Databricks just spent $1 billion to acquire Neon, and the most revealing detail isn't the price tag. It's this: 80% of databases on Neon's platform are now provisioned by AI agents, not humans.
That's not a forecast. That's what's already happening.
80% of Neon's databases are already provisioned by AI agents, not humans—revealing how agent workloads are becoming the primary consumer faster than most infrastructure was designed to handle.
We run web agents at production scale, and this pattern matches what we're seeing. Agents have become the primary consumers of infrastructure. Systems built for human operators hit a wall when that flip happens.
When Machine Speed Becomes the Baseline
Traditional databases assumed humans would provision them. A developer takes a few minutes to spin up an instance. Fine when your customer is human.
Neon provisions a fully isolated Postgres instance in 500 milliseconds. Agents demand it. A web agent monitoring hotel inventory creates databases to cache site structures that change hourly, tears them down when sites shift layouts, spins up new ones when regional variations appear. Databases optimized for persistence can't handle infrastructure that's fundamentally ephemeral by design. An agent optimizing supply chains generates more database queries in an hour than analyst teams produce in a week.
"The $100-billion-plus database market is facing unprecedented disruption."
— Ali Ghodsi, Databricks CEO
Agents expose architectural assumptions that no longer hold—assumptions about persistence, provisioning speed, and resource consumption patterns.
Why Architecture Beats Features
Neon's scale-to-zero architecture means thousands of ephemeral databases only consume resources when actually running. Traditional databases charge for always-on capacity because they assumed persistence. Agents create databases programmatically and tear them down just as fast. The economics invert.
Every system agents touch at production volume shows similar stress:
- APIs designed for human-paced requests struggle with agent concurrency
- Authentication systems built around session management break when agents need programmatic access at scale
- Data pipelines optimized for batch processing can't keep up with agents adjusting pricing across thousands of products based on what competitors changed in the last hour, not yesterday's batch run
Infrastructure experts predict 2026 will force enterprises to confront scaling AI systems. Infrastructure maturity has become the bottleneck.
The Strategic Split
The $1 billion Neon acquisition signals where value accumulates: in infrastructure that treats agents as primary consumers from the ground up.
Infrastructure designed for human operators carries architectural assumptions that features alone can't overcome. Neon's compute-storage separation, instant branching, and sub-second provisioning reflect architectural choices that only make sense when agents are your primary customer.
Eighty percent of Neon's databases already come from agents. As agent workloads become dominant, what becomes operationally feasible changes. Agents that can provision databases in milliseconds instead of minutes can explore solution spaces that weren't viable before: monitoring markets that shift too fast for human-paced infrastructure, testing strategies across thousands of scenarios simultaneously, adapting to web surfaces that change hourly.
The strategic choice for enterprises: build on infrastructure designed for yesterday's human operators, or infrastructure architected for agent workloads that are already here.
-
Multi-agent orchestration surge: Gartner reported a 1,445% increase in multi-agent system inquiries from Q1 2024 to Q2 2025, signaling how organizations are shifting from single large LLMs to coordinated specialist agents.
-
Database-per-customer economics: Neon's Agent Plan enables platforms to spawn thousands of isolated databases through database branching, making database-per-customer architectures economically viable at scale.
-
Infrastructure cost inversion: After the Databricks acquisition, Neon reduced storage pricing by 80% and compute costs by 15-25%, reflecting how agent-native infrastructure fundamentally changes database economics.
-
Agentic AI Foundation formation: The Linux Foundation launched the Agentic AI Foundation with Anthropic contributing Model Context Protocol, establishing open standards for agent interoperability as infrastructure matures.