Inside the Agentic Stack: Why Supabase and Salesforce Are Buying AI Infrastructure, Not AI Features

The first wave of generative AI was about features: a chat box, a summary button, an assistant in the sidebar. Features are easy to copy. The second wave is about agents that act on their own, and agents put very different demands on the software underneath them.

An agent might spin up a fresh database for a single task, call three different models to finish it, charge a customer, and then disappear. Multiply that by millions and the bottleneck is no longer the model. It is the infrastructure that has to keep up. That is where the strategic buyers are spending.

On 2 October 2026 Supabase announced it is acquiring Turso, alongside a separate US$150m funding round led by GIC with CapitalG, IronArc and SquarePeg. The purchase price was not disclosed. Our full teardown is in Supabase to Acquire Turso.

The numbers behind the deal explain it. Supabase says it now adds more than 4 million new databases a month, and roughly 70% of them are created by AI agents and coding tools rather than people.

• The old modelOne dedicated database server per project. Fine for human developers who create a handful of projects. Far too slow and expensive when agents create millions of short-lived databases.
• The Turso modelA rewrite of SQLite that runs enormous numbers of lightweight, isolated databases on shared infrastructure, in Turso’s cloud or inside a customer’s own cloud account (BYOC). Cost tracks actual use instead of provisioned servers.
• The combined offerPostgres for the core application, Turso for the swarm of per-agent databases around it. Turso founder Glauber Costa becomes Supabase’s Head of Agentic Services.

The unit economics are the point. When a database costs almost nothing to create and idle, an agent can have its own one for every task. That keeps data isolated and simple, and it turns database volume into a growth engine instead of a cost problem.

Salesforce signed a definitive agreement (opens in a new tab) to acquire Listen Labs on 29 September 2026, with closing expected in Salesforce’s fiscal fourth quarter (November 2026 to January 2027). Terms were not disclosed.

Listen Labs’ agents design research studies, recruit from a network of more than 50 million participants, interview them in more than 120 languages and synthesise the findings, turning months of qualitative research into days. Annualised revenue grew 15 times in the nine months after launch, according to its January 2026 Series B announcement.

Why does a CRM want this? Systems of record capture what customers did. Agents also need to know why. Owning a scaled engine for real customer conversations, and simulating how customers will respond, gives every Salesforce agent better context than a competitor can scrape. Our analysis: Salesforce to Acquire Listen Labs.

Read together with Stripe and OpenRouter and Salesforce and Fin, the direction is clear. The platforms with distribution are buying the layers agents cannot work without.

• Measure machine usageIf agents already drive a growing share of your usage, as at Supabase, document it. Agent-created volume is now one of the most persuasive numbers in an infrastructure sale.
• Own a layer, not a featureStorage, routing, billing, identity and first-party data are hard to replicate. Thin features on top of someone else’s model are not.
• Expect strategic termsMost of these deals were undisclosed and founder-retaining. Strategic buyers are paying for teams and architecture as much as revenue.