The pure-play AI observability platform built from the ground up for LLMs and agentic systems, now the largest-ever funded company in its category.
arize.com ↗Arize AI was founded in 2020 in Berkeley, CA by Jason Lopatecki (CEO, former AI strategy lead at TubeMogul/Adobe) and Aparna Dhinakaran (CPO, former lead of Uber's model lifecycle management system). The two founders built Arize from firsthand frustration with deploying ML in production without adequate tooling for debugging, evaluation, and monitoring. The company initially found product-market fit in traditional ML observability — model drift, performance degradation, labeling errors — and rapidly pivoted its platform to address LLMs and agentic AI following the ChatGPT inflection. It is the only pure-play AI observability company that spans the full lifecycle from development experimentation through production monitoring.
In February 2025, Arize closed a $70 million Series C — the largest-ever investment in the AI observability category — led by Adams Street Partners with participation from M12 (Microsoft's venture fund), Datadog, PagerDuty, OMERS Ventures, and Sinewave Ventures. Existing investors Battery Ventures, TCV, Foundation Capital, and Swift Ventures also participated. Total raised is $131 million across four rounds (Seed $4M 2020, Series A $19M 2021, Series B $38M 2022, Series C $70M 2025). The company plans to use capital to scale evaluation and monitoring for multi-agent systems, expand its open-source Phoenix library, and hire engineering talent.
Arize operates a dual-product strategy: Arize AX (enterprise-grade platform for AI engineers) and Phoenix (open-source LLM observability library built on OpenTelemetry). Phoenix has become the de facto standard for LLM tracing and evaluation in the developer community, with broad adoption enabling Arize to build a commercial pipeline from open-source users. The June 2025 Observe conference previewed Agent Visibility (multi-agent system visualization), Agent Trajectory Evaluation (assessing the quality of agent decision paths, not just outcomes), Prompt Learning (automated prompt optimization), and Alyx (an in-product AI copilot with trace troubleshooting and MCP integration). ADB, a purpose-built database for processing billions of traces, underpins the enterprise platform.
Arize AI matters because it identified the hardest unsolved problem in enterprise AI deployment — the black box between what an LLM agent is supposed to do and what it actually does in production — and built dedicated infrastructure to address it before the category was understood. Traditional APM and observability tools were designed for deterministic software; LLMs are non-deterministic, context-dependent, and fail in ways that can't be caught by latency monitors or error rates alone. Arize's Phoenix open-source library creates a developer distribution moat that is difficult for incumbents to replicate, while its enterprise platform adds evaluation frameworks, compliance logging, and multi-agent visualization that enterprises need for AI governance. The strategic participation of Datadog and Microsoft (M12) in the Series C signals that even the largest infrastructure players see Arize as either a partner or acquisition target.
Arize AI is the category leader in dedicated LLM and agentic AI observability, holding a first-mover advantage over specialized competitors including Galileo ($68M raised), Patronus AI ($20M raised), and LangSmith (Langchain's commercial offering). Its open-source Phoenix library is the most widely adopted LLM observability toolkit by developer count, creating distribution depth that pure commercial vendors cannot match. The competitive risk is dual: infrastructure incumbents like Datadog (which participated in Arize's Series C) could embed sufficient LLM observability into their platforms to satisfy mainstream enterprise buyers, while Arize must prove that its evaluation and multi-agent capabilities are differentiating enough to justify standalone adoption at scale. Arize's federal contract win (AFWERX) and Microsoft M12 participation suggest a path toward government and Microsoft-ecosystem expansion that could extend its moat beyond the developer community.
206 companies across 10 categories — the most comprehensive AI security company tracker.
Browse All Companies →