|
Welcome back! Cathy here. Earlier this week, I wrote about startups moving some of their workloads from Amazon Web Services and other traditional cloud providers to young challengers such as Nebius. That’s in part because startups are having trouble getting the Nvidia chips they need on AWS, sometimes because those chips are only available in too expensive packages. Capacity can be so scarce that an Nvidia representative told one small AI startup it should rent spare GPUs from a facility associated with the Qatari government, an executive at the startup told us. And even neoclouds like Nebius don’t have the capacity to serve all potential customers. For instance, as we reported in our piece, The Biological Computing Company, an AI startup that signed up to use Nebius servers for cost reasons six months ago went to AWS in the past month or two, according to Alex Ksendzovsky, the startup’s CEO. He said Nebius didn’t have the capacity the startup needed, and AWS also dropped its GPU rental price. Dan Lawrence, general manager of the Americas at Nebius, said that for every Nvidia AI server chip in its facilities, there are four to five companies that would gladly use it. That’s forcing Nebius to pick and choose which ones get the goods. “We have capacity meetings three times a week where we talk about the customers that are coming and which ones we want to pick,” Lawrence said. The supply-demand imbalance likely influenced Nebius when it decided to hike prices earlier this year, and “we are still selling out across all chip types at the higher prices,” Chief Revenue Office Marc Boroditsky told analysts in May. (Server and chip component shortages are also pushing up prices that neoclouds and other firms pay for Nvidia server racks, as we explained in this article.) Nebius’ willingness to cater to numerous small customers stands in contrast to other neoclouds that overwhelmingly prefer to rent out large clusters of graphics processing units to a handful of mega customers. But Nebius also has struck some big deals with Microsoft and Meta. Lawrence says around 75% of Nebius’ business comes from companies that have already maxed out the servers they can rent from traditional clouds, while the other quarter are moving all their cloud workloads to Nebius. Lawrence said Nebius’s startup customers are generally AI model makers, robotics companies or sell coding generation tools, such as Cursor and Cognition. He said while Nebius does give out credits to help startups get started, its strategy to attract new business isn’t centered around freebies. “Rather than offer free compute, we find well-funded startups that can scale and grow,” he said, and later added: “We don't consider ourselves a low-cost provider.” Vibe-Coding in the Cloud Beyond GPUs, AWS is also facing new competition from an older class of cloud software startups that help businesses develop websites or apps, including Render and Vercel, which are both eight or more years old. These firms can make it easier for nontechnical customers to work with AI tools, compared to traditional cloud providers, some companies and people who work with them say. “AWS has always been stronger at providing the building blocks,” said Randall Hunt, CTO at Caylent, which helps customers use AWS. “They give you the Legos and you get to build whatever your mind can imagine, whereas Vercel is much more of the ‘Hey, here's the exact template to follow and this is how you should build.’” That’s why Trusted Health, a nurse staffing company with around 200 employees, is shifting around 75% of its AWS spending to Render before the end of the year, said Marcelo Silveira, a vice president of architecture. Silveira said the switch will help make it easier for nontechnical staff to use AI tools, without relying on the help of Trusted Health’s engineering staff. Those engineers can now spend more of their time on AI-related projects, rather than working on cloud-related tasks, he said. For its part, AWS says it offers a bevy of tools to help startups navigate its tech, including AWS Startup Advisor, an AI assistant, and offers its own AI that can help nontechnical customers and coders alike develop apps and sites.—Phoebe Liu contributed to this article.
| 
