stocktwits[dot]com.<