Why Klarna chose graph DBs to shed SaaS

Welcome to Runtime! Today: Neo4j co-founder and CEO Emil Eifrem explains how Klarna used graph databases to chart its own enterprise software path, Oracle buries its head further in the sand as customers start asking about two recent security breaches, and the latest funding rounds in enterprise tech.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

Time for a little graph theory

One reason enterprise software companies are rushing to provide their customers with generative AI agent-building platforms is that they are terrified about the ramifications of those customers using other agent-building platforms to replace their lucrative SaaS businesses with homegrown alternatives. Like everything in this world, the truth is a little more complicated, but it does seem clear we are at a turning point for enterprise software and things could look very different in a few years.

Last month Klarna CEO Sebastian Siemiatkowski took to X to clarify earlier statements about how the Swedish fintech company used generative AI to replace its enterprise software vendors, which rattled the market last year. It turned out that Klarna wasn't using large-language models to replace SaaS apps; instead, it embraced graph databases built by Neo4j, co-founder and CEO Emil Eifre m told Runtime in an interview last month at the HumanX conference.

Graph databases store data by establishing relationships between different types of data, somewhat similar to the ideas behind vector databases but very different in implementation.
Neo4j — which has raised a little over $800 million and recently reported $200 million in annual recurring revenue — started out as a "system of record database" for one-off corporate applications, Eifrem said, but started to notice a recurring pattern in how customers used the database.
"If you look at any organization, it is a number of things that are deeply connected, a number of networks. It is people and processes, products and information and money that flows across an organization," he said.
And those connections can be represented by graphs that help companies understand how decisions made in one part of the company have a ripple effect across other departments, such as how a supply-chain disruption can impact everything from inventory and pricing to staffing and marketing.

Klarna grew like most startups, building a business around department-specific SaaS apps with "most of them having their own ideas and concepts and creating an unnavigable web of knowledge that required a tremendous amount of Klarna specific expertise to operate and utilize," Siemiatkowski said on X. What Klarna did last year was take that organizational data — spread across thousands of those SaaS apps — and move it into graph databases, Eifrem said.

Klarna and Siemiatkowski wanted to have "a consistent data model," Eifrem said, and they concluded that "the one data model that can unify our sprawling data and systems is the graph model." (Klarna declined to comment for this newsletter, citing quiet-period regulations as it prepares for an IPO.)
Klarna then used LLMs to build an internal application around those graph databases that employees can use to find answers to questions using natural-language commands.
"So no, we did not replace SaaS with an LLM, and storing CRM data in an LLM would have its limitations. But we developed an internal tech stack, using Neo4j and other things, to start bringing data=knowledge together," Siemiatkowski said last month.

But SaaS companies like Salesforce and ServiceNow are convinced customers want to build generative AI apps on their platforms because so much customer data is already there. Storing that data somewhere else — in a graph database or even just a data lake — could make it much easier for customers to build apps that take CRM data once stored in Salesforce, for example, and mix it with data from other SaaS apps to chart their own course.

That could have enormous ramification for SaaS companies that are used to being the "single source of truth" for their customers.
"I hear from a lot of CIOs — Fortune 500, Global 2000 CIOs — and the level of frustration that I've heard about Oracle over the years, it's almost reaching that level of frustration with their massive spend on SaaS services," Eifrem said.
Couple that with the growing excitement around the potential of AI coding assistants and editors to make developing software cheaper and easier, and the ground starts to shift.
"Between that frustration and the increased productivity of in-house engineering teams. I think we're moving into — in this classic build-versus-buy (dilemma) — a build era for the enterprise," Eifrem said.

Meanwhile, on Lanai

Oracle's public silence concerning two recent data breaches entered its second week Monday, all but guaranteeing a steady drip-drop of media coverage as customers attempt to figure out what's going on. It still has yet to acknowledge a breach of its cloud infrastructure systems first reported on March 21st, and on Friday Oracle began sending letters to customers of its Oracle Health service informing them of a separate breach that occurred in February, according to Bleeping Computer.

The FBI is investigating the breach involving the former Cerner division, according to Bloomberg, and it appears that patient data was stolen. Some medical companies have been hit with ransomware demands in response to that breach, but no one seems to understand the full extent of the incident.

And when it comes to the first breach, Oracle appears to be skating on its blanket denial of a breach in "Oracle Cloud" from two weeks ago by the fact that the breach occurred on an older generation of its cloud infrastructure that it now calls "Cloud Classic," according to security researcher Kevin Beaumont. "This is a serious cybersecurity incident which impacts customers, in a platform managed by Oracle," Beaumont wrote, and while the company appears to be communicating with key customers in private, its lack of transparency falls well below industry norms for handling any security incident that results in the loss of customer data.

Enterprise funding

OpenAI raised $40 billion in new funding, valuing the money-hemorrhaging foundation model company that started the generative AI boom at what feels like a high-water mark of $300 billion.

Island scored $250 million in Series E funding to continue developing its enterprise browser, which gives security-conscious companies a way to lock down all corporate web activity.

Crusoe landed $225 million in a new credit facility, adding to a $600 million Series D round the upstart AI cloud provider raised late last year.

Temporal Technologies raised $146 million in Series C funding for its developer platform, which helps companies deploy more reliable applications.

Retym launched with $75 million in Series D funding, giving it a total of $180 million in overall funding as it develops DSP networking chips for data centers.

Straiker launched with $21 million in new funding for its security software designed for generative AI applications as well as agents.

The Runtime roundup

Broadcom is planning to substantially increase the minimum number of processor cores required for a VMware license from 16 to 72, according to CRN, which could spur a lot of medium-sized customers to seek alternatives.

AWS now supports Model Context Protocol in its Bedrock service, adding to the number of large AI and infrastructure companies that are betting on MCP becoming a standard for AI agent connectivity.

Thanks for reading — see you Thursday!

This post was updated to correct the spelling of Neo4j CEO Emil Efriem throughout.