After Harvey, vertical AI’s next $10B winner might be in agriculture

Vertical AI startups like Harvey ($11B) and Abridge have proven that wrapping a foundation model with proprietary data and domain workflow logic creates billion-dollar companies. Agriculture is the next frontier, with a fragmented $500B data layer and USDA's $300M Palantir deal validating the thesis. GrowersTech is building the domain-specific intelligence stack from inside the industry to capture this opportunity.

TL;DR Harvey crossed $11B by wrapping legal data around a foundation model. Agriculture’s fragmented $500B data layer is now the next vertical AI frontier, with USDA’s $300M Palantir deal validating the thesis and GrowersTech building the domain-specific intelligence stack from inside the industry. In legal, Harvey crossed an $11 billion valuation and Legora is racing to claim the European market behind it. In healthcare, Abridge has built a multi-billion-dollar business turning clinical conversations into structured medical records. In customer service, Sierra is now valued at over $15 billion, making it one of the fastest companies in AI history to reach that mark. The architecture in every case is identical: a foundation model https://thenextweb.com/news/5-things-about-hottest-new-trend-ai-foundation-models wrapped in proprietary data, a domain ontology, and the workflow logic of a single complex industry. Vertical AI https://thenextweb.com/news/business-ai-solutions-beginners-what-is-vertical-intelligence has become the most reliably category-defining bet in the current AI cycle. The question every investor is now asking is which sector goes next, and how big the prize is when it does. One answer keeps coming back: agriculture. The numbers underneath it are easy to miss. McKinsey has estimated that connecting agriculture’s fragmented data could add $500 billion to global GDP. U.S. crop farmers alone spend roughly $72 billion a year on seed, fertilizer, and crop protection. The AI-in-agriculture market is forecast to more than triple this decade, growing from $2.43 billion in 2025 to over $8 billion by 2031, according to Mordor Intelligence. And every dollar of value in that expansion is gated by a single bottleneck: the data layer running underneath every decision in the global food system is broken. What Harvey actually proved Harvey’s $200 million round at $11 billion was not won by having a stronger foundation model. It was won by the layer around it, trusted legal data, an ontology of how legal work actually flows, evaluation frameworks tuned to legal accuracy standards, and matter-specific context that ChatGPT cannot synthesize on demand. Its A&O Shearman partnership produced agentic, multi-step AI for antitrust filings, fund formation, and loan review. Its LexisNexis alliance bolted in trusted content and workflow tooling. None of that was a model story. All of it was a context story. MIT Project NANDA’s 2025 GenAI Divide report found that 95% of organizations are getting zero return from their GenAI initiatives https://thenextweb.com/news/mckinsey-ai-productivity-paradox-enterprise-roi-capex . The diagnosis was not that the models are weak. It was that horizontal AI rarely transforms complex industries on its own. Harvey solved that for law. Abridge solved it for clinical documentation. Sierra solved it for customer service. The companies positioned to solve it for agriculture are now the ones investors are watching most closely. The U.S. government just put $300 million behind the same thesis. A $300M signal from USDA In 2026, the U.S. Department of Agriculture launched its “ One Farmer, One File ” initiative, an effort to unify systems across the Farm Service Agency, Natural Resources Conservation Service, and Risk Management Agency into a single farmer record. Shortly after, USDA and Palantir https://thenextweb.com/news/palantir-thales-air-space-intelligence-faa-smart-ai-air-traffic announced a $300 million Blanket Purchase Agreement supporting the National Farm Security Action Plan. The deal validates what agri-tech operators have argued for years: agriculture’s data problem is now infrastructure-level https://thenextweb.com/news/2-trillion-ai-infrastructure-problem-shashidhar-bhat . Palantir is the horizontal solution. The vertical version, built from inside agriculture, with agronomic logic encoded as primary architecture, is what the next category-defining company in this space will look like. That is the bet behind GrowersTech, the Israeli–American group that combined the Agmatix data platform with GROWERS, the U.S. retailer-loyalty company that captures the transactional layer of the American agriculture supply chain. Its core engine, Axiom, is built around a neuro-symbolic AI architecture: a knowledge graph layered with pre-trained agronomic ontologies, fused with field-level data and the transactional signals flowing between input manufacturers, retailers, and farmers. “ Agriculture doesn’t have a data problem. It has an intelligence problem, ” said Ron Baruchi, the company’s CEO. “ The data exists. What’s missing is infrastructure that understands what it means. ” Why generic AI breaks in the field A generic model knows what nitrogen is. It cannot tell you that the right amount changes depending on the growth stage, the soil type, what was planted in that field the previous year, and what the weather will look like for the next 90 days. Agronomic decisions are made in context. That context is what every horizontal AI deployment in agriculture has failed to capture, and it is what makes the sector so unforgiving for generic tools. GrowersTech starts from the opposite direction. Its ontology, the structured representation of how agriculture actually works, is pre-trained by agronomists before any customer data enters the system. Relationships between crops, soils, products, and outcomes are encoded as primary architecture. New deployments configure that ontology rather than rebuilding it. The system now structures more than 1.5 billion field-trial data points, drawn through research partnerships with leading agricultural universities. The company’s modeling work has been published in Nature, a credibility marker rarely earned by private agri-tech vendors. The platform is already running inside global crop input manufacturers, large food and beverage supply chains, U.S. ag retailer cooperatives, and government agriculture ministries, covering product performance prediction, sustainability modeling, retailer loyalty intelligence, and accelerated R&D trials. Each deployment adds to the underlying data graph, which compounds the model’s accuracy across every other use case. That is the same flywheel that turned Harvey into legal-industry infrastructure. How big the prize gets The case for vertical AI in agriculture is structurally larger than the legal case. Agriculture’s economic footprint is global. Its data is more heterogeneous. Its decision points are more numerous. Legal AI compresses the work of expensive professionals. Agricultural AI compresses the loss between decisions and outcomes across a $500 billion value chain. Harvey hit $11 billion because legal was the first vertical AI category to prove out. The companies that win the next round will be valued against a far larger denominator. The open question is no longer whether vertical AI in agriculture produces a category winner. It is which company scales first, and how much bigger the prize gets when it does.