Nokia and Databricks prove cloud-agnostic data platform for AI networks
Nokia and Databricks have completed a joint proof of concept (PoC) demonstrating a unified, substrate-agnostic data platform intended to underpin AI-driven autonomous networks. The two companies say the work confirms that telecoms operators can consolidate fragmented data environments and deploy real-time analytics at scale, without the need to recode workflows for each infrastructure target.
The PoC centred on a real-time performance management use case, simulating analytics ingestion at a throughput intended to match the demands of a tier-1 carrier. Engineering teams validated three core capabilities: data pipelines created once and redeployed across Databricks and an open-source stack comprising Apache Flink, Kafka, and Iceberg; a custom compiler that automatically translates abstract Python-based transformation logic into native formats such as Delta Live Tables or Flink SQL at deployment time; and an agentic data fabric layer that responds to natural language prompts to generate, validate, and deploy new data products without manual intervention.
The architecture in detail
The vendor-neutral design separates core data logic from platform-specific connectors, a pattern intended to limit lock-in and allow operators to swap underlying infrastructure as commercial arrangements change. Zero-copy sharing and query-time data products are also baked into the architecture, meaning derived metrics are computed on read rather than duplicated across systems, keeping cross-domain data consumption lightweight.
Oguz Sunay, CTO of AI and Autonomous Networks at Nokia, said the collaboration was "a big step as we work toward building the types of data foundations required for next-generation autonomous networks." Nevash Pillay, Global Head of Telecommunications Industry at Databricks, framed the problem as one of consistency: operators managing increasingly complex networks need a more coherent way to exploit the data those networks generate.
Market context
The PoC lands at a moment when network operators are under significant pressure to monetise their investments in 5G infrastructure, and AI-driven automation is widely positioned as the mechanism for doing so. The challenge described, hundreds of siloed operational and business support systems running incompatible data architectures, is well-documented in the industry and has given rise to a crowded vendor landscape. Ericsson, Huawei, and a range of cloud-native telco software specialists are pursuing similar data fabric and AI-ops propositions, while hyperscalers are pushing their own managed analytics stacks directly at carriers.
The cloud-agnostic angle is notable. Many operators are deliberately pursuing multi-cloud strategies to avoid the leverage that a single hyperscaler relationship can create, and the ability to deploy identical pipelines across AWS, Azure, Google Cloud, or private infrastructure without rework addresses a genuine procurement concern. Databricks' existing relationships with AT&T and other large carriers, cited in the company's own customer roster, give it credible distribution into this segment.
Standards and regulatory read-across
Autonomous network ambitions sit squarely within the scope of ETSI's Zero-touch Network and Service Management (ZSM) framework and the TM Forum's Open Digital Architecture, both of which define the interoperability and automation standards that large operators reference in procurement. A platform that can demonstrably align with these frameworks will carry more weight with operators than one that requires bespoke integration. Neither company cited conformance with these standards in the release, which is a gap the collaboration may need to address as it moves from PoC to commercial deployment.
Nokia and Databricks said they plan to continue joint development aimed at broadening autonomous network capabilities. No commercial product launch date, pricing structure, or named operator pilot was announced alongside the PoC results, so the next meaningful milestone will be a named carrier deployment at production scale.