The continual learning platform for agents — AI observability, LLM evaluation, and agent improvement (Phoenix OSS + Arize AX), being acquired by Dynatrace for $915M
HQ: Berkeley, CA (Dynatrace press release cites San Francisco) | Founded: 2020 | Employees: ~150-220 (Tracxn ~223 / PitchBook ~135 by source; 26 open roles Aug 2026) | Stage: Pending acquisition by Dynatrace — definitive agreement Aug 13, 2026, $915M ($815M cash + replacement equity), expected to close ~Q2/Q3 FY2027 | Website: https://arize.com
Arize AI was founded in 2020 in Berkeley, CA by Jason Lopatecki and Aparna Dhinakaran, two ex-Uber engineers who had built Michelangelo, Uber's machine-learning platform, and experienced firsthand how painful it was to troubleshoot models in production. Their thesis: AI teams need visibility into why models fail — drift, quality decay, hallucinations — before failures reach customers, and this need compounds as LLMs and agents go mainstream. Arize shipped the first AI-observability footprint across both traditional ML and then generative AI, open-sourcing the Arize Phoenix evaluation/tracing library and creating OpenInference, the standard GenAI semantic conventions layered on OpenTelemetry. The company raised institutional rounds (a Series A in 2022 led by Battery Ventures, a Series B in 2023, and a $70M Series C in Feb 2025 — the largest in AI observability — led by Adams Street with M12/Microsoft, OMERS, Datadog, and PagerDuty among participants), acquired evaluator startup Velvet in March 2025, and shipped a first-to-market audio-evaluation capability for voice assistants, agent-native debugging for coding harnesses (Cursor, Claude Code, Codex, GitHub Copilot, OpenCode), and 'Alyx', an AI-engineering agent that runs evals, debugs, and improves agents. On August 13, 2026, Dynatrace announced a definitive agreement to acquire Arize for $915M in a cash-and-equity deal; both founders join Dynatrace, with Lopatecki continuing to lead the Arize team. The platform is certified SOC 2 Type II, ISO 27001, PCI DSS, HIPAA, and GDPR-compliant, and now processes over 1 trillion traces annually while the Phoenix OSS project clocks ~5M downloads a month.
Arize built the de facto open standard for AI observability and got the distribution flywheel right: Phoenix is the most-adopted OSS observability/eval library for AI dev (5M monthly downloads), OpenInference is the OpenTelemetry semantic convention for GenAI, and the managed Arize AX converts that developer gravity into enterprise ARR trusted by Fortune 500s and AI-native builders. Its differentiated wedge is the closed loop — observe traces, run evals at scale, then use the signal (and the 'Alyx' AI-engineering agent) to push fixes, which is why the tagline became 'continual learning for agents'. The $915M Dynatrace deal (~$815M cash plus equity for employees, roughly a 4-5x on ~$150-230M raised) is a classic 'pick-and-shovel wins in an expanding category' exit: Dynatrace needs the eval/observability layer and developer trust that Arize owns to win the >$10B-by-2030 AI observability market, and it's paying ~200bps ARR-growth accretive. Risks: total-funding estimates vary by source, the OSS-led wedge has thin direct monetization, and the competitive field (Datadog LLM Observability, LangSmith, Langfuse, Braintrust, Helicone) plus model-provider-native tracing could compress the standalone case — which is exactly why being absorbed into an observability giant is the strategically sound endgame.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki