Snowflake's $6bn AWS bet shows who powers enterprise AI

Snowflake’s latest quarter looked, at first glance, like another AI-adjacent software rally. The more important detail sat inside the company’s expanded AWS collaboration: a five-year commitment to spend $US6 billion (about $9.2 billion) with Amazon Web Services, alongside a broader push to run more workloads on Graviton processors and AWS GPU capacity. For enterprise buyers, that is a cleaner signal than the share-price reaction. Not in marketing copy. In committed cloud spend.

On the numbers alone, Snowflake had room to make that case. Reuters reported that the company lifted its FY27 product-revenue outlook to $US5.84 billion (about $8.9 billion) after first-quarter revenue reached $US1.39 billion (about $2.1 billion), up 33 per cent. That supports the insider view that enterprise AI is moving beyond pilots and into budget lines.

Underneath the quarter sits a second reading, and by the third paragraph it is the more interesting one. Stronger guidance is one thing, signing away billions in future cloud spend is another. Snowflake is proving that AI demand can lift data platforms, while also showing how much of that lift still depends on hyperscaler economics.

Snowflake’s pitch in this cycle is not that it builds the biggest model. It is that enterprises already keep governed data inside the platform, and that agents become useful only when they can operate against that data securely, at scale and with enough compute behind them. In that sense, the quarter fits the broader capex story digitalblog has been tracking from the software layer. The spending is not only flowing to chipmakers and data-centre operators. It is also flowing to the software companies that can turn cloud capacity into production AI workflows, and to the clouds that still collect rent on nearly every serious deployment.

Why the AWS commitment matters

At $US6 billion, the pact is large enough to work as strategy rather than supplier housekeeping. Snowflake said the deal expands Marketplace collaboration, regional availability, migration support and infrastructure for agentic AI, all on top of a customer base that already leans heavily on AWS. That does not read like a temporary buying spree. It reads like a company deciding that the fastest path to monetising AI demand is to go deeper into the cloud platform where much of its install base already lives.

Snowflake chief executive Sridhar Ramaswamy framed the move as an execution step rather than a branding exercise in the company’s announcement:

AI has generated enormous excitement, but for enterprises, the real challenge and opportunity is turning intelligence into action.
— Sridhar Ramaswamy, Snowflake

From the analyst side, that optimism has limits. A richer top-line outlook only partly answers the question of who captures the economics. If Snowflake needs to commit $US6 billion to secure the right mix of CPUs, GPUs, migration help and marketplace reach, then its growth is not becoming less dependent on AWS as AI expands. It is becoming more visibly tied to AWS’s roadmap and pricing.

The skeptic view goes further. TechCrunch argued that the agreement is notable precisely because it favours Amazon’s home-grown Graviton chips, while Sherwood noted that Snowflake is still balancing competitive pressure from hyperscalers with the cost of AI build-out. That does not mean the deal is a negative for Snowflake. It means the quarter is best read as evidence that cloud-native software still grows inside hyperscaler gravity. The software layer may be capturing new demand, but it is not escaping the landlord.

Why Graviton changes the economics

For all the attention on GPUs, Graviton is the more revealing detail. Much of enterprise AI spending will not land on model training clusters. It will sit in the less cinematic work around them: data preparation, warehousing, orchestration, inference support, governance and repeated queries across large stores of enterprise information. If Snowflake can push more of that workload onto processors with better price-performance, it has a clearer answer to the analyst’s main question about whether AI demand can translate into durable software economics.

AWS chief executive Matt Garman made the production-deployment argument explicit in the same Snowflake release:

Enterprises are rapidly moving from experimenting with AI to putting intelligent agents to work.
— Matt Garman, Amazon Web Services

From a builder’s viewpoint, that shift matters. Snowflake cited Hex as an example of customers building on the partnership, and Hex co-founder and CTO Caitlin Colgrove said Snowflake on AWS already acts as a foundation for customers trying to move quickly with data. Elsewhere in the same cohort, SiliconAngle reported that Unravel Data is launching optimisation tooling across Snowflake, Databricks and BigQuery, another sign that enterprises now want help managing AI-era data workloads across multiple platforms rather than merely buying access to a model.

Even so, the skeptic’s question is not resolved, only partly answered. Better CPU efficiency can help Snowflake control costs and make more AI workloads commercially sensible. But every Graviton win is also a win for Amazon. The Register’s reporting on AWS’s new Graviton-powered Redshift instances underlines the same point from a neighbouring product line: Amazon is steadily turning custom silicon into a lever for data-warehouse performance and economics. If that pattern holds, Snowflake’s upside is real, but it arrives through infrastructure designed, priced and ultimately owned by its cloud partner.

Seen from Australia, that is why the quarter looks more important than an earnings beat. It suggests the AI cloud race is broadening from chip suppliers and model vendors to the data platforms that sit closer to enterprise workflows. Yet it also suggests the winners in that layer may be the companies most willing to sign up for deeper, longer and more expensive relationships with the underlying clouds. For Australian CIOs and developers tracking where enterprise AI budgets are heading, Snowflake’s result is a reminder that the software story and the infrastructure story are now the same story told from different altitudes. AI demand is lifting enterprise data platforms. It is also binding them more tightly to the hyperscalers they depend on.