answrsnow
Revenue Operations · Explainer

How Does Snowflake Handle the Cost of Anthropic and OpenAI Inference at Scale in 2027?

By Kory White · Updated Oct 2026 · 8 min read
A Claude chat window open on a laptop
Direct answer

Snowflake does not run Anthropic and OpenAI inference at a flat price. It resells it through Cortex AI and bills customers in AI Credits per million tokens, with a separate rate for input and output tokens on each model. The rates sit in the Snowflake Service Consumption Table, and an AI Credit costs $2.00 with global routing or $2.20 with regional routing, per Snowflake's pricing page.

On its own side, Snowflake has been open that this is a margin question. On the Q2 fiscal 2027 call, the CFO said fast-growing AI workloads "carry a lower contribution margin today" and the company set full-year non-GAAP product gross margin guidance at 74%. Management's levers are model choice, an "auto" mode that routes each task to a suitable model, and open models it runs itself. What Snowflake pays Anthropic and OpenAI is not published. The two $200 million partnership announcements do not give payment terms.

For a buyer, the practical answer is to treat tokens like a metered utility: pick the cheapest model that does the job, watch usage in the account views, and cap spend with the controls Snowflake documents. The sections below show where each figure comes from, and say plainly where a number is not public.

How Cortex AI bills LLM calls

Snowflake's cost documentation says Cortex AI functions incur compute cost based on the number of tokens processed. A token is roughly four characters, though that varies by model. The details that move a bill:

The pricing page separates AI Credits from Platform Credits. It lists $2.00 per AI Credit with global routing (cross-region inference allowed) and $2.20 with regional routing, where requests stay in the account's home region or chosen geographies. ACV-based discounts on AI Credits apply automatically, but the page says capacity discounts do not apply to AI Credits.

Prices change. Everything below is read from the Service Consumption Table effective October 15, 2026. Check the current table before you budget.

What Claude and OpenAI models cost in AI Credits

Table 6(a) of the Snowflake Service Consumption Table lists AI_COMPLETE rates in AI Credits per one million tokens. A sample of the Anthropic and OpenAI rows, as input then output:

Multiply by the AI Credit price above to get dollars. The table is long and moves with each new model, so use it for the exact row you plan to call. Three patterns stand out:

Prompt caching has its own table. For claude-sonnet-4-6, Table 6(b) lists 1.50 for input, 7.50 for output, 1.875 for cache writes and 0.15 for cache reads, so repeated prompt prefixes can be re-read at a fraction of the input rate. The pricing page confirms that prompt caching is supported with reduced-rate cache reads.

Some rows carry promotional pricing. A footnote on one OpenAI row, openai-gpt-5.6-sol, says its price may increase by 25% on input and 50% on output on November 21, 2026. Other footnotes name December 1, 2026 and January 1, 2027 for other providers' models. If a workload depends on a promotional rate, plan for the step-up.

The Anthropic and OpenAI partnerships

Both deals are headline-sized, and neither release states what it costs Snowflake in a given quarter.

Not published: the payment terms of either $200 million agreement, what Snowflake pays per token to either company, and any AI revenue run-rate. The Anthropic release states no pricing, the OpenAI release states no pricing, and management gave no run-rate figure on the Q2 call.

What Snowflake has said about the margin impact

In its Q2 fiscal 2027 results on September 2, 2026, Snowflake reported product revenue of $1.49 billion, up 37% year over year, and non-GAAP product gross margin of 74.7% for the quarter. Its outlook for full-year non-GAAP product gross margin is 74%.

On the earnings call (an auto-generated transcript, so wording may contain errors), the CFO tied that outlook to a higher mix of fast-growing AI workloads that carry a lower contribution margin today. The CEO said cost has become a real concern as models get more powerful, and described three responses:

That is the closest Snowflake comes to describing how it absorbs inference cost: it pairs each task with a cheaper model where it can, runs open models on its own capacity, and relies on vendor price cuts. The calls did not quantify the cost of Anthropic or OpenAI inference separately.

Cost controls customers can use

These are the controls Snowflake documents. None of them changes the per-token rate; they change how much you consume and who can consume it.

The AI Functions cost-management guide does not cover budgets or resource monitors, so do not assume a hard cap exists on AI Functions without building one.

What this means for buyers

Questions people ask

How does Snowflake charge for Claude and OpenAI models?

In AI Credits per million tokens, with separate input and output rates per model, listed in the Snowflake Service Consumption Table. An AI Credit is $2.00 with global routing and $2.20 with regional routing, per the pricing page.

How much does Snowflake pay Anthropic and OpenAI?

It has not published that. Each partnership was announced as a multi-year, $200 million agreement, but the releases do not state payment terms or per-token costs.

Does Snowflake support prompt caching for Claude and OpenAI models?

Yes. The pricing page says prompt caching is supported with cache reads billed at a reduced rate, and Table 6(b) of the Service Consumption Table lists cache write and cache read rates per model.

Can I set a hard spending cap on Cortex AI Functions?

Snowflake's AI Functions cost guide shows how to build alerts, per-user monthly token budgets and runaway-query cancellation, but it does not describe a built-in hard cap, budgets or resource monitors for AI Functions. Resource budgets are documented for Cortex Agents.

Is AI hurting Snowflake's margins?

Management says AI workloads carry a lower contribution margin today. Non-GAAP product gross margin was 74.7% in Q2 fiscal 2027, and the full-year outlook is 74%, per the results release and the earnings call.

Sources

Prices, models and cost controls change often. Figures here were read from the linked pages on October 17, 2026; confirm against the current Snowflake documents before you commit budget.