Snowflake results reflect enterprise demand for alternatives to Anthropic in AI coding

Snowflake's quarterly results, released Wednesday, have become a signal of a broader shift in the enterprise artificial intelligence market: the search for alternatives to depending on a single language model provider. While headlines about AI for coding are usually dominated by Anthropic, OpenAI and Cursor, Snowflake, a data management company, has shown with its figures that many other providers can offer competitive products in this space.

According to the article, this trend responds in part to Anthropic having increased its pricing power earlier this year, which has led corporate CIOs to look for ways to control spending with both Anthropic and OpenAI. Rather than being locked into a single model provider, companies prefer coding tools capable of connecting to multiple language models.

Snowflake's coding assistant, CoCo (formerly known as Cortex Code), saw its customer accounts grow at least 28% quarter over quarter in the period ended in July, reaching 9,100 accounts (with the caveat that different teams within the same company can open several accounts). According to a Snowflake spokesperson, CoCo currently uses Anthropic and OpenAI models on its backend, and is also integrating open-weight and open-source models such as DeepSeek-V4-Flash and GLM-5.3.

On the earnings call with analysts, Snowflake's executive vice president of Product, Christian Kleinerman, explained that "many, many customers" have said that after making significant purchase commitments with a particular model provider, they ended up concluding that they "should have been able to use other models," and that choosing Snowflake gives them that flexibility, something the company considers one of its main advantages. This view aligns, according to the article, with that of executives at Palantir, Microsoft and Salesforce, who agree that companies need a trusted intermediary connecting their business with OpenAI and Anthropic.

Snowflake CEO Sridhar Ramaswamy also noted that the top-tier AI capabilities built into CoCo have significantly reduced the difficulty of migrating data at scale from legacy systems to modern databases such as Snowflake's, thereby reinforcing the company's core data management business.

The report also points to a broader industry dynamic: cloud software providers that charge based on actual resource consumption are in a more favorable operating position than those relying on legacy per-user applications. Snowflake's product revenue accelerated for the third consecutive period in the July quarter, growing 37%, allowing it to sidestep for now the debate over outcome-based pricing that does affect competitors such as Salesforce. Claude Code, the market leader in AI for coding, operates mainly on a per-seat subscription model, although for large customers it increasingly incorporates usage-based billing.

The text recalls that earlier this summer it was already reported that, faced with rising Anthropic bills, some customers had begun to evaluate alternatives or build their own coding tools. It is unclear whether Claude Code will manage to hold its dominant position, but the article lists several moves by large software companies entering the AI coding arena: Atlassian launched its coding agent for Jira in July, and Salesforce introduced Slack Code in August. Even OpenAI has lowered its own barriers, allowing customers to use non-OpenAI models within its Codex coding assistant. The article notes that, although it is technically possible to connect Claude Code to other AI models, Anthropic does not encourage the practice.

For Snowflake, CoCo works not only as a standalone product but as a commercial lever: CFO Brian Robbins said on the call that customers using CoCo tend to buy more of the company's data management products. The market rewarded these results with a jump of nearly 17% in Snowflake's share price on Thursday, bringing its year-to-date gains above 60%.

The article also includes a second piece of information, from Cohere's research division, on the reach of AI agents in the U.S. labor market. A study titled "The Early Footprint of Automation," completed in May, analyzed the public tools developed by programmers for AI agents aimed at automating tasks. The research found that existing tools already target work associated with roughly half of the 923 occupations catalogued by the U.S. Department of Labor (the study did not include internal, non-public tools developed by companies themselves).

The report's authors note that, seen from the supply side, there is still a large number of occupations for which no automation tools have been developed, though they acknowledge that the fact that nearly half of all occupations are already being targeted by AI agents may come as a surprise. Among the occupations at greatest risk of displacement, the study cites graphic designers and web developers, as well as certain medical roles such as biostatisticians and clinical data managers, for which AI agent tools already exist that can handle highly specialized aspects of the work. According to Cohere Research head Marzieh Fadaei, this suggests that the number of jobs in those occupations could shrink, while the salaries of those who keep those positions could actually rise.

Taken together, both parts of the article sketch the same pattern: on the one hand, large software companies are capitalizing on their customers' desire not to depend on a single AI model provider, becoming trusted intermediaries between businesses and giants such as OpenAI and Anthropic; on the other, agentic automation is already affecting, albeit unevenly, a very significant share of U.S. occupations, with effects that could show up sooner in employment levels than in the wages of those who remain in those roles.