So you wanna be an AI cloud provider

Welcome to Runtime! Today: CoreWeave's S-1 reveals incredible growth but an unhealthy dependence on Microsoft and debt, Klarna's CEO clarifies exactly what it did with its Salesforce data, and the latest funding rounds in enterprise tech.

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

Oh what a tangled web

CoreWeave is one of the biggest winners of the generative AI frenzy, timing its pivot away from cryptocurrency mining right as OpenAI's ChatGPT sparked a gold rush that depended on GPUs. On Monday it confirmed plans to cash in on that serendipity by going public, which has seemed an almost-quaint strategy during the last several years of enterprise tech.

It's no surprise that the GPU rental business has been pretty good: CoreWeave recorded a 700% jump in revenue during 2024 according to its S-1, ending the year with $1.92 billion. In just the last three months, CoreWeave took in $747.4 million, which is more than three times as much revenue as it recorded in all of 2023.

• The company said it also has commitments from customers to spend $15 billion and "as of December 31, 2024, our committed contracts had a weighted-average contract duration of approximately four years," it said in the filing.
• CoreWeave now operates 32 data centers in the U.S. and Europe, up from just three when co-founder and chief product officer Brian Venturo spoke with Runtime in May 2023.
• The company is losing a significant amount of money thanks to the expense of building out GPU data centers, but its gross margins are 76%, which is outstanding for a hardware business.
• CoreWeave believes it has an efficiency advantage over the Big Three and other AI cloud providers thanks to the software it has built around those GPU servers, which "enables the provisioning of infrastructure, the orchestration of workloads, and the monitoring of our customers’ training and inference environments to enable high availability and minimize downtime."

But one surprise from CoreWeave's S-1 was the degree to which it relied on a literal handful of companies for most of its growth during 2024. Microsoft alone accounted for 62% of its revenue during 2024, and Microsoft and another company (which wasn't named, but seems likely to be Meta given how many times it was mentioned in the filing) accounted for 77% of its revenue.

• Microsoft probably accounts for a lot of that $15 billion in committed spend, but it appears to be moving into a very different stage of its AI buildout entering 2025, distancing itself from OpenAI and acknowledging that the industry likely built too much capacity during the initial frenzy.
• And it will continue to invest in its own GPU infrastructure to serve its own cloud customers, which could reduce its need to rely on CoreWeave for excess capacity.
• When your biggest customer is also one of your biggest competitors, things can get pretty complicated.

CoreWeave also outlined the degree to which it has used debt to finance its GPU buildout, declaring that it "pioneered and scaled innovative financing structures." Those innovative financing structures have, of course, not faced the test of time, and CoreWeave currently has about $8 billion in debt on its balance sheet.

• The company said it expects to make its money back on those server investments over 2.5 years, which is pretty aggressive for the cloud business.
• But that assumes growing demand for its services, which now seems a little shaky after DeepSeek showed what could be done on the cheap and tariffs rocked the stock market Monday and Tuesday.
• It's not hard to see companies that have been struggling to get generative AI apps into production thinking twice about how much they're spending to try and make AI work if their main lines of business start to deteriorate.
• "If demand slips, however, Coreweave could find itself unprofitable and sitting on unused gear that was kicking off GAAP costs like a dog sheds water," as Alex Wilhelm put it in his Cautious Optimism newsletter.

And all that SaaS

Swedish fintech Klarna has been the darling of the generative AI community ever since it released a statement in February 2024 claiming its OpenAI-powered customer-service assistant was doing the work of 700 people, and then doubled down in September by saying it had shut down its Salesforce and Workday accounts in favor of homegrown AI-powered services, which caught Marc Benioff's eye. But on Monday, CEO Sebastian Siemiatkowski said that reports that LLMs are killing SaaS have been greatly exaggerated.

Klarna did in fact end its relationship with Salesforce, but "we did not replace SaaS with an LLM, and storing CRM data in an LLM would have its limitations," Siemiatkowski said in a post on X. Instead, it "decided to start consolidating," moving data that was spread across various SaaS apps — "most of them having their own ideas and concepts and creating an unnavigable web of knowledge," he wrote — into a new internally developed system built around Neo4j's graph database.

That did allow Klarna to retire a bunch of SaaS applications, and he warned established SaaS companies that they face a problem: "They started as companies with a clear opinion of how to do things, but over time, as they try to satisfy every whim of any random person working at any large enterprise, they become somewhat of a glorified database and lose their opinion. Opinionated software is worth something, as opinions represent an experience of what works, what produces results," he wrote.

Enterprise funding

Anthropic raised $3.5 billion in new funding, which values the foundation model company at a whopping $61.5 billion.

Cloudsmith landed $23 million in Series B funding for its software supply-chain management service.

QuantWare scored €20 million ($21.3 million) in Series A funding to continue building processors and chip interconnect technology for quantum computing.

LlamaCloud raised $19 million in Series A funding and announced the general availability of its flagship product, which helps companies build AI agents from unstructured data.

Anagram landed $10 million in Series A funding for its enterprise security training software.

Genesis Computing scored $5 million in seed funding as it develops AI agents that can be used to extract insights from enterprise data.

The Runtime roundup

VMware urged customers to patch three different zero-day vulnerabilities that are currently being exploited by hackers.

AWS created a new group focused on agentic AI led by Swami Sivasubramanian and moved its Bedrock AI services under its computing group, according to Reuters.

Microsoft blamed a widespread Microsoft 365 outage over the weekend on a "problematic code change," according to Bleeping Computer.

Thanks for reading — see you Thursday!