Databricks’ AI growth and the changing software interface
Databricks’ numbers are striking, but their most useful lesson is narrower than the usual “AI will kill SaaS” argument. On Monday, the company announced a $5.4 billion revenue run rate, growing 65% year over year, with more than $1.4 billion from AI products. TechCrunch reported the figures and spoke with co-founder and CEO Ali Ghodsi about what they mean. They describe one company’s performance; they do not establish a general market trend.
Ghodsi’s point was that AI had increased usage at Databricks, rather than simply eroding the value of software businesses. At the same time, he wanted the company seen as an AI company, not just as a SaaS vendor. That tension is worth examining. AI can be a new product category, but it can also change how customers reach software they already depend on.
For companies considering AI investments, the practical question isn’t whether SaaS is dead. It’s whether a new interface makes an existing workflow materially easier, and whether the system underneath remains useful. Databricks offers a case to study, not a ready-made forecast for companies or the wider software sector.
Databricks connects AI to existing data
Databricks is known for its cloud data warehouse, where businesses store information for analysis. In TechCrunch’s account, Ghodsi pointed to Genie, an LLM-based interface, as a driver of warehouse usage. Instead of writing a query in a specialized language or relying only on a prebuilt report, users can ask questions in natural language. Ghodsi’s example was asking why warehouse usage and revenue spike on particular days.
That changes who can approach the data. An employee who doesn’t know a query language might still be able to ask a useful question. The warehouse is not replaced in this account; the path to it becomes more conversational. Ghodsi attributed some usage growth to Genie, but the report does not quantify Genie’s contribution to revenue or establish that it explains the company’s overall growth.
There’s a familiar enterprise lesson here: valuable data isn’t valuable to every team if access is difficult. A language interface may lower that barrier. It doesn’t, by itself, guarantee that the answer is correct, that the user has appropriate access, or that an organization should let a model take action. Those are implementation decisions, not conclusions supported by the revenue figures.
Systems of record may outlast familiar screens
Ghodsi argued that businesses are unlikely to discard core systems of record and replace them with quickly built AI-generated alternatives. These systems hold essential records across areas such as sales, customer support, and finance. He described them as difficult to move and noted that model makers are not offering databases to become those systems of record.
The nearer-term challenge, in his view, is the user interface around those records. People may use natural language rather than navigate a vendor’s screens; agents may connect through APIs or plugins. A system can remain important while its familiar interface becomes less central to daily work.
That could change what protects a software incumbent. Ghodsi described workforce familiarity with products such as Salesforce, ServiceNow, and SAP as a historical moat. If people increasingly interact with software through language, that learned familiarity may matter less. The product could become less visible even while its underlying data and functions remain necessary.
That is not a prediction that every incumbent will lose. Databricks itself is an example of an established platform adding an LLM interface, with Ghodsi connecting that product to greater usage. But a well-known screen isn’t an unbreakable defense, either. AI-native competitors may build for models and agents from the start. Customers will still judge whether those products work for them; neither an AI label nor an incumbent’s history settles the question. See also why enterprises need an intermediary layer.
How do AI startups make money?
The Databricks example offers a grounded, if incomplete, answer to “how do AI startups make money?” They sell a product customers use and pay for, ideally tied to a valuable workflow or resource. Here, the reported example is not simply a chatbot detached from business operations. It is an AI product that gives users another way to work with a data warehouse.
TechCrunch reported that more than $1.4 billion of Databricks’ run rate came from AI products. The story does not break that amount down by product, explain pricing, or provide product-level margins. So the figure signals significant reported activity, but can’t tell us how profitable those sales are or how much came from Genie specifically.
For a startup, a model is not a business on its own. The product has to solve a problem well enough to earn ongoing use, and its economics need to work for both provider and customer. Databricks’ report illustrates the importance of connecting AI to a product customers already have reason to use. It doesn’t supply a universal revenue formula for founders. Relatedly, how AI startups can make money depends on the specific product and customer value delivered.
The report also describes Lakebase, a database designed for agents. Ghodsi said that after eight months in the market it had generated twice the revenue the data warehouse had at the same age. He qualified the comparison by calling both products “toddlers.” That qualification matters: it is an early comparison from the company’s CEO, not evidence that agent-oriented databases will generally outgrow established warehouses.
Still, Lakebase points to the kind of product question founders may pursue: if agents need access to data and software, what infrastructure will they use? A product built around that need could be a business, but early traction is not proof of durable demand. Buyers and investors should separate the possibility from the evidence available so far.
Funding gives Databricks room, not a blueprint
TechCrunch reported that Databricks closed a previously announced $5 billion funding round at a $134 billion valuation and secured a $2 billion loan facility. Ghodsi said the company wasn’t immediately preparing another raise or an IPO. He described the capital as protection if markets weakened again, giving the company years of runway.
That is relevant context for a venture-backed technology company, but it should be kept in proportion. A large round and a loan facility can give Databricks room to invest; they do not prove that capital alone creates customer demand. Nor does this financing establish that AI startups generally can access similar terms. The report’s figures describe Databricks’ financing, not a market-wide pattern.
For founders, the useful comparison isn’t the size of the cheque. It’s the relationship between product use and the business. Does the tool address a real bottleneck? Does it fit the data and systems a customer already relies on? Does it bring more users into a workflow, as Databricks says its AI products have helped do? Those questions are more useful than treating a valuation as validation of every AI business model.
Enterprise leaders can apply the same discipline when assessing vendors: ask what the product changes in the work itself, and what evidence supports the claimed benefit. Databricks’ scale is not a benchmark every company should chase. Its funding story is specific to the company and should not be mistaken for a financing playbook for startups.
What enterprise buyers should watch
The case raises three practical questions. Can natural-language tools make data platforms accessible to employees who don’t write queries? Can a company preserve its system of record while changing the interface people use? And will products designed for AI agents become durable businesses, rather than promising early experiments? The TechCrunch report provides examples and Ghodsi’s views; it doesn’t settle those questions for the market.
For enterprises, an AI feature deserves scrutiny beyond the demo. Which data can it reach? Who can use it? What happens when its answer is wrong? Does it make a recurring workflow easier, or merely add a new surface to an old one? Those are buying questions, and the answers will differ by organization.
Databricks’ reported $5.4 billion run rate is a strong signal about Databricks. It is not a verdict on SaaS as a whole. The more defensible takeaway is that AI can be both a product line and a new route into an established platform. Enterprise software may change dramatically at the point of interaction while the systems beneath it continue to matter.
Source: TechCrunch report on Databricks’ growth, AI products, and financing.
databricks ai growth and the changing software interface
databricks ai growth and the changing software interface