Meta acquires AI analytics startup Sundial

The Sundial deal is less a typical tech acquisition than a homecoming. Julie Zhuo, a former Meta VP of Product Design, and Chandra Narayanan, who previously worked in analytics roles at Facebook, Instagram, and Sequoia Capital, left to found Sundial around 2021 - and six years later Meta is the company buying them back ^[1]. Zhuo wrote in her announcement that Sundial was built on the belief that better decisions come from using data well, framing the acquisition as a continuation of that mission rather than an ending. Narayanan will now lead Meta's data analytics tooling and agents work inside the new DATA unit, reporting to chief data officer Alex Schultz ^[4]. For Meta, re-absorbing a team it originally trained sidesteps the usual risk of building equivalent in-house tooling from scratch, pairing fresh outside experience with institutional fluency in Meta's own data systems.

The acquisition is framed internally as a response to a specific problem. Schultz said AI has created effectively unlimited demand for insights, and that Meta wants to define the future of AI analytics ^[2]. Coverage ties this directly to Meta's reported $130-plus billion in AI infrastructure spending, arguing the deal supports finer-grained measurement of whether that investment is paying off ^[3]. In that light, Sundial is not primarily a customer-facing product Meta wanted to own - it is an internal instrument for making sense of the output of Meta's own AI systems at the pace those systems now generate data.

The flip side of folding Sundial into Meta is that it stops serving everyone else. Reporting flags that current customers face product discontinuation with no announced data-portability or migration timeline ^[5]. That list reportedly included OpenAI ^[3], whose relationship with the tool is now uncertain once the product stops serving outside customers. The acquisition therefore reads as acqui-hire-shaped: Meta wants the team and the technology for itself, and is comfortable with the product disappearing for the market it used to serve.

Sundial's pitch was not just a chatbot over a data warehouse - it rested on a semantic layer defining shared business metrics consistently, work serious enough that Sundial joined Snowflake's Open Semantic Interchange initiative to help standardize data-metric definitions across AI tools ^[6]. Commentary on the deal argues this semantic consistency is what separates a trustworthy AI analytics agent from a dangerous one: a human analyst knows to ask clarifying questions, while an agent that guesses at an ambiguous metric produces a confident, wrong number ^[3]. A YouTube breakdown of the deal struck a similar note, characterizing it less as a contest over model quality and more as a fight over who controls the data infrastructure and metric definitions that AI systems reason over - a dimension of the AI race that gets far less attention than raw model capability.