Nebius (NASDAQ: NBIS), a European infrastructure and cloud company focused on artificial intelligence, has announced an agreement to acquire Tavily, a provider of “agentic search” (search oriented to agents) with the goal of integrating real-time web access capabilities into its platform. The deal, announced on February 10, 2026, is expected to close “within the next few weeks” if customary closing conditions are met, though the company has not disclosed the transaction amount.
Nebius’s thesis is clear: as companies move from pilots to actual deployments of autonomous agents, fast and low-cost inference is no longer enough. A robust search/grounding layer is also needed to allow agents to query the web, verify information, and reduce errors—without forcing developers to “patch together” multiple providers and APIs within the same stack.
What Nebius is buying (and why now)
Tavily positions itself as a search infrastructure provider for agents—a component that enables autonomous systems to make internet queries, retrieve relevant results, and feed them into the agent’s reasoning process. In the announcement, Nebius frames Tavily as a complementary piece to its “Token Factory”: if that layer delivers high-performance inference, search provides access to up-to-date information many agents need to be effective in real-world tasks.
This move aligns with a trend seen in enterprise architectures: agents performing workflows (research, support, operations, compliance, logistics) tend to fail less when combining reasoning capabilities with information retrieval and verification mechanisms. In other words: it’s not just “more models,” it’s “better data access and external connections.”
An acquisition aimed at building a “platform,” not just IaaS
Nebius emphasizes that it does not want to be just an IaaS provider. The narrative focuses on building an integrated platform so that vertical startups and large enterprises can train, fine-tune, and run agentic systems more smoothly. Real-time search is a challenging component to implement well: it impacts latency, cost control, security, compliance (what is queried, how it’s filtered, how it’s audited), and output quality.
The announcement also confirms that the Tavily team, including founder and CEO Rotem Weiss, will join Nebius, and that the Tavily brand will continue to operate as a product.
How much is the deal worth?
Nebius has not disclosed the price.
However, several media outlets have provided estimates: Bloomberg suggests around $275 million, citing sources close to the transaction. Other reports from Israeli press and agencies have mentioned figures in that range, with potential variables tied to performance targets.
Metrics Nebius considers as “adoption proof”
In the announcement, Nebius highlights that Tavily has achieved more than 3 million downloads of its SDK per month and has a community of over one million developers, along with mentions from clients such as IBM and AI ecosystem companies like Cohere and Groq. These figures are significant indicators of traction, especially in a market where many “stack pieces” emerge as libraries or APIs that gain adoption before evolving into enterprise products.
Why “agentic search” is becoming strategic
The promise of agents is to perform complex tasks with some degree of autonomy: browsing information, verifying sources, making conditional decisions, and executing actions (creating tickets, placing orders, making changes, generating reports). The bottleneck occurs when the agent needs current or verifiable data and can no longer rely solely on the static knowledge embedded in the model. That’s where layers of search, retrieval, and grounding come in: they are not luxuries but essential for system reliability.
Nebius also supports this perspective with market projections (growth from approximately $7 billion in 2025 to $140–$200 billion at the start of the 2030s, according to cited studies). While these forecasts for emerging markets should be taken with caution—market predictions are often optimistic—they illustrate why there’s a race to “own” the full agent stack.
What it means for clients: less “glue” and more control
If Nebius manages the integration well, the enterprise client’s proposition is straightforward:
- Fewer vendors needed to build agents (inference + search + operations under one umbrella).
- Better control over latency and costs (avoiding chaining external services with dispersed billing).
- Potential improvements in auditing and governance: knowing what is queried, when, under what rules, and how evidence is stored.
Still, the real challenge isn’t the announcement itself but execution: technical integration, governance, security, observability, and creating a development experience that simplifies rather than complicates what it aims to streamline.