Snowflake shares closed 36 per cent higher on 28 May, the data-cloud company's best single day since it went public. Two things moved them. The fiscal first-quarter numbers beat Wall Street, and Snowflake committed to spend US$6 billion with Amazon Web Services over five years, including AWS's custom silicon for AI work.
The quarter
Revenue rose 34 per cent year on year to US$1.39 billion, ahead of the US$1.32 billion analysts expected. Adjusted earnings came in at US$0.39 a share against a US$0.32 forecast. Snowflake lifted full-year product-revenue guidance to US$5.84 billion, about 31 per cent growth, and widened its non-GAAP operating margin by three points to 12 per cent.
Why the AWS deal mattered to the market
A US$6 billion outlay would normally read as a cost, but investors took it as a demand signal. The logic is that Snowflake is buying that much compute — much of it on Amazon's own chips — only because its customers are running enough AI workloads on the platform to require it. Finance chief Brian Robins told analysts that AI tools such as the company's Cortex Code product were driving what he called a step-change in AI revenue, per SiliconANGLE. That is a vendor characterisation; the revenue line will show whether it holds.
The wider read
Snowflake's jump helped lift the broader software sector the same day, per CNBC. The market has spent two years punishing software companies for AI spending that shows no clear return. A large compute commitment tied directly to revenue was the proof point investors wanted to see.
The revenue line has not answered yet
The finance chief's step-change in AI revenue was flagged above as a vendor characterisation that the revenue line would test. Three months on, that test has still not been run in public.
At the same May results, Snowflake guided second-quarter product revenue to between US$1.415bn and US$1.420bn, about 30 per cent year on year. The actual figures were due in late August and had not been published as of 23 August.
The market repriced Snowflake on 28 May based on a beat, a raised guide, and a spending commitment. None of those is a confirmed result. The first quarter that could validate the Cortex Code claims was still pending when the shares jumped 36 per cent.
The AWS commitment read as demand. It is also a fixed obligation
Investors read the US$6bn as a demand signal, reasoning that Snowflake would not buy that much compute without customers to serve. The logic holds only if that demand arrives on schedule.
Disclosure since has shown how large that bet has become across the sector. Five hyperscalers have reported roughly US$1.2 trillion of lease obligations, of which about US$725bn has not yet commenced and therefore sits in the notes rather than on the balance sheet, with Goldman expecting around 35 per cent of 2027 AI capital expenditure to be debt-funded.
A multi-year compute commitment is a demand signal when consumption keeps pace, and a fixed cost when it lags. The contract is the same either way; which reading applies is settled only in hindsight. The disclosure practice across the sector puts much of that exposure where a balance-sheet reader does not look.
The private comparable repriced in the same quarter
Snowflake's multiple is marked daily. Its closest private competitor's is set in a funding round, and that round happened in August.
Databricks raised US$5bn at a US$190bn valuation on a US$7bn run-rate, with its chief executive using the occasion to argue that AGI had been reached, that it was not superintelligence, and that the world was largely unchanged apart from the inference bill.
Two companies in the same layer re-rated upward in the same quarter, one by public market and one by private round. That pattern is more consistent with a sector-wide repricing of data infrastructure than with a verdict on any single product line, which is a reason to be careful about reading the 28 May move as the market validating Cortex Code specifically.
Separately, the 2024 breach reached a guilty plea
The most consequential Snowflake story of the quarter for anyone running data on the platform has nothing to do with earnings.
Connor Riley Moucka pleaded guilty on 6 August over the campaign that took data from more than 165 organisations using Snowflake, including over 100 million AT&T call and text records. No Snowflake vulnerability was ever involved. The entry route was stolen credentials against accounts that had no multi-factor authentication.
The plea closes a criminal matter and changes nothing about the exposure. Every ingredient of that campaign is still available to the next person who tries it, and the control that would have stopped it is a setting, not a purchase.
View author profile → · Editorial policy
