Snowflake spent eleven years convincing enterprises to trust it with their data, and last week it signed a deal that bets the next contest is about controlling what AI agents do with that data once they start acting on it. On May 27, the company announced it would acquire Natoma, a two-year-old startup that builds an enterprise gateway for the Model Context Protocol, the standard AI agents use to call tools, APIs and applications. The terms were not disclosed.
It landed inside a single, busy day. Snowflake also reported first-quarter fiscal 2027 product revenue of $1.33 billion, up 34% year over year, and signed a five-year, $6 billion compute commitment with AWS. The stock jumped roughly 38%. Natoma was the smallest of the three announcements by dollar value and the most revealing about where Snowflake thinks the next platform fight sits.
What Natoma Actually Does
The Model Context Protocol, introduced by Anthropic, has become the common way to connect an agent to the systems it needs to do work, and that connectivity is where the governance problem starts. As an enterprise wires up dozens of MCP servers, agents start calling tools across email, Slack, CRMs and databases with inconsistent permissions, thin audit trails and no central place to set policy. Snowflake points to its own research finding that 96% of organizations still struggle to scale AI across the enterprise, and frames the gap as shadow AI and data exfiltration risk.
Natoma sits in front of that mess as a centralized MCP gateway. Before an agent takes an action, the platform checks who requested it, what permissions apply and whether the action is allowed, then logs it. It ships with a verified library of more than 100 prebuilt MCP servers and handles authentication, fine-grained authorization and monitoring for what the company calls non-human identities. Founded in 2024, Natoma had raised a $7 million seed round from Greylock and Index Ventures and employed 27 people. Chief executive Pratyus Patnaik sold his prior company, atSpoke, to Okta in 2021 and spent two years there working on identity, which is the lineage that matters here.
Why Snowflake Wants The Control Plane
Snowflake’s pitch is that intelligence without governance creates risk, and that agents need context, permissions and policy guardrails rather than raw data access. Once the deal closes, the company plans to wire Natoma into Cortex Agents, Snowflake Intelligence and its Cortex Code assistant, so those tools can reach SaaS apps, cloud environments, VPCs and on-prem systems through one governed gateway. Chief executive Sridhar Ramaswamy framed the result as Snowflake products becoming a single interface for daily work, querying data, updating CRM records and composing email included.
The strategic move is the extension of Snowflake’s governance perimeter from data assets to the actions agents take. That is a meaningful change in what the company sells. Constellation Research analyst Michael Ni argued that data platforms won the analytics era and whoever governs agents, context and autonomous actions wins the agentic one, casting Natoma as the missing layer between insight and execution. The same logic explains why nearly every large vendor, from application suites to orchestration players, is now racing for the same runtime governance position.
The Limitations Worth Naming
The deal has not closed, and the press release carries the usual forward-looking caveats about integration risk and customer retention, which carry real weight given the mismatch in scale. Natoma is 27 people and a seed-stage product being folded into a company managing a $6 billion infrastructure commitment in the same quarter. Identity-and-access products live or die on breadth of integrations and battle-tested edge cases, and a young gateway has not yet been stressed at the scale Snowflake customers will demand.
There is also a positioning tension I keep coming back to. MCP is an open standard precisely so agents are not locked to one vendor’s tools. A governance layer that lives inside Snowflake risks pulling that openness back toward a single control plane, which is good for buyers who already standardize on Snowflake and awkward for everyone running a genuinely multi-vendor agent stack. And the category is crowded. Okta, where Patnaik came from, is chasing non-human identity directly, while Astrix, Venafi and a wave of MCP-gateway startups want the same chokepoint, so Snowflake has bought a head start into a fight that is far from settled.
What Decision-Makers Should Take From It
If you run an enterprise data estate on Snowflake, this is a reason to slow down before buying a standalone agent-governance tool, because a native option is coming and consolidation will be cheaper to operate. If you run a multi-cloud or multi-vendor agent strategy, treat Natoma as one bet in a contested market and keep your gateway choice portable until integrations and audit depth are proven in production.
The broader signal is about where value is migrating. As the reasoning model turns into a commodity input, the position worth defending becomes the layer that decides what an agent may touch and keeps a record of what it did. Snowflake just paid to own that layer for its own customers, and the scale of the surrounding announcements suggests it expects the agentic control plane, rather than the data warehouse, to define its next decade.
