Meta has acquired Sundial, an AI-native analytics startup founded by former Meta executives, Axios reported on October 23, 2026. A Meta spokesperson confirmed the deal; no price was given. Sundial's team, including co-founder Chandra Narayanan, joins Meta, and the product will not continue as a standalone offering.
It is a small deal by Meta's standards, but it shows where AI spending is heading: from building models to measuring whether they pay off. This post sets out what the available sources confirm, what they do not, and what to watch. We could not open Axios's original article, so the details come from Sundial's own earlier announcement, the founders' public statements and secondary write-ups, and we mark which is which.
TL;DR: what builders and analysts are asking
| Question | Answer |
|---|---|
| Is it confirmed? | Yes; a Meta spokesperson confirmed to Axios |
| Price? | Not disclosed |
| Who runs it? | Folded into a new Meta data analytics org led by Alex Schultz |
| Is the product continuing? | Not as a standalone product, per reports |
| Total funding raised by Sundial | $23 million (Series A of $16 million in July 2025) |
| Customers | OpenAI and Gamma are reported as past clients; their future status is unclear |
| Why now? | Meta's AI spending is above $130 billion this year, per reports |
What Sundial built
Sundial was founded in 2021 and, per its July 2025 funding announcement, described itself as an AI analytics platform combining dashboards, notebooks and data pipelines in one product. That release put total funding at $23 million, including a $16 million Series A led by DJ Patil of GPV.
Julie Zhuo, a co-founder and former Meta vice president of design, summed up the pitch on X: Sundial "automated away many of the obstacles involved in getting to meaningful analysis," listing cleaning data, creating semantic layers, writing SQL and storytelling. A semantic layer is the shared definition of business terms, such as what counts as an active user, that lets AI tools and people query data consistently. It is the unglamorous piece that decides whether an AI agent answering a business question gets the right number.
Secondary coverage adds that Sundial supplied tools for semantic layers and SQL writing to AI companies including OpenAI and Gamma, and that it joined the Snowflake-led Open Semantic Interchange initiative in April 2026, which aims to standardize how metrics are defined across AI tools. We could not verify those details against a primary document, so treat them as reported.
What Meta said and did not say
Confirmed by reports: Meta acquired the company; a spokesperson confirmed to Axios; the team joins Meta; Sundial will be part of a new data analytics organization under Alex Schultz, who shared this in a post on LinkedIn.
Not stated in anything we could read: the price, the size of the team joining, whether all employees are staying, and what happens to current customers. The last point is the live question. If OpenAI was a customer, as reports say, a customer relationship with a competitor of Meta becoming part of Meta's stack is unlikely to continue unchanged. Neither company has commented in the sources we found.
Why would Meta want an analytics company?
Reports tie the deal to Meta's AI spending, which they put above $130 billion for the year. Posts on X this week claimed Meta reported its first negative free-cash-flow quarter since 2012 on that spending, a claim we have not checked against Meta's filing, and commentary noted that knowing which models, clusters and ad products actually generate returns is becoming as important as building them. That is an inference by commentators, not a Meta statement, but it is a plausible reading: a company spending at that scale needs fine-grained measurement of cost per workload and return per product.
Analytics also has an internal-use logic. Sundial's founders came from Meta's growth and data culture, so the team knows the company's data habits. A tool that lets non-analysts ask questions in plain language over governed data is exactly what a large company with thousands of product teams wants, and a team that already built it saves time over a from-scratch effort.
Why a semantic layer matters for AI agents
It helps to see the problem Sundial targeted. Ask an AI agent "how many active users did we have last week?" and the answer depends on definitions that live in people's heads or scattered SQL: what counts as active, which time zone, whether test accounts are excluded, and which table is the source of truth. A human analyst knows to ask. An agent that guesses produces a confident, wrong number.
A semantic layer writes those definitions down once, in a form both people and software can read, so every dashboard and every agent computes the metric the same way. Cleaning messy data, defining metrics and writing correct SQL are the three chores that eat most of an analyst's week, and they are the three Zhuo listed in her description of the product. Automating them with AI is attractive for any large organization, and especially for one whose teams ship thousands of experiments and need consistent measurement across them.
The risk is the usual one with AI analytics: a fluent answer that is quietly wrong. The remedy is the same as for any model output, which is to check results against known numbers and keep humans responsible for the definitions. Whether Meta's version of the tool will expose that kind of auditing is not something the announcements address.
Meta's pattern of small AI acquisitions
This is the latest in a run of smaller buys that add capability rather than headline scale. Last month Meta acquired Swedish startup Stilla for agent tool integration, covered in Meta acquires Stilla and first reported by Axios. Earlier, Meta's purchase of Manus drew attention, with an unwinding story we followed in Manus independent operations resumed. Meanwhile Meta has been shipping its own models and products, such as Muse Wonder and a Muse Spark 1.3 update, and using them internally; see Muse Health Research for one example of agents working with large data.
The common thread is that Meta is buying small teams with focused skills and folding them into internal organizations. Whether that yields products outsiders can use, or stays internal, is not answered by any of these announcements.
What people are asking
Does this change anything for Sundial's current users? Probably yes, soon. Reports say the product will not continue standalone, so current customers should expect a wind-down or migration notice. Nothing in the coverage gives dates; check your vendor communications and plan an export of dashboards, definitions and queries.
Is Meta building an enterprise analytics product? Nothing indicates that. The reports describe an internal data analytics organization. Meta sells advertising and increasingly AI features; an analytics product for outside businesses would be a new direction and has not been announced.
Does it say anything about AI agents and data? It underlines how much agent reliability depends on data definitions. Agents that query warehouses fail in boring ways when metric definitions are inconsistent, which is why semantic layers are getting attention. Our coverage of Databricks MemCurrent touches on the same data-context theme for agents.
Is the $130 billion figure verified? It comes from news summaries describing Meta's AI infrastructure spending expectations. Check Meta's earnings materials for the exact guidance.
What to do if you use AI analytics tools
- Write down your metric definitions. If your semantic layer lives only inside a vendor's product, export it in a portable form, so a change of ownership does not leave you rebuilding your business logic.
- Watch for acquisition clauses. Check whether your contract has data portability and notice terms if your analytics vendor is bought.
- Test AI answers against known numbers. Keep a short list of metrics you can verify by hand, and run your AI analytics tool against it regularly.
- Prefer open standards where you can. Initiatives like the Open Semantic Interchange exist to make definitions portable across tools.
What would settle the open questions
- A Meta or Sundial post stating the terms for existing customers and timelines.
- Axios's original report, which may contain details not repeated elsewhere.
- Any disclosure of price in a filing, though a deal this size is unlikely to appear in one.
- Signals about whether the capability becomes a product beyond Meta.
Bottom line
Confirmed: Meta bought Sundial, the team is joining, and the product is folding into a new internal data analytics organization. Not confirmed: price, headcount, customer treatment, and any plan beyond internal use. The deal is small, but it points at the unglamorous layer that decides whether AI answers about a business can be trusted.
Details are accurate as of October 23, 2026 and are drawn partly from secondary reports of Axios's exclusive.