
Share
Agents don't provision like humans do, so Cockroach Labs is rethinking how database capacity gets paid for. Continuum turns a company's entire database estate into one elastic, usage-priced system instead of a pile of overprovisioned boxes.
Cockroach Labs just shipped something that attacks a problem most database teams have quietly accepted as the cost of doing business: you provision for peak load, and then you pay for that peak capacity whether you ever hit it or not. The new product is called Cockroach Continuum, and it treats an enterprise's entire database estate, every database across every application, as a single elastic system rather than a pile of individually sized instances.
The timing isn't subtle. AI agents don't behave like human developers. They run continuously, spin up new applications and databases at a pace no ops team can staff for, and don't care about your capacity planning spreadsheet. Gartner's number here is the one worth sitting with: AI agents per Fortune 500 company could jump from fewer than 15 today to 150,000 by 2028. That's not a gradual ramp you provision ahead of. That's a step function that breaks the old model.
"CockroachDB was built for the day everything goes wrong at once," said Spencer Kimball, Cockroach Labs co-founder and CEO. "That used to mean an outage, a traffic spike, or just bad luck. With thousands of agents acting at once, it's becoming the norm. The database industry has always asked customers to guess their peak and pay for it whether they hit it or not. Agents don't work that way, and neither should the bill."
The core move here is architectural: instead of managing each database as its own isolated unit with its own provisioned headroom, Continuum runs the whole estate as one distributed system that shares capacity underneath. Five capabilities make that work:
Worth noting this all sits on top of CockroachDB's existing engine, which has a decade of production runtime behind it. Cockroach Labs is explicit that the resilience and correctness guarantees the database is known for don't get traded away for the new elasticity. That's the harder engineering problem, honestly. Pooling capacity across thousands of tenants while keeping consistency guarantees intact isn't a rebrand, it's a distributed systems challenge that depends entirely on the underlying engine already being solid.

The business case leans on a real supply problem, not just a hypothetical one. Compute is getting scarce and expensive at the exact moment AI workloads are demanding more of it. The company points to GitHub as a concrete data point: GitHub planned for a 10x increase in compute needs last October, then found itself needing 30x by February as AI-assisted development took off faster than anyone modeled. Hyperscaler capex has crossed $600 billion in 2026, chasing grid power and data center capacity that takes years to build out, not months. In that environment, infrastructure sitting idle "just in case" is a much more expensive habit than it used to be.
This context matters for anyone evaluating infrastructure spend right now. Database licensing and provisioning models that made sense when compute was cheap and abundant start to look like unnecessary overhead when the underlying resource itself is constrained. Continuum is Cockroach Labs' bet that enterprises will increasingly want their database layer to behave more like a utility, metered and elastic, rather than a fleet of fixed-size boxes they have to keep guessing the size of.
Alongside the product, Cockroach Labs also rolled out a new visual identity. The company is moving its branding away from the cockroach-as-survivor framing toward something built around adaptability, using a modular "biopixel" design language meant to signal components that can flex independently while reinforcing the whole system. It's a small detail relative to the technical announcement, but it tracks with the broader pitch: a company trying to reposition itself as infrastructure for an agentic era rather than just a resilient SQL database.
Continuum is a genuine architectural shift, not just a pricing tweak slapped on top of the existing product. Virtualizing databases across a shared pool and decoupling storage from compute via Plenum are the kind of changes that actually let consumption-based billing work honestly, rather than becoming a marketing wrapper around the same provisioned capacity.
The timing argument is credible. Agent-driven workloads really do break human-scale provisioning assumptions, and the GitHub compute numbers back that up with a concrete, recent example rather than a vague forecast.
The open question is execution at scale. Pooling capacity across thousands of isolated tenants while preserving the correctness and resilience CockroachDB is known for is a nontrivial distributed systems problem, and auto-scaling compute, arguably the piece that matters most for realizing the cost savings, is still listed as "coming soon" rather than shipped. Enterprises evaluating this will want to watch how that capability lands before fully betting their estate on the utility-pricing model.
Tags
Original Sources
Cockroach Labs Introduces Continuum to Unify Enterprise Database Operations - BigDATAwire
↗ https://www.hpcwire.com/bigdatawire/this-just-in/cockroach-labs-introduces-continuum-to-unify-enterprise-database-operations
About the author
Kai built ML infrastructure at a Bay Area startup before developing an obsession with transformer architectures and inference optimisation that eventually pulled him out of product work entirely. A stint at a compute research lab sharpened his instinct for what actually matters in a model release versus what is marketing. He writes from the inside — from the perspective of someone who has debugged the systems he is describing at three in the morning. He is allergic to hype and instinctively drawn to the unglamorous plumbing questions that everyone else skips over.
More from The Engineer →This Week's Edition
18 November 2026
29 articles
Related Articles

Elsevier Gives ClinicalKey AI Full EHR Access, Betting on Patient Context Over Chatbots
Products & Applications · 6 min

BMO Claims Quantum Utility Milestone in AI Feature Selection, Using 100x Less Data
Products & Applications · 5 min

National Academies Report Calls for Faster, Tougher Retraction Standards
Models & Research · 5 min
Related Articles

Elsevier Gives ClinicalKey AI Full EHR Access, Betting on Patient Context Over Chatbots
Products & Applications · 6 min

BMO Claims Quantum Utility Milestone in AI Feature Selection, Using 100x Less Data
Products & Applications · 5 min

National Academies Report Calls for Faster, Tougher Retraction Standards
Models & Research · 5 min
More Stories
© 2026 Cedar & Bloom. All rights reserved.