Harvey
Seat-based enterprise SaaS subscription; list ~$1,000-2,000/lawyer/month for mid-market firms, volume-discounted well below list for large Am Law 100 deployments (200+ seats); ~$288K annual minimum spend, 12-month commitments, ~20-25-seat minimums, land-and-expand within firms
The five valuation marks from Jul-2024 ($1.5B) through Mar-2026 ($11B) are company-confirmed; the Dec-2023 Series B (~$715M) valuation is from press reporting. Roughly 7.3x re-rating in ~20 months. Cumulative ~$1.22B raised across ~10 rounds since the Nov-2022 OpenAI Startup Fund seed.
Earnings, margins, COGS & capex
Hyper-growth vertical-AI SaaS: ARR roughly tripled inside ~9 months to a Sacra-estimated ~$300M by May 2026, on 1,500+ customers and 142,000+ lawyers including ~50% of the Am Law 100. Growth is seat-expansion-led (median seat count reportedly doubles within 12 months) and increasingly agentic (25,000+ custom AI agents run for clients). Profitability and margins are undisclosed; the economics hinge on the spread between seat pricing and foundation-model inference cost, which the multi-model routing strategy is partly designed to manage.
Revenue trend
Margins
structurally below classic SaaS due to per-query LLM inference COGS
assumed negative - growth-stage burn
assumed negative
COGS structure
Dominated by foundation-model inference (OpenAI, Anthropic, Google) plus cloud hosting and human 'embedded legal engineering' delivery teams - a heavier services/COGS mix than pure SaaS. Multi-model strategy (adopted May 2025, adding Anthropic Claude and Google Gemini alongside OpenAI) partly a COGS + leverage hedge against single-vendor pricing; models served via provider clouds (Claude on AWS Bedrock, Gemini on Vertex AI).
Capex
Minimal owned capex; asset-light. Spend concentrated in R&D, model fine-tuning/eval (proprietary 'BigLaw Bench' benchmark), and global go-to-market / legal-engineering headcount rather than physical or data-center assets.
Latest earnings
n/a
No public guidance. Capital raised (Mar 2026) earmarked for scaling AI agents and expanding embedded legal-engineering teams globally.
- ARR (May 2026, Sacra est.)
- ~$300M
- Customers
- 1,500+ across 60+ countries
- Lawyers on platform
- 142,000+
- Am Law 100 penetration
- ~50%
- Custom AI agents run for clients
- 25,000+
- Weekly active users
- ~4x YoY
- Monthly queries
- ~5.5x YoY
Growth drivers
- Seat expansion inside existing firms (median seat count reportedly doubles within 12 months)
- Agentic workflows — autonomous multi-step legal tasks (25,000+ custom agents) - as the upsell beyond single-shot drafting/research
- International expansion (60+ countries)
- Adjacency into enterprise/professional services (50 asset managers; PwC deployment across 60+ countries) beyond pure law firms
- Frontier-model capability gains lifting the ceiling on legal-task reliability
Bull & bear
Harvey is the emerging system-of-record for AI-native legal work, winning the highest-value, stickiest customers first and expanding seats and use cases faster than incumbents can retrofit AI - a durable vertical-AI compounder if it converts brand and workflow depth into pricing power as models commoditize.
- Owns the hardest segment: penetrating ~50% of the Am Law 100 is a reference moat competitors can't cheaply replicate
- ARR tripled in ~9 months to a Sacra-estimated ~$300M with seat counts doubling within firms - land-and-expand is working
- Model-agnostic auto-routing turns foundation-model progress into a tailwind, not a dependency, and improves COGS leverage over time
- Massive, underpenetrated TAM: global legal + professional services is a multi-hundred-billion labor pool ripe for AI substitution
- Best-capitalized independent in the category, able to outspend on R&D, GTM and legal-engineering delivery
- Expansion beyond outside counsel (PwC across 60+ countries, ~50 asset managers, in-house teams) proves the platform travels past pure Big Law
Harvey is a fast-growing application layer renting its core intelligence from three model vendors, sandwiched between content-moat incumbents (Thomson Reuters, LexisNexis) below on price and the model labs above on capability, at a ~37x-trailing-ARR valuation that prices in near-flawless execution.
- No model moat: the intelligence is OpenAI/Anthropic/Google IP; defensibility rests on workflow + data, which incumbents also have
- Incumbents own the proprietary legal corpus (Westlaw, Lexis) Harvey must license or approximate - and they're shipping agentic AI (CoCounsel Legal built on Anthropic's Claude Agent SDK; Lexis+ Protege) at a fraction of Harvey's price
- Undisclosed gross margins likely well below SaaS norms; heavy legal-engineering services mix caps scalability
- Valuation re-rated $1.5B (Jul-2024) -> $3B (Feb-2025) -> $5B (Jun-2025) -> $8B (Dec-2025) -> $11B (Mar-2026) - ~37x trailing ARR - discounting hyper-growth persisting with zero accuracy/liability stumbles
- Big-Law buyers can insource on the same public models or consolidate onto their existing Thomson Reuters/LexisNexis vendor, compressing Harvey's wedge
- Two of the three key growth marks ($195M end-2025, $300M May-2026) are Sacra estimates, not company-disclosed - the momentum narrative rests partly on third-party modeling
What it is worth
Private last-round mark + ARR-multiple cross-check (no public price / no DCF-grade disclosure). ~$11B on a Sacra-estimated ~$300M ARR = ~37x trailing ARR run-rate.
~$4-6B
if growth decelerates, incumbents (TRI/RELX) win on price+content, and gross margins disappoint, the wrapper-risk discount re-rates it toward Legora-adjacent levels or below.
~$11B
roughly the current mark holds; growth stays high but the multiple compresses toward the low-30s as ARR scales into it.
~$15-20B+
if ARR compounds past $600M-$1B in 12-24 months with improving margins and agentic upsell, sustaining a 25-30x forward multiple; sets up a premium IPO.
The Mar-2026 $11B mark (~$200M headline round, ~$288M reported) implies a ~37x trailing-ARR multiple, rich even for hyper-growth vertical AI and the top of a fast ladder ($1.5B Jul-2024 -> $3B Feb-2025 -> $5B Jun-2025 -> $8B Dec-2025 -> $11B Mar-2026). The number is only supportable if ~150-200%+ growth persists and Harvey converts brand/workflow lock-in into durable, SaaS-grade gross margins as models commoditize - neither yet proven publicly.
SWOT
Strengths
- Category-leading brand and reference logos in elite law (~50% of Am Law 100), the hardest, stickiest segment to win
- Sacra-estimated ~$300M ARR at ~200%+ growth - among the fastest-scaling vertical-AI apps
- Deep, blue-chip capital base (~$1.22B raised — Sequoia, GIC, a16z, Kleiner Perkins, Coatue) funding multi-year burn
- Model-agnostic architecture (OpenAI + Anthropic + Google) with auto-routing reduces single-vendor dependence and lets it route to best/cheapest model per task
- Embedded legal-engineering delivery model deepens switching costs and workflow lock-in
Weaknesses
- Undisclosed and likely pressured gross margins - LLM inference COGS sits between it and SaaS-grade economics
- Thin proprietary moat at the model layer; core intelligence is rented from third parties it does not control
- Heavy services/legal-engineering mix dilutes pure-software scalability
- Valuation (~37x trailing ARR) leaves no room for a growth stumble
- Concentrated in a slow-to-adopt, risk-averse buyer (Big Law) sensitive to accuracy/hallucination and malpractice risk
Opportunities
- Expand from law firms into the far larger enterprise/in-house + professional-services (audit, tax, consulting) TAM
- Agentic workflows moving from assistant to autonomous task completion - higher value capture per seat
- International white space (EMEA/APAC) where incumbents are weaker
- Owning proprietary legal data/eval and workflow layer as models commoditize
Threats
- Incumbents with proprietary content moats — Thomson Reuters (Westlaw) CoCounsel and LexisNexis (RELX) Lexis+ with Protege - bundling agentic AI into existing firm relationships at lower price points
- Well-funded pure-play rivals (Legora at $5.55B) and Word-native tools (Spellbook) attacking mid-market and EMEA
- Foundation-model providers (OpenAI/Anthropic/Google) moving up the stack into vertical legal apps - the platform risk of being a 'wrapper' (Anthropic has begun shipping legal-focused offerings)
- Pricing pressure — Harvey's list (~$1,000-2,000/mid-market seat) is a multiple of CoCounsel Core (~$225/user/mo) and Legora (~$300-800/user/mo)
- AI liability/regulatory scrutiny and a single high-profile hallucination incident in a live matter
Moats, dependencies & bottlenecks
Moats
~50% of Am Law 100; hardest segment to win, high trust barrier for new entrants
Embedded legal-engineering teams and firm-specific customization raise cost to rip out
Data/eval flywheel, but not a content moat like Westlaw/Lexis
~$1.22B raised outspends most rivals, but capital is not a durable moat
Core intelligence is rented from OpenAI/Anthropic/Google - explicitly NOT owned
Dependencies
Core technology supplier Product intelligence and COGS both governed by model vendors; multi-model strategy (May 2025) diversifies but does not remove the dependency
Inference and hosting; Claude served on AWS Bedrock, Gemini on Vertex AI - pricing and capacity exposure
Slow, risk-averse buyers; adoption gated by accuracy/liability comfort
Unprofitable growth-stage; reliant on continued private-market access at rich multiples
Lacks a captive proprietary corpus at Westlaw/Lexis scale; must license or partner for authoritative citations
Advantages
- First-mover brand and the deepest elite-law customer base in the category
- Fastest ARR ramp among independent legal-AI players (Sacra-estimated ~$300M)
- Model-agnostic auto-routing across the three frontier labs
- Best-funded pure-play, enabling sustained R&D + GTM outspend
- Land-and-expand seat economics with proven in-firm expansion (median seats double in 12 months)
Weaknesses
- Rented, non-proprietary core intelligence (platform/wrapper risk)
- Undisclosed and likely sub-SaaS gross margins
- No proprietary legal-content moat vs Thomson Reuters/LexisNexis
- Premium pricing exposed to cheaper, content-backed incumbents
- Valuation prices in flawless multi-year execution
Bottlenecks
- Accuracy/hallucination tolerance in live legal matters - the ceiling on autonomous agent adoption
- Gross-margin drag from per-query inference COGS as usage scales
- Long enterprise sales + security-review cycles in conservative firms
- Talent: scarce, expensive legal-engineering + ML staff for delivery
- Dependence on foundation-model capability roadmaps it does not control
Top signals & trends
Top signals
bullish if sustained · Sacra-estimated ~$300M May-2026 print is the key momentum proof point
bearish risk · First real read on wrapper-economics durability
Content-moat incumbents shipping agentic AI at lower price into existing relationships
Legora at $5.55B (Mar-2026 Series D led by Accel) is scaling fast in EMEA/mid-market and now the US
Direct platform-risk to the application layer; Anthropic has begun entering legal
Rapid up-rounds ($3B->$5B->$8B->$11B in ~13 months) raise eventual public-market scrutiny of the multiple
Trends
Expands value capture per seat; Harvey's stated use-of-funds (25,000+ agents live)
Cheaper/better models lift capability but erode the app-layer's scarcity and pricing power
Validates TAM but intensifies competition and pricing pressure
Westlaw/Lexis pairing proprietary corpus with agentic AI is the central competitive threat
Raises accuracy bar and adoption friction; a single incident carries outsized risk
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Foundation-model supplier + first institutional backer (OpenAI Startup Fund, Nov 2022 seed/Series A); was exclusive model provider pre-May 2025
Claude models added May 2025 as part of multi-model strategy; served via AWS Bedrock
Gemini models via Vertex AI + cloud; Google Ventures led the $100M Series C (Jul 2024, $1.5B valuation)
Cloud/hosting option + Microsoft 365 integration surface for legal workflows
Cloud infrastructure; AWS Bedrock serves Harvey's Claude inference
Indirect - GPU compute underlying all model inference Harvey consumes
Core buyer; ~50% of Am Law 100 are customers (partnerships, not public)
Professional-services deployment across 60+ countries - beachhead beyond pure law firms
~50 asset managers use Harvey for diligence/compliance work
Corporate legal teams - expansion TAM beyond outside counsel
Casetext-built CoCounsel integrated with Westlaw/Practical Law; proprietary content moat + agentic 'CoCounsel Legal' (beta Apr 2026, built on Anthropic's Claude Agent SDK) from ~$225/user/mo (Core) - the primary incumbent threat
Lexis+ AI rebranded to Lexis+ with Protege (Feb 24, 2026); citation-grounded research on a captive corpus, bundled into existing firm relationships
Strongest pure-play rival; $550M Series D at $5.55B (Mar 2026, led by Accel), strong in EMEA/multi-jurisdiction, now expanding in the US, priced below Harvey (~$300-800/user/mo)
Contract-focused legal AI; UK-origin, enterprise + in-house counsel
Microsoft Word-native contract drafting/review; popular with solo/small firms and in-house teams
Vertical specialists (personal-injury, litigation, in-house) attacking specific workflows below Harvey's enterprise tier
Coopetition - suppliers today, potential vertical-app entrants; Anthropic has begun shipping legal offerings - platform risk to the application layer