Turba Labs, which aims to optimize AI infrastructure, has raised $52 million
By Sarah Klearman
The data center build-out is exploding across America. But what if AI didn't need more -- more chips, more energy, more resources -- to meet the intensifying demand for computing power?
Enter Palo Alto, Calif.-based Turba Labs, which has raised nearly $52 million in combined seed and Series A financing. The two-year-old startup's goal: to double the world's available compute without building a single new data center.
Turba Labs creates a digital twin of the infrastructure within data centers. Its platform models customers' artificial intelligence use cases and then directs those demands toward the hardware best suited for their level of complexity.
The company's thesis is that optimizing existing AI infrastructure could unlock significantly more compute than is currently available, said co-founder and co-Chief Executive Hans-Juergen Schmidtke. By customizing how AI workloads interact with the underlying hardware, Turba Labs' platform could double the availability of compute. Schmidtke said exactly how much capacity the platform unlocks will vary depending on the infrastructure involved.
Turba Labs' platform could help make more supply available to meet mushrooming demand for the compute necessary to run AI models. Demand already outstrips supply, fueling a data center build-out that is on track to become the largest economic bet in U.S. history, The Wall Street Journal has reported. And scarcity of the power, hardware and skilled labor required to build data centers -- as well as the willingness of local jurisdictions to approve them -- is further complicating progress, according to a report published by Bain & Co. in September.
There are also shortages in various components of AI infrastructure, said Schmidtke, describing a "round robin of scarcity."
Schmidtke should know: As the former director of engineering at Meta Platforms, he oversaw large-scale deployment of AI infrastructure for the company.
"There are few people in the world that have more insights into procurement, write-off cycles, maintenance -- the questions that anyone who buys chips for data centers [asks] -- than him," said Simon Schmincke, a Berlin-based partner at VC firm Creandum, which led Turba Labs' Series A.
Co-founder and co-CEO Patrick Jahnke most recently led a data center team at SAP, optimizing its AI usage, among other things. SAP is one of Turba Labs' customers, a company spokesperson said.
The co-founders' understanding of the hardware and software served as the foundation for the "infracode," as Schmidtke described it, on which Turba Labs' platform is built. The company's long-term goal is to usher in the commoditization of compute.
"Compute is going to be the new oil. Not a single business in the world won't run with AI-enabled intelligence in the next three to five years," Creandum's Schmincke said. "It is impossible that we solve the problem with more chips. Energy, water, permits -- all of this is constrained. So we need to start thinking about: How can we get more out of the existing?"
Turba Labs works not only with compute providers like data center operators and neoclouds, but enterprises wanting to project and optimize their compute usage, as well as hardware makers who want to project how their products will perform.
The company has 65 employees across its headquarters in Palo Alto and an office in Heidelberg, Germany, where Jahnke is based. It will use the Series A funding to expand its engineering and product teams, among other things, a spokesperson said.