Notion has acquired ZeroEntropy, an AI infrastructure company specialising in efficient, task-specific models. ZeroEntropy’s full team will join Notion, and chief executive Ghita Houir Alami will lead a new Model Research group inside the company.
The acquisition is intended to give Notion more control over the models and inference systems behind its AI features. Rather than relying only on large general-purpose models, Notion wants to build smaller components optimised for specific jobs such as search, ranking and information retrieval.
ZeroEntropy’s role in Notion search
The two companies have already worked together on Notion’s unified search. ZeroEntropy developed a reranker that helps decide which retrieved results are most relevant before an answer is generated. Notion says this work made unified search up to 30 per cent faster and reduced reranking latency by 85 per cent while maintaining answer quality and lowering inference cost.
Those figures are company-reported and may vary with the query, workspace and workload. They nevertheless illustrate why specialised models can matter. In a search pipeline, a large model does not need to perform every step. A smaller reranker can handle one narrow decision quickly, leaving more expensive models for tasks where their broader reasoning is useful.
For users, the infrastructure is mostly invisible. The practical effects should appear as faster answers, more relevant results and AI features that can be offered at a sustainable cost. Search quality is particularly important in Notion because answers may draw from pages, connected apps and large stores of workplace knowledge.
A new model research group
ZeroEntropy brings experience in custom computing kernels, inference systems and techniques for specialising models. Notion says the new group will focus on models that are fast, affordable and designed around the work people perform in its product.
This is a different emphasis from training a single frontier model to do everything. Task-specific systems can be smaller and cheaper to run, and they may be easier to evaluate because the expected behaviour is narrower. The trade-off is that a product needs a reliable way to route each task to the right component and monitor the combined system.
Notion has not disclosed the financial terms of the acquisition. It has also not announced a new customer-facing product or a timetable for features produced by the Model Research group. The immediate change is organisational and technical: the ZeroEntropy team and its capabilities are becoming part of Notion.
What customers should watch
The acquisition could improve Notion AI’s speed and economics, but enterprise customers will also want clarity about evaluation, data handling and model lifecycle management. A faster ranking system is valuable only if it continues to surface the right material and respects existing workspace permissions.
Teams should watch for measurable changes to search latency and relevance, especially in workspaces with many connected sources. Administrators may also want information about regional processing, retention policies and whether new specialised models change existing security or compliance arrangements.
Notion’s published numbers focus on reranking latency and overall search speed. They do not show how the system performs across every language, industry or workspace size. Independent testing with representative queries remains the best way for customers to judge whether search has improved for their own content.
Why the acquisition matters
AI product competition increasingly depends on infrastructure as well as headline model capability. Retrieval, ranking, caching and efficient inference determine how quickly a system responds and how much it costs to serve. Bringing those skills in-house may let Notion tune the complete experience more closely.
The deal also signals that Notion sees bespoke models as a core part of its long-term product, not merely a supporting integration. If the new research group succeeds, users may see quicker workplace search and more specialised assistance without every task being sent to the largest available model.
For now, the acquisition should be treated as an investment in the foundations of Notion AI. Its customer impact will become clearer when Notion ships new features or publishes further performance evidence from the combined team.