Snowflake's CoCo Turns Consumption Into AI Revenue Proof

Background

Snowflake (SNOW) sells storage and compute on a consumption basis, billing customers in credits tied to query volume and data processed rather than per-seat licenses. The company has layered an AI and agent stack on top of its data platform, including CoCo, a coding agent billed on token consumption, and CoWork, positioned as part of the "system of intelligence" for what CEO Sridhar Ramaswamy calls the agentic enterprise. The stock has surged roughly 91% over six months to the low $320s, with Q2 FY27 earnings due September 2, 2026.

Key Findings

The Bear Case on Consumption Has Inverted

The prevailing assumption since the 2022-2023 optimization cycle is that usage-based pricing is a liability: when customers cut costs, consumption revenue falls immediately, while seat-based SaaS enjoys contractual insulation. That logic ran in reverse then, and it runs in reverse now. AI agents that query data continuously generate incremental credits with every interaction, and Snowflake captures that volume automatically, with no renegotiation, no upsell motion, no per-seat AI SKU to sell. Seat-based peers face the opposite problem: agent usage explodes but revenue is capped by license counts, forcing awkward repricing exercises. The same mechanism that made Snowflake fragile in a cost-cutting cycle makes it the cleanest AI revenue pass-through in software.

CoCo Is Metered Revenue, Not a Loss-Leader Feature

Most enterprise software vendors have bolted AI assistants onto existing products as retention tools or flat-fee add-ons, making the AI's actual P&L contribution invisible. Snowflake did the opposite. CoCo is billed on token consumption, with usage charged pay-as-you-go and tokens billed as AI feature credits on both on-demand and capacity contracts. CoCo interactions also trigger standard Cloud Services compute when querying metadata, adding a second consumption stream. The company built per-user quotas, budgets, and observability tables that log each interaction, token count, and cost, which is billing infrastructure, not demo infrastructure. Oppenheimer's specific citation of strong CoCo consumption in its recent target commentary suggests this metering is already registering meaningful usage, and five major sell-side firms have raised targets on AI-driven workload demand across infrastructure software.

Ramaswamy's Agentic Enterprise Pitch Is a Consumption Pitch in Disguise

The skeptical read on Ramaswamy's media circuit, including his recent Six Five appearance, is that "agentic enterprise" is marketing gloss on a data warehouse. But the architecture he describes maps directly onto the billing model. His argument is that the bottleneck in the agentic era is not storage but making the right data visible and accessible to models at the moment of decision, with coding agents as the foundation and the data platform as an orchestration layer spanning multiple models, formats like Iceberg, and clouds. An orchestration layer where agents continuously query data and invoke tools is, mechanically, a credit-generation machine. The strategic pitch and the monetization mechanism are the same thing, which is rare in current AI positioning.

The Financials Already Show an Inflection, Ahead of the Agent Story

The bear framing holds that Snowflake's growth deceleration is structural and AI is a narrative overlay. The verified numbers say otherwise. Revenue growth reaccelerated from 25.7% year over year in Q1 2026 ($1.0B) to 31.8% in Q2 2026 ($1.1B) to 33.5% in Q1 2027 ($1.4B), a rare re-steepening for a company at this scale. Gross margins held in a tight 66.5% to 67.8% band across five quarters, meaning the incremental AI workloads are not degrading unit economics. The caveat is real: Snowflake remains GAAP unprofitable, posting a -$295.6M net loss and -$0.86 diluted EPS in Q1 2027. But the operating margin trajectory improved from -42.9% in Q1 2026 to -23.4% in Q1 2027, so the reacceleration is coming with narrowing losses, not widening ones.

September 2 Is the First Hard Test, and the Bar Is High

The near-term X chatter treats SNOW as a technical setup, a "former leader" consolidating near its 21-day EMA after the run. That framing misses what actually matters: Q2 FY27 on September 2 is the first quarter where agent-driven consumption, CoCo specifically, can show up in product revenue at scale. There have been no new fundamental announcements or filings in recent days; the stock is digesting a 91% six-month gain into the print, with insider selling by former CEO Frank Slootman disclosed just before this window. A stock priced for AI inflection needs product revenue and consumption commentary that validates Oppenheimer's read. Anything that sounds like "AI adoption is early" reprices the multiple, not the thesis.

Implications

Snowflake

Snowflake becomes the rare software name where AI agents show up directly in revenue rather than in a slide deck. The strategic imperative is execution on the September 2 print: management needs to quantify agent-driven consumption, or at minimum let the product revenue number and guidance do the talking. The token-metered CoCo billing architecture also gives Snowflake a template to monetize every future agent product (CoWork included) without inventing new pricing, compounding the structural advantage as the agent portfolio expands.

Seat-Based SaaS Peers

Vendors monetizing per seat face an inversion of their historical edge. Agents multiply usage but not licenses, forcing them to retrofit consumption-style AI pricing onto contracts built for humans. Snowflake's model requires no such transition, which turns the agentic shift into a relative competitive repricing across enterprise software, not just a Snowflake-specific catalyst.

Sell-Side and Investors

Oppenheimer's CoCo consumption callout, echoed by five firms raising targets on AI workload demand, has front-run the evidence. That makes the September 2 print asymmetric: confirmation validates a new monetization framework for the whole consumption-model cohort, while a miss lands on a stock up roughly 91% in six months with GAAP losses still near $300M a quarter. Investors treating this as a moving-average trade are underweighting a binary fundamental catalyst eight days out.

Conclusion

The consumption model that made Snowflake a casualty of the optimization era has flipped into its clearest AI advantage: agents query continuously, credits accrue automatically, and CoCo's token-metered billing routes that activity straight into product revenue. The reacceleration from 25.7% to 33.5% growth alongside narrowing operating losses shows the platform inflecting before the agent story fully lands, and Oppenheimer's consumption callout suggests CoCo is contributing rather than pending. Ramaswamy's agentic enterprise framing is credible precisely because the architecture and the billing model are the same mechanism. September 2 decides whether this is a proven monetization edge or a well-priced hypothesis, but the structural point stands either way: in an agent-driven stack, usage-based pricing stops being the bear case and becomes the distribution channel for AI revenue.