From $1 million to $7 billion in ARR in 10 years! Databrick, a rising star in the AI industry, explains the core competitive advantages of enterprise-level AI.

From less than $1 million to over $7 billion in annual recurring revenue, Databricks has achieved a rare expansion in enterprise software over a decade. According to Ron Gabrisko, the company's Chief Revenue Officer, what truly underpinned this growth was not a single model, but rather data context, corporate governance, usage-based pricing, and a sales system that was continuously rebuilt as revenue levels increased.

Databricks is a cloud-based unified data, analytics, and artificial intelligence (AI) platform. Founded in 2013 by the original team behind Apache Spark, its core purpose is to help enterprises break down data silos and integrate the storage, processing, business intelligence (BI) analysis, and machine learning/AI workflows of massive amounts of data onto a single platform.

Turner Novak, host of the tech startup interview program The Peel, recently had a lengthy conversation with Ron Gabrisko, Chief Revenue Officer of Databricks. Gabrisko joined the company when its revenue was less than $1 million and subsequently led the development of its global sales and market expansion system, guiding Databricks to grow its annual recurring revenue to over $7 billion. This experience gave him familiarity with the company's early product-market fit process and direct involvement in globalization, pricing, and organizational expansion.

Ron Gabrisko's vision for Databricks is straightforward: enterprises aggregate data scattered across different systems and then use this data for analysis, prediction, and AI applications. Content recommendation on streaming platforms, fraud detection and loan approval in banks, and new drug development in pharmaceutical companies all involve the same fundamental task: transforming more data into more accurate judgments. Databricks aims to integrate data engineering, analytics, machine learning, and artificial intelligence into a unified platform, while simultaneously addressing permissions, governance, and operational efficiency.

Three anti-consensus bets lay the foundation for growth.

All seven co-founders of Databricks hold PhDs from UC Berkeley. Ron Gabrisko attributes Databricks’ growth over the past decade to three early choices made by the seven co-founders: fully committing to cloud computing, adhering to open source, and putting data and artificial intelligence on the same platform.

Today, these directions have become an industry consensus, but in the early stages of a company's establishment, most large enterprises still deployed their core systems on-premises. Banks and financial institutions were concerned about the security, regulation, and scalability of cloud services, and many CIOs even believed that critical data would never be migrated to the cloud.

Ron recalls visiting major Wall Street financial institutions in the company's early days, attempting to develop large clients. At that time, many bank CIOs explicitly stated they would "never go to the cloud." However, ten years later, these banks have become one of Databricks' largest client groups. Ron draws a parallel between this historical context and today's corporate reluctance to adopt AI, arguing that resistance to new technologies often stems from security, regulatory, and legacy system inertia, rather than problems with the technology itself—"Once you solve the challenges of security, governance, and compliance, the market will explode."

Databricks' decision to offer only cloud services is essentially a bet that enterprise infrastructure will eventually migrate entirely. This allows companies to avoid maintaining traditional on-premises software and scale along with the cloud infrastructure. Open source has helped projects like Spark, Delta, and MLflow quickly gain traction with developers, but Ron Gabrisko emphasizes that open source users don't automatically become paying customers. The sales team still needs to understand the difficulties users encounter in actual deployments and why they are willing to pay.

Customer feedback focused on managed services, security, reliability, governance, and scalability. Databricks thus provides an enterprise-grade platform built on top of open-source technology, transforming complex systems that customers previously had to assemble and maintain themselves into ready-to-use services. Open source expands the technological scope, while commercial products address management and operational challenges in production environments.

The moat of enterprise-level artificial intelligence lies in the data context.

In the face of a rapidly iterating model market, Ron Gabrisko does not believe that a single model can constitute a lasting barrier.

As businesses gain access to a wider range of models, their capabilities tend to converge. What truly determines the quality of AI output is the model's ability to connect to the company's proprietary data and accurately understand customers, revenue, processes, and industry rules. Without this context, the model can only provide generalized answers; with context, it can generate predictions and action recommendations directly relevant to the business.

This is also the position Databricks hopes to occupy long-term. The company started in the data science and machine learning market, and subsequently used capabilities like Unity Catalog to unify the management of data, models, notebooks, and other assets. Model vendors may change, but an enterprise's data, permission structures, and business semantics will persist. Whoever can securely organize these assets and allow different models to access them within the same governance framework is closer to the core infrastructure of enterprise AI.

The interview revealed that Databricks' revenue from AI-related products alone has reached approximately $1.7 billion. This figure indicates that enterprise customers are willing to pay for production-grade AI, but they are not buying general-purpose chat tools detached from their own data; rather, they are purchasing systems capable of performing queries, calculations, predictions, and agent-like tasks in a controlled environment.

Context control selection and cost determine whether something can enter production.

Ron Gabrisko summarized the most important concerns for enterprises when deploying artificial intelligence into four aspects.

First, there's the context . Artificial intelligence must access enterprise data in order to understand the business and improve the reliability of predictions.

Secondly, there's control . Enterprises need to understand who has access to which data, models, and agents, and how to use that data, as well as how to prevent the leakage of sensitive information. For regulated industries such as banking, healthcare, and government, access, auditing, and compliance are prerequisites for products to enter the production environment.

The third issue is choice. Complex tasks may be better suited to cutting-edge closed-source models, while other tasks may be better suited to open-source models. Enterprises also want to retain cloud platform options to avoid being locked into a single vendor.

The fourth issue is cost . Artificial intelligence projects can consume a lot of computing resources in a short period of time, and management must set budgets, usage rules and guardrails, and continuously measure the business value created by each application.

These constraints haven't reduced the application scenarios. Ron Gabrisko mentioned that clients have already used AI to accelerate new drug discovery, automate loan and fraud detection, and help retailers optimize promotions, shelf space, delivery, and inventory. The real challenge lies in connecting the model to enterprise data in an efficient and controllable manner, and continuously monitoring the quality of the output.

Turn Databricks into Client Zero

Databricks used its own business system as the first use case for its artificial intelligence products.

Ron Gabrisko uses Genie to predict revenue, claiming the prediction error can be controlled to approximately 1% to 2%. He also has the system identify customers most likely to churn, analyze the reasons, and provide remedial suggestions, or determine which products have stronger stickiness and which accounts still have room for growth. Before client meetings, Genie can also organize account information, product usage, and potential opportunities.

Unlike traditional dashboards that only display preset metrics, Genie allows employees to ask questions directly in natural language, with the system handling queries, calculations, charts, forecasts, and model calls in the background. All sales staff can use this tool on their mobile phones to reduce administrative work such as forecasting, data entry, and account management, and to demonstrate products directly to customers.

This zero-customer mechanism serves another purpose. Companies first redesign their sales processes using their own products, and then bring the proven practices to customers. Artificial intelligence is therefore not merely a tool for automating the original processes, but rather prompts teams to rethink which tasks should be done by humans, which judgments can be delegated to systems, and how to improve efficiency while retaining necessary oversight.

Growing from millions of dollars to 7 billion dollars requires four sales systems.

Ron Gabrisko believes the most common mistake in enterprise software sales is starting to sell before understanding the customer. Large enterprises typically involve budget managers, business units, technical teams, management, and approval processes, with sales cycles potentially lasting 6 to 12 months or even longer. Effective sales must first understand the customer's organizational structure and business problems before explaining how the product can reduce costs, shorten timelines, or mitigate risk.

Sales organizations cannot use the same approach across all revenue stages.

When revenue grows from zero to $10 million or $20 million, the primary task is to find a product-market fit; when moving from $20 million to $100 million, the company needs to establish repeatable sales strategies; when moving from $100 million to $1 billion, international markets, partners, and channels begin to become growth priorities; once the company reaches a scale of several billion dollars, the leadership team, culture, systems, processes, and continuous innovation determine whether the organization can continue to expand.

In its early days, Databricks hired about 40 sales staff per quarter.

This wasn't about blindly scaling up without demand, as Spark already had a large user base. The sales team's task was to reach out to open-source users, understand why customers were willing to pay, and identify the industries and company types most likely to buy. They were both generating revenue and undertaking customer discovery.

The company’s revenue subsequently grew from less than $1 million to about $13 million to $15 million, and then crossed $50 million, $100 million and $250 million.

This experience also illustrates that product-driven growth cannot replace enterprise sales. Open-source users are willing to use the technology, but they may not be proactive in purchasing commercial services. Sales teams must proactively connect free usage with chargeable features such as hosting, security, and governance to help customers move from individual or small team trials to enterprise purchases.

Complex products must be sold by people with technical expertise.

Databricks requires its sales staff to be able to personally demonstrate the product. Its primary buyers are data engineers, data scientists, and other technical professionals, making it difficult to build credibility if salespeople cannot understand their clients' problems. Therefore, Ron Gabrisko considers technical skills, experience in the data and AI industries, pay-as-you-go sales experience, startup experience, and resilience under pressure as key hiring criteria.

Candidates are required to submit a business plan, present their vision for artificial intelligence to the CEO, and complete a Genie demo tailored to their industry. Ron Gabrisko is cautious about resumes with a history of frequent job changes within two years, as such candidates may not have experienced the challenging phases of a company. During background checks, he avoids vaguely asking whether a candidate is excellent, instead requesting references to provide relative rankings, willingness to work together again, and specific performance under pressure.

This hiring standard serves Databricks' enterprise sales model. Salespeople need to bring their knowledge to the customer site, able to discuss business outcomes and explain product deployment. For complex technology companies, sales is not just packaging the product, but rather an integral part of helping customers understand and adopt it.

Free proof-of-concept for expanding services

Most of Databricks' early proof-of-concept and pilot projects were free.

Ron Gabrisko believes the goal of a pilot program is to demonstrate the value of the team and product, not to generate revenue from short-term projects. However, free doesn't mean indefinite investment. Before a project begins, the capabilities to be demonstrated, clear success criteria, and completion deadlines must be agreed upon, and sufficient support from high-level management is required to ensure that a successful pilot program can move into formal procurement.

This approach forms a typical path for implementation and expansion. Databricks first proves its value in a department or use case, cultivates internal advocates, then expands to other tasks within the same department, subsequently covering more departments, and eventually entering the entire enterprise. Whether the pilot program is charged is not the key point; the key point is whether it can be converted into larger, long-term contracts.

Volume-based pricing allows revenue to grow in tandem with customer value.

Databricks' early annual contracts ranged from approximately $15,000 to $18,000. To increase the average order value, the company initially increased platform and user fees, but the per-seat pricing model led clients to limit the number of users, actually reducing overall product usage. After eliminating user fees, more employees could directly use the platform, resulting in a rapid increase in overall usage.

Based on this, Ron Gabrisko developed a pricing principle: prices should be as close as possible to the value received by the customer.

If the price is higher than the value, customers won't buy; if the price is significantly lower than the value, the company won't get a reasonable return. For cloud services and AI software, computational volume, query volume, tokens, or other real usage are generally more appropriate than the number of seats. As data volume, query demand, and agent activity increase, the value received by customers and the revenue of suppliers can grow in tandem.

Enterprise-level requirements must be completed in the early stages of product development.

Many software companies initially serve startups and tech companies eager to experiment, but to build a multi-billion dollar business, they must eventually enter the banking, healthcare, retail, consumer goods, and government markets. These clients conduct system audits of security, compliance, permissions, auditing, and data segregation. If these requirements are not addressed early in the product development process, retroactive implementation often slows down sales and product development.

Data governance is also a fundamental obstacle to the implementation of artificial intelligence.

Enterprise data is often scattered across legacy systems, old formats, and proprietary formats. New tools can help build data pipelines, discover datasets, and add tags, but bringing data into a location where it can be securely accessed by artificial intelligence remains a complex undertaking. Technology platforms, business domain knowledge, and governance capabilities need to be integrated into the production environment simultaneously.

International expansion amplifies these requirements. Ron Gabrisko advises companies to first acquire a small number of overseas clients remotely, and then invest heavily once their sales strategies in the domestic market are largely stable. Europe, the Middle East, and Africa cannot be considered single markets, and Japan and India also have different languages, cultures, and purchasing habits. Local managers are familiar with clients and talent, while experienced staff from headquarters possess product knowledge and internal expertise; both types of personnel need to work together to build regional teams.

Frontline deployment engineers connect technology and business

Enterprise AI projects need people who understand both business goals and technical implementation.

Databricks' frontline deployment engineers are capable of programming, building products, conducting pilots and setting up environments, and working closely with the marketing team. They need to determine what customers truly want to achieve, choose the appropriate technology path, and consider upgrades, costs, and maintenance in advance.

Ron Gabrisko believes that a good solution cannot rely on a large external team repeatedly rewriting it in the long run. It must be scalable, cost-effective, and ultimately maintained by the customer's own team. The value of frontline deployment engineers lies in their ability to transform one-off, customized projects into a sustainable, operational product.

Going public is just one step in the growth process.

Regarding the long-standing public interest in the IPO timeline, Ron Gabrisko stated that Databricks' IPO is only a matter of time. The company already operates like a publicly traded company, regularly disclosing its financial situation, holding board meetings, and having its business reviewed by an audit committee, but management is not in a hurry to complete this process.

Ron Gabrisko sets the company's long-term goal at the trillion-dollar level, making the IPO just one part of the journey. Compared to that goal, $7 billion in annual recurring revenue doesn't mean growth is nearing its end. He describes the company's current stage as the first half of the second inning of a game, meaning enterprise AI is still in its early stages, and Databricks needs to continue updating its product, organizational, and market strategies.

Ron Gabrisko aspired to be a professional baseball player in his early years. Although this plan didn't materialize, the habits of training, competition, and perseverance he developed have carried over into his management work. In his view, what truly matters is whether the team remains willing to take responsibility under pressure. Databricks' decade-long expansion has proven that technology choice can determine the starting point, but turning technology into sustainable revenue requires long-term investment in sales, pricing, governance, and organizational capabilities.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.