OpenAI-backed Thrive Holdings raises $2B to bring AI to the enterprise
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Thrive Holdings raised $2 billion at a $12 billion valuation from SoftBank, D1 Capital Partners, and Altimeter Capital, with the New York Times first reporting the news
- OpenAI took an ownership stake in the Thrive Capital spinout in December 2025 and embedded employees at portfolio companies to accelerate AI adoption, a hands-on model that has itself become a sellable strategy
- Current, Thrive's accounting arm covering 50+ firms and 2,000+ professionals, processed more than 7,000 tax returns at 98% accuracy via its self-improving TaxAI agents and cut tax prep times by over 30%
- Shield, Thrive's IT arm with roughly 20 companies, reports 36x faster help-desk resolution times and doubled the number of custom AI agents deployed over the past month
- Thrive Holdings is using part of the raise to launch a third platform targeting regulatory services for physical infrastructure — covering data centers, manufacturing, healthcare, power, water, and transportation
- Founding member Anuj Mehndiratta framed the bet around U.S. infrastructure bottlenecks, saying complexity, not field work, is where AI can compress timelines without lowering safety standards
- The raise reflects a wider pattern: OpenAI and Anthropic have each partnered with large PE firms — launching The Deployment Company and Ode respectively — to embed elite engineers into enterprises
- Thrive Holdings now counts more than 70 businesses across its two operating platforms, all built on the thesis that fragmented, regulation-heavy industries are AI's best near-term foothold
Why it matters: Thrive's proof points — 7,000 tax returns at 98% accuracy, 36x faster help-desk resolution, and 70+ portfolio companies — are rare, audit-ready evidence that the embedded-engineer deployment model works. The new infrastructure-permitting vertical gives OpenAI a channel into regulated industries that standalone AI products typically fail to crack.
Ask SkimNews
