Snowflake Inc. SNOW

Key data

The business

Snowflake rents companies a place to keep all their data and charges them for the computing they burn every time they run a query against it. It is pure consumption, not per-seat: the meter runs on usage, so revenue rises when a customer analyzes more, and falls when they tune their queries to spend less. The data itself sits on Amazon, Microsoft or Google clouds underneath. The moat is data gravity, once a company's records and the dashboards, pipelines and permissions built on them live inside Snowflake, ripping them out is a year-long project nobody volunteers for. The newest lever is Cortex, the AI layer that lets customers ask questions of that data in plain language, plus the February 2026 purchase of Observe, which pulls machine and application logs into the same platform. The thing a customer actually holds is an invoice that scales with how curious they were last month.

The numbers

Revenue is reaccelerating while the loss margin shrinks, which is the whole story in one line. Quarterly sales grew 35% year over year in the July quarter, up from the low-30s a year earlier, and the net loss margin went from 26% of revenue to 12%.

The Q4 FY26 figures are derived as the fiscal year less the first nine months, since that quarter is not in the vendor feed. The inflection is Q2 FY27: revenue jumped $156M sequentially, the biggest step in the series, and the loss margin halved. Gross margin held at 67%.

Growth ran at 29% for two straight years and has now stepped back up toward 35%, while the full-year loss finally stopped widening. That is the compounding the price demands. The last watch-item on this name asked whether growth would stay above 30% while stock pay's share of sales fell; this quarter answered on the compounding side, with growth at 35% and the loss margin halved, not on the overpaid side.

The catch sits in the cash. The company throws off real free cash flow, but it hands staff more in stock than that cash is worth, then spends more still buying stock back to keep the share count from ballooning.

Stock pay is estimated at 30% of sales, the trailing-twelve-month rate. On the company's own adjusted earnings the stock runs about 80 times FY2029 estimates, but those estimates add the $1.4B of stock pay back as if it were free. The variant here: the market prices a re-accelerating compounder that converts to real profit, and the single print that settles it is whether the GAAP operating loss keeps halving while revenue holds above 30%.

Management

Insiders sold $299M of stock across 87 transactions over the last twelve months and bought nothing. Chairman and former CEO Frank Slootman took out $46M across two sales in July and August, and co-founder Benoit Dageville sold $16.8M in August. The plan status is not disclosed, so scheduled selling cannot be separated from discretionary here, which matters because a lump sum hides which. The buyback has slowed from $1.9B in FY2025 to $0.87B in FY2026, and its main job is mopping up the dilution from the $1.4B of annual stock compensation rather than shrinking the count. Pay is tied heavily to that stock, so a falling multiple hits employee retention, not just holders.

How it fails or surprises you

Consumption gets optimized. Because the meter runs on usage, a customer that tunes its queries or shifts cold data elsewhere spends less without leaving. If net consumption growth slips toward 25% over the next two or three quarters, the revenue line stalls even as logos hold. The first print is sequential revenue: another $150M step confirms demand, a flat one signals tuning.

Stock pay never closes the gap. The clearest strike against the bet is in its own cash table: $1.41B paid to staff in stock against $1.12B of free cash flow. If growth eases toward 25% while that pay stays near 30% of sales, the GAAP loss stops shrinking and the "profit" stays a non-GAAP fiction funded by dilution. Watch operating margin and stock pay as a share of sales together.

AI reignites consumption (right tail). Cortex and the Observe logs give customers new reasons to run compute against data already sitting in Snowflake, a second wave the market treats as unproven and does not pay for at 20x sales. If AI workloads push revenue growth back above 40% and drag operating margin across zero within a year, the multiple and the earnings re-rate at once. The tell is growth accelerating and the operating loss turning positive in the same quarter.

Closing thoughts

Two or three more quarters like the last one, and it's settled. The loss margin has gone from 26% to 12% of revenue in four quarters; if growth holds above 30% while that margin keeps shrinking at the same pace, Snowflake crosses into GAAP profit within a year and the 20x-sales price looks ordinary in hindsight. An ambiguous print, growth in the low 30s but the loss stuck near 12%, leaves you paying a premium multiple for a business that runs on stock nobody counts as a cost, and the fatter tail then is down, because the peer sells for four times sales and the floor is a long way below $337. What is actually at risk if consumption softens is the whole re-rating; what the right tail is worth, if AI does what the bulls claim, is the same multiple on a much larger and finally profitable base.

The bet is still that companies keep pouring data into Snowflake and keep paying to run queries against it, and that AI makes them run more. It breaks if consumption growth drifts toward 25% while the stock pay stays near 30% of sales, and the one pair of numbers that tells you first is quarterly revenue growth against the operating loss margin. Stop believing the story the quarter growth falls under 30% and the loss margin stops shrinking in the same report.