26 Jul 2026 3 min read Western Hemisphere

Notion Buys AI Reranking Startup ZeroEntropy, Open-Sources Its Models

Ghita Houir Alami, CEO of the Company.

Notion has acquired ZeroEntropy, a San Francisco-based AI search infrastructure startup co-founded by Moroccan engineer Ghita Houir Alami, who serves as CEO, and CTO Nicholas Pipitone. The entire ZeroEntropy team is joining Notion, with Houir Alami taking charge of a newly formed Model Research group at the productivity software company. Financial terms of the deal, announced this week, were not disclosed.

The acquisition formalizes a relationship that began commercially: Notion had been running its unified AI search on ZeroEntropy's reranking model, zerank-2, for roughly two months before the deal, a swap that replaced systems from larger, better-funded competitors. As part of the transaction, ZeroEntropy is open-sourcing its full model lineup under an Apache 2.0 license on Hugging Face, allowing existing customers to self-host without a commercial license. The startup's own products, including its zsearch search engine, will stay operational until September 4, 2026, before being shut down; new customer signups have already been disabled.

Market Context

ZeroEntropy, founded in 2024 and a graduate of Y Combinator's Winter 2025 batch, built its business on a specific bet: that narrow, repeatable steps inside AI pipelines — reranking, embedding, classification, query rewriting — don't require frontier-scale models, and that small, task-specific models can run them faster and more cheaply. That thesis is gaining traction industry-wide as AI companies look to cut inference costs without sacrificing output quality, and it's central to why Notion, rather than licensing the technology, chose to bring the twelve-person team in-house.

For Notion, the deal continues a run of AI-driven growth. The company reported roughly $600 million in annual recurring revenue in 2025, with about half attributed to its AI products, and completed a $270 million secondary share sale in January 2026 that valued the company at $11 billion, with backers including Sequoia Capital, Index Ventures, and Singapore's GIC. Multiple reports suggest Notion, co-founded in 2013 by CEO Ivan Zhao alongside Simon Last and Akshay Kothari, is positioning for a potential IPO by late 2026.

What Stands Out

"The future of AI products will not be powered by a single model" — Akshay Kothari, Notion co-founder and president.

Regional Relevance

For the United States: The deal underscores how AI-native software companies are increasingly opting to acquire specialized model teams outright rather than license third-party infrastructure — a build-via-buy strategy that lets platforms like Notion control costs and roadmap as they scale toward a potential public listing. San Francisco remains the epicenter of this activity; both acquirer and target are based there, reinforcing the city's grip on AI infrastructure talent even as that talent increasingly originates elsewhere in the world.

For Morocco and the broader MENA region: The acquisition is a visible win for Moroccan tech talent on the global stage, with Houir Alami — raised in Rabat before leaving at 17 for France's École Polytechnique and later UC Berkeley — now leading model research at an $11 billion U.S. company. But it also reflects a recurring pattern: founders from the region build and raise capital in the U.S. rather than at home, and exits accrue to Silicon Valley rather than to domestic ecosystems. ZeroEntropy's deal is the second such case in two weeks, following Egyptian-founded Stakpak's acquisition by Vercel, raising questions about whether this pipeline strengthens the region's tech reputation or simply exports its most promising builders.

The Other Side

Is an undisclosed, all-team acquisition really a strong outcome, or a soft landing? ZeroEntropy raised only $4.2 million in seed funding before being absorbed roughly a year later. Without disclosed terms, it's unclear whether investors and the twelve-person team saw a meaningful return or whether this functions closer to an acqui-hire that lets a promising but commercially unproven startup avoid winding down.

Why open-source everything right before shutting the product down? Releasing all models under Apache 2.0 is generous to developers, but it also removes any future licensing revenue and signals that ZeroEntropy's standalone business — rather than its team and technology — was not what Notion wanted to keep running.

Does small, task-specific model infrastructure actually hold up as AI labs keep shipping cheaper frontier models? ZeroEntropy's pitch rested on narrow models beating general-purpose ones on cost and latency for specific tasks. If frontier labs continue driving down the price of large models, the cost argument for specialized infrastructure could narrow, testing whether Notion's bet pays off long-term.