Legal AI startup Legora raises $550M at an $8.8B valuation
AspireAI · Legora
Legora, the agentic operating system for legal professionals, announced a $550 million Series E at an $8.8 billion post-money valuation, up from $5.6 billion at its previous round. The round was led by Atreides Management, with new investors including Baron Capital, Capital Group, Fidelity Management & Research Company, Glade Brook Capital Partners and J.P Morgan, and every major existing investor participating.
Growth, per the company: ARR went from $1 million to $100 million in less than 18 months, the next $100 million took less than six months, and ARR now stands above $200 million. Legora states this is faster than any other enterprise software company selling through a direct sales team.
The most telling detail in the release is the customer mix. In-house legal teams were only a small fraction of the customer base a year ago and made up more than 40% of new customers in the third quarter. In the United States, now Legora's largest market by revenue, more than half of the AmLaw 50 are customers, and Legora opened its first US office only in spring 2025.
The capital funds expansion across North America, Europe and Asia-Pacific, plus deeper investment in the Legora aOS, the agentic operating system where lawyers and their agents work, including new products for agentic in-take, horizon scanning and legal research.
Why it matters: legal has become the clearest vertical where AI has moved past drafting assistance into multi-step agents that carry out work. That in-house teams are now the fastest-growing slice suggests adoption is spreading inside companies, not only at law firms.
Honest limitation: every growth figure above is self-reported by Legora and unaudited, and ARR in legal AI says nothing about whether those agent hours convert into billable work for customers.
Source: Legora newsroom press release, October 21, 2026.
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.