2025 recap: Reflections on building data infrastructure for the AI era

Axiom has always been more than an observability tool. It’s a data platform for machine data. But 2025 was the year that distinction started to matter.

Observability tools help you analyze logs, metrics, and traces. The infrastructure underneath Axiom does something different: it scales to handle petabytes effortlessly, finds needles in hundred-billion-row haystacks in seconds, and provides the foundation for emerging applications. That foundation is what makes everything else in this post possible.

This is the story of what we built in 2025 and where we’re heading. The thread running through everything: Axiom’s data infrastructure is the best foundation for hyperscale enterprises. Observability is one proven application. AI engineering is the next. And Axiom itself is evolving with intelligence that will change how teams extract value from every byte.

Every capability we shipped this year stands on infrastructure we’ve built from scratch over many years. It’s not adapted from general-purpose systems. It’s purpose-built for machine data at scale.

Haydex addressed a problem that becomes critical at hyperscale. Finding specific events in petabytes of data means searching "haystacks" that expand by terabytes daily. Traditional approaches slow down linearly with volume. Haydex uses field-scoped filters that prune thousands of blocks with a single read, distributed indexing across Lambda functions, and aggressive optimization at every bottleneck. We see 175+ billion rows per second in production, peaking at 670+ billion with our zero-matches cache.

Edge deployments solved a different kind of challenge. Some workloads can’t leave a region due to data sovereignty requirements, regulatory constraints, or latency sensitivity. The unified edge architecture we built this year unlocked regional deployments and opens the path for more flexible deployment models ahead.

Behind these features: tens of petabytes ingested monthly, millions of GB-hours of query compute - the equivalent of centuries of cumulative querying across our customers - and revenue that more than doubled as more teams finally found out what complete, affordable visibility feels like. This infrastructure is what lets us promise capabilities other platforms cannot match.

Observability was the first application of Axiom’s infrastructure. We had logs from day one and distributed traces a year later. But customers were desperate for the same remarkable efficiency and performance for their metrics data, too.

MetricsDB fills this gap. A purpose-built metrics datastore that treats hyper-cardinality as a design principle, not a limitation. Teams can ingest OpenTelemetry metrics at petabyte scale, track billions of unique time series without cost penalties, and query metrics alongside logs and traces right inside Axiom’s Console. Now in public preview, MetricsDB completes Axiom’s observability story: all three pillars are unified.

Spotlight changed how investigation works in Console. Traditional workflows are reactive: something looks wrong, you iterate on queries hunting for the pattern that explains it. Spotlight flips this. Select a region on a chart and it automatically compares those events against baseline behavior, analyzes every field, identifies significant differences, and generates an AI-powered summary. From “something’s wrong” to “here’s why” in seconds. Seriously.

These aren’t just features. They’re what observability can look like when it’s built on thoughtful infrastructure designed for this workload from day one.

Software is evolving faster than anyone imagined even two years ago. AI features have moved from novelties to competitive necessities. The brightest engineering teams are building multi-step AI workflows, agentic systems, and products that learn from every interaction.

These teams face a new class of problems. Traditional testing doesn’t work when correctness is probabilistic. You ship a prompt change, watch metrics, and hope nothing breaks. When something does break, most are left guessing which change caused it. That’s shipping on vibes.

Axiom’s AI engineering toolkit is our answer. Not a separate product, but a specialized lens over Axiom’s data platform for teams shipping generative AI to production.

Rich telemetry capture for prompts, completions, tool calls, and costs. Pre-built dashboards that appear automatically with your first AI telemetry. Offline evaluations that let you test changes against curated collections and catch regressions before they reach production. The same infrastructure that powers observability at scale, applied to the unique challenges of probabilistic systems.

This is where software development is heading. Axiom is building the infrastructure and workflows to support it.

We’re not just building tools for other teams adopting AI. We’re evolving Axiom itself.

Axiom’s MCP server launched in September and connects AI agents directly to organizational event data. Agents can query, analyze, and reason through complex debugging scenarios without manual query crafting. It works with any MCP-compatible agent and includes domain-specific prompts for common workflows. This was our first step toward something bigger.

In 2026, customers will start to feel this evolution more directly. Axiom will provide a focused space to move through queries in a structured investigative flow, capture how investigations unfold, and return to work later with full context preserved. Intelligence features will shift Axiom from reactive, human-led search toward proactive, agent-assisted investigation.

The insight driving this work: humans have always been the limiting factor in extracting value from machine data. You collect everything, but insight depends on someone asking the right questions at the right time. That constraint is about to change. Axiom’s intelligence features will help organizations realize more value from every byte they collect.

2026 priorities follow directly from this evolution.

Completing the AI engineering loop. User feedback capture, annotation workflows for domain experts, and online evaluations that create continuous improvement cycles. The toolkit will mature from observability into a complete system for confident AI development.

Console intelligence. The investigative layer that makes observability collaborative and reusable, with AI assistance that helps teams move faster than human-led search alone allows.

MetricsDB general availability. Taking the complete observability platform from preview to production-ready.

To our customers: you trusted us with your data and pushed us to solve harder problems. The feedback shapes everything we build.

2025 was a year of building foundations and making bets on where software development is heading. We’re ready for 2026.

Happy holidays from the Axiom team. We hope you can log off, while we keep logging everything for you.