Sridhar Ramaswamy on Snowflake, AI Competition and Leading Through Change

Snowflake CEO Sridhar Ramaswamy argues that AI infrastructure and applications increasingly overlap, creating opportunities and risks for companies building on model providers. In this February 2025 conversation with Harry Stebbings, he discusses leadership, enterprise AI, competition and product distribution. His assessment of Snowflake and its rivals reflects his role leading the company.

Ramaswamy did not imagine becoming a CEO while studying for his bachelor’s degree or PhD. He once considered becoming a professor, then moved into software engineering when research became less appealing. His career advice is to aspire broadly while taking small steps and recognizing opportunities.

For young graduates, including his own sons, he suggests finding work they care about that society also values. AI is changing knowledge work, so he urges people to be open, adaptable and willing to put in the effort to learn rather than deny the change.

He expects software engineering and other knowledge professions to be affected substantially, but says it is too early to know whether engineering becomes narrower or more valuable. As software spreads into more fields, he still sees opportunities for people who embrace the technology.

With age, physical challenges have become harder for him, and he acknowledges that time to immerse oneself in an entirely new field is limited. He remains committed to learning and adapting mentally, and sees AI tools as a way to explore more subjects.

Asked how an intense CEO can bring a team along, he says people need to understand the scale and fleeting nature of an opportunity. He describes high expectations at Snowflake as a choice about the kind of mission and growth employees want to pursue.

Not everyone scales with a growing organization, he says, and personal circumstances change. When a team doubles, strengths that made a manager effective at the previous size may become obstacles. He regards it as his responsibility to give people time and opportunities to adapt.

Ramaswamy says leaders should not postpone difficult conversations: delay can worsen a conflict and deprive the other person of useful feedback. He tries to state expectations respectfully and directly, acknowledging that the conversation is awkward and that he may not have every answer.

A reduced scope does not always mean a person should leave. He recalls a Google manager whose responsibilities became too broad as a business expanded; he narrowed the role so the person could focus, and says the manager later became a successful vice president.

He rejects the idea that richer leaders are necessarily better. Wealth can make a leader less sensitive to risk, while decisions affect employees, customers and shareholders as well as the executive. He says leaders need to remember those different interests.

Stebbings asks where lasting value will be created as AI models improve and applications become easier to build. Ramaswamy says the boundary between infrastructure and applications is blurry: a model provider or large technology company may enter a startup’s category itself.

He sees value in companies with customer relationships and a clear product that adopt AI fast enough to defend themselves. He places Snowflake in that category because it helps customers collect, analyze and use data. That is the CEO’s strategic view, not proof that the company cannot be disrupted.

Ramaswamy argues that OpenAI’s consumer opportunity is not solely about having the best or cheapest model. He points to ChatGPT’s product features, distribution and user relationships as potential sources of staying power. His comparison with Anthropic is an opinion made at the time, not a final ranking of the companies.

Stebbings asks whether a free alternative such as DeepSeek could pull users away. Ramaswamy distinguishes a model from a complete product with features such as image creation, uploads and code execution. His claim that users would not switch is a prediction about behavior, not evidence that switching cannot happen.

He praises OpenAI’s ability to attract users and publicity, while saying other model developers have challenged assumptions about how difficult progress must be. He separates the company’s consumer product success from uncertainty about individual models and future ambitions.

He agrees that building a startup on top of a model provider can be uncomfortable when that provider may launch competing applications. Stebbings asks whether Nvidia could similarly enter Snowflake’s data layer. Ramaswamy says the company must also watch cloud providers that already offer competing data products.

He argues that capital alone cannot reproduce a product with strong market fit, while acknowledging that Snowflake must keep innovating. He calls Databricks a credible competitor and says the two companies have different strengths. These comments come from the CEO of a direct competitor.

Asked whether Databricks is ahead in AI workloads, he distinguishes established machine learning from newer generative AI. He concedes Snowflake had been catching up in machine learning, but claims it was ahead on several AI fronts. The interview offers no independent benchmark to verify that competitive ranking.

He describes Snowflake’s work on data transformation, unstructured and structured data, and an agentic framework called Snowflake Intelligence. He says customers were using Snowflake AI in production and that the company still needed to communicate its capabilities more widely.

Ramaswamy says a public company faces constraints a private competitor may not, including scrutiny of free cash flow and quarterly results. He makes critical claims about Databricks’ spending and sales organization; these are competitor allegations in the interview, not independently verified facts.

He argues that constraints can create clarity: rather than give the AI team a blank check, Snowflake focused on specific work. Public-market volatility can also affect employees whose finances depend on their stock, so he says the company must operate responsibly.

Asked whether he would take Snowflake private, he says no. Public visibility provides liquidity and forces the company to show its work, he argues, even when market reactions are uncomfortable. His criticism of a competitor’s changing narrative is his own interpretation.

On enterprise AI adoption, he expects growth to be gentler than an uninterrupted hockey stick, but says the technology already creates value. He uses note transcription and summarization after meetings and points to internal chatbots that answer questions using structured data.

At Davos, he says executives asked for practical utility: what can the technology do for their organizations? He describes combining structured and unstructured information to help automate parts of insurance underwriting. That is an example of a potential workflow, not evidence that every deployment produces a positive return.

Stebbings observes that incumbents appear to be innovating quickly. Ramaswamy thinks technology leaders have learned from previous platform shifts and failures, and have both an incentive and the capital to respond to AI.

He argues that major companies adapted reasonably well to mobile, even though the shift also created new businesses such as Uber and Lyft. At Google, he recalls his team working for years to bring mobile monetization closer to desktop levels.

Asked whether the infrastructure spending boom is a productive bubble or a waste, he says it is too early to know. Buildings and power can retain value for other uses; rapidly depreciating hardware can lose value quickly. He expects some innovation opportunities to remain for specialized companies.

He believes a provider such as OpenAI cannot build every workflow and gives Harvey as an example of a company he thinks can defend a niche. This is his view of future competition, not a guarantee about either company.

On Snowflake’s growth, he describes widening the product scope from analytics into data ingestion, engineering, machine learning and AI access. He says acquisitions can support that work, but rejects an unrelated acquisition-led roll-up in favor of product innovation.

He says Snowflake paid about $150 million for Neeva, the company he founded, and believes the purchase has more than paid for itself. His judgment is not an independently calculated return. As Neeva’s founder and Snowflake’s CEO, he has a personal and corporate interest in the assessment.

He expects AI and applications built by customers on Snowflake to become meaningful sources of future revenue. He cites data applications built by major financial and technology customers and wants Snowflake to participate in the value those customers create. The revenue impact remains a forecast.

Asked whether AI will favor a few general models or many specialized ones, he looks to search history. He says Google built a central consumer entry point through deliberate distribution deals and then incorporated services such as image search into the main product.

For consumer AI, he thinks ChatGPT has the strongest starting position as an entry point. Enterprise workflows, he argues, are more fragmented and may support more specialization. Both are his predictions about a changing market.

He recalls Yahoo and AOL as important early distribution partners for Google and says the company at one point paid AOL more than it earned from those referrals. His broader lesson is that a good product benefited from deliberate business and distribution decisions.

At Google, he remembers Larry Page and Sergey Brin pushing discussions relentlessly and examining assumptions. He carries forward that focus on first principles, energy and finding the right business outcome.

In the quick-fire round, Ramaswamy calls “founder mode” a shorthand for effective leadership rather than something reserved for founders. On parenting, he says presence matters more than searching for a perfect book or tool.

He believes people often underestimate their capacity to change. Looking back, he has few regrets about taking risks in his career, and credits his spouse for supporting decisions that involved moving and changing roles.

He admires Salesforce’s brand and says Snowflake also needs to explain itself beyond its reputation as a data warehouse. The hardest part of his role is balancing the drive for team results with treating people well.

Asked what he is proudest of, he says being a good father. He and his wife tried to be present, set an example and help their children become thoughtful and independent. He closes by emphasizing that leadership and family both require attention to people.