發生了什麼事
Nebius Group N.V. announced the acquisition of Inferize, a startup that builds technology to cut AI model cold‑start , and will integrate its team and software into the Nebius Token Factory stack.
According to a TechGraph report, Nebius Group N.V. (Nasdaq: NBIS) has completed the acquisition of Inferize, a startup founded in January 2026 that focuses on optimisation. The deal adds Inferize’s technology and engineering team to Nebius’s Token Factory, the company’s AI‑cloud platform for launching and scaling large models.
Inferize’s core offering targets the "cold‑start" problem – the time required for a model to load onto GPUs before it can serve requests. These delays can leave GPUs idle during demand spikes, new instance launches, or weight updates in reinforcement‑learning loops, forcing platforms to keep spare capacity as a safety net.
Nebius CTO Danila Shtan is quoted as saying the acquisition will make Token Factory more responsive to demand changes and improve overall GPU utilisation. Inferize co‑founder and CEO Guy Bortnikov added that integrating the technology will help Nebius customers reduce the cost of idle GPUs.
The report notes that Inferize built a working prototype within three months of its founding and will now work across the Token Factory stack, beginning with integration of its cold‑start reduction technology.
為什麼這很重要
Cold‑start delays waste GPU capacity and raise costs for AI‑as‑a‑service providers; Inferize’s solution promises tighter scaling of resources, better token economics, and more responsive AI services for Nebius customers.
and GPU idle time are major cost drivers for AI‑cloud operators. By reducing cold‑start latency, Nebius can offer customers tighter scaling, potentially lowering the price per token and improving the economics of serving high‑throughput workloads.
Improved utilisation also benefits developers who run large language models or vision models on Nebius, as they can rely on more predictable performance without over‑provisioning hardware.
The acquisition signals Nebius’s broader strategy to build a vertically integrated stack, complementing earlier integrations with Eigen AI and Clarifai. This could position Nebius as a more competitive alternative to larger cloud providers that already offer inference‑optimisation services.
互動機制：它實際上是如何運作的
以互動方式探索這項發展背後的基礎技術。
Which component of an AI application is the machine-learning model itself?
接下來看什麼
The rollout of Inferize’s technology within Token Factory, pricing or service‑level changes for Nebius users, and competitive responses from other AI cloud providers.
The timeline for integrating Inferize’s technology into Token Factory, including any beta programs or early‑access releases.
Whether Nebius will adjust its pricing model or token‑economics to reflect the expected efficiency gains.
Reactions from existing Nebius customers and whether they adopt the new capabilities at scale.
Competitive moves from other AI‑cloud platforms that may introduce or accelerate their own ‑optimisation features in response.