.
Image: Getty Images
Enterprises have often dreamed about AI systems that can reason across their most sensitive data, execute multistep tasks, and explain their logic while remaining inside a highly governed environment.
Snowflake and Anthropic are betting they can finally crack the code.
Through a multiyear, $200-million (R3.4 billion) expansion of their agentic AI partnership, the companies plan to deliver an operational “control plane” that uses Anthropic’s latest Claude models, such as Sonnet and Opus 4.5, to power enterprise intelligence.
The announcement landed alongside Snowflake’s Q3 earnings for fiscal year 2025, which showed the company maintaining strong momentum. Snowflake reported $1.21 billion in revenue, up 29% year over year, driven by $1.16 billion in product revenue. The company now operates at a $100 million AI run rate (year to date) while adding a record 615 new customers.
But as the race to dominate enterprise agentic AI accelerates, not everyone is convinced that Snowflake’s momentum guarantees staying power. “Snowflake is still in the early innings of seeing if the traction will stick,” says William Falcon, founder and CEO of Lightning AI. “For a database like Snowflake, they’ve hopefully learned from others’ mistakes and invested in Anthropic to try and avoid similar problems.”
That scepticism frames what makes Snowflake’s approach so interesting. Instead of treating AI as an external service that companies must funnel their data toward, the company wants the intelligence layer to reside where the data already lives. Its philosophy is to “bring AI to the data.”
“By deeply integrating Claude into Snowflake Intelligence and Cortex AI, we’ve collapsed that sprawl into a single governed environment where the model runs directly where a company’s data already lives, securely with full business context and without ever moving that data or introducing risk,” says Vivek Raghunathan, senior vice president of engineering at Snowflake.
The hallmark of this collaboration is a new class of AI agents capable of multistep reasoning on governed corporate data through Snowflake Intelligence, the company’s enterprise intelligence agent powered by Claude Sonnet 4.5. Under the hood of Snowflake Intelligence sits Cortex Agents, the Snowflake Horizon Catalog, and a layer of semantic models.
Analysts can ask complex questions in natural language, developers can build intelligent agents without stitching together infrastructure, and business teams can get deep insights with citations and traceability.
In practice, the integration means a business user can ask a natural-language question, such as “What is driving churn in our Northeast customer segment?” and Claude will determine which datasets are relevant, write and execute the SQL, and explain how it arrived at its conclusion. In highly regulated industries such as healthcare, financial services, or life sciences, that combination of deep reasoning with end-to-end governance is especially transformative.
“In regulated environments, ‘here’s the answer’ isn’t enough. You need ‘here’s how I got there’,” says Katelyn Lesse, head of API at Anthropic. In areas like financial reconciliation, companies routinely juggle data from disparate systems that rarely align cleanly, with exceptions that demand human judgment.
Lesse noted that earlier approaches either overlooked this nuance or relied so heavily on manual review that any promised efficiency gains disappeared. “Claude can work through those discrepancies and flag where it’s uncertain, which is just as important as getting the answer right.”
Enterprises can also design custom multi-agent systems, with Snowflake Cortex Agents serving as the scaffolding for production-ready data agents powered by Claude. These agents can retrieve, interpret, and reason across structured and unstructured data with greater precision and efficiency.“We don’t see or access customer data because Claude operates within Snowflake’s security perimeter, so the customer’s data stays private,” Lesse added.
Raghunathan notes that Snowflake uses Claude internally across engineering, sales, and operations. Developers rely on Claude Code to accelerate development and code production cycles, while its sales teams use a Claude-powered assistant to unify data across the organisation and shorten deal timelines.
The companies say that early customer results are already showcasing what an enterprise built around AI agents might look like. Customer communications platform Intercom now uses Claude through Snowflake Cortex AI to power its Fin AI Agent. Likewise, Simon Data, a composable customer data platform, uses Claude on Snowflake to unearth patterns that conventional analytics overlooked, while maintaining governance across customer datasets.
The race to dominate enterprise agentic AI has intensified pressure across the technology landscape. Snowflake’s rising AI revenue has seized market attention, but experts argue that its strategy, while meaningful, does not fundamentally reshape enterprise AI’s competitive frontier. Gregor Stewart, chief AI officer at SentinelOne, believes the Anthropic alliance strengthens Snowflake but does not vault it ahead of rivals.
“Databricks has a stronger internal team and just as good relationships and arrangements with the frontier labs. In some ways, Snowflake is just catching up to them,” he adds.
“I see hyperscalers using models and generic compute to build ‘one-size-fits-all’ assistants that lack the specific business context residing in the data layer. In contrast, Snowflake is positioning itself as the governed brain where the actual work happens, rather than just the infrastructure where the model runs.”
The positioning, AI where the data lives, is the philosophical gulf separating Snowflake and Anthropic from Databricks, Microsoft’s Copilot ecosystem, Google’s integrated cloud stack, and AWS.
The companies are betting that enterprises will increasingly favour systems that minimise data movement, maximise security, and deliver reasoning directly within existing governance boundaries.
“Enterprises have been burned by AI projects that demanded new infrastructure, new skills, new risk, and delivered unclear ROI. Snowflake’s revenue run rate validates that ease of adoption beats raw capability,” says Ian Riopel, CEO of Root. “Against Databricks: ‘intelligence in SQL’ beats ‘build custom pipelines’; Against Microsoft and Google: ‘AI in your existing flow’ beats ‘adopt our new flow.’ The reality most vendors miss is that enterprises aren’t looking for another platform to master; they want 100 times the efficiency with the same knowledge and access their employees already have.”
If that thesis proves correct, Snowflake and Anthropic may be constructing more than a partnership—an architecture for how enterprise software will work over the next decade. In that vision, agentic AI doesn’t sit beside business systems; it becomes the operating system. And both companies are intent on owning the moment when enterprises decide to make that shift.
“Our AI strategy is inherently open. We support models from several leading providers so enterprises can orchestrate multi-agent systems without being locked into a single cloud or model provider,” Raghunathan added. “This is what makes Snowflake a true AI control plane.”
ABOUT THE AUTHOR
Victor Dey is a tech analyst and writer covering AI, data science, startups, and cybersecurity. A former AI and tech editor at VentureBeat, his work has also appeared in New York Observer, Entrepreneur Magazine, HackerNoon, and more.
FAST COMPANY