PhoenixAI Inc., formerly known as CelerData, raised $80 million in Series B funding on June 11, 2026, to develop its artificial intelligence-native database. Sky9 Capital led the round, with participation from Atypical Ventures and Olive Technology Ventures, to scale governance for regulated industries as enterprise data demands shift toward agentic AI.
Agentic AI has moved from prototyping to production, fundamentally changing how databases are queried. Instead of small workloads running on virtual machines, swarms of AI agents now send thousands or millions of questions per second. Modern data stacks frequently strain under this concurrent load.
How analytical databases handle agent swarms
Transactional databases record individual operations like inserting rows or updating balances. They prioritize reliability and normalized data, making them poorly suited for the unstructured, conversational queries AI agents use. An agent asking for top customers by revenue growth over 90 days requires scanning millions of rows across multiple tables, a task optimized for analytical databases.
Analytical databases sacrifice write speed for fast, complex reads across massive datasets. They sit alongside transactional systems, which remain the primary system of record. PhoenixAI positions its technology as a layer on top of existing enterprise resource planning systems to help AI agents reason across live data faster. For professionals working in AI for IT & Development, this separation between the system of record and the system of insight dictates how new infrastructure must be architected.
"Most of today's analytical databases were architected for a world that no longer exists, where humans ran dashboards on flat tables and complexity was someone else's problem," said PhoenixAI President Rick Underwood. "When thousands of agents need to query, reason and act on petabytes of live data simultaneously - any question, simple or complex - the database is either the bottleneck or the breakthrough."
Technical architecture and market competition
PhoenixAI claims its rebuilt database delivers subsecond latency and high concurrency on live data. The company uses a "no pipeline" architecture, pulling fresh data directly from Kafka, an open-source event streaming platform. This allows agents to ingest updated information in seconds rather than waiting for traditional batch processing.
Competitors are adjusting their platforms to capture this market. Snowflake recently launched agentic features, Databricks is pushing real-time capabilities with Delta Live Tables, and ClickHouse Cloud has improved its concurrency handling. IT leaders evaluating these architectural shifts might review an AI Learning Path for IT Managers to assess integration strategies for real-time data layers.
Why this matters for IT and development professionals
Database bottlenecks will directly limit the scale of AI agent deployments in production environments. Development teams must evaluate whether their current analytical databases can handle thousands of concurrent, complex queries without degrading performance. Adopting or integrating agentic-ready data layers will be a prerequisite for scaling AI operations beyond isolated experiments.
Your membership also unlocks:
