New investors include Galvanize, Brevan Howard Macro Venture Fund, and ArcTern Ventures, alongside existing backers Celesta Capital, Drive Capital, Eric Schmidt’s First Spark Ventures, and Micron Ventures.

CEO Faris Sbahi told Fortune the company’s software platform is already being used by more than half of the top 10 semiconductor companies by revenue, as it targets one of the industry’s biggest challenges: the rising cost and complexity of designing advanced AI chips, where even small errors can lead to expensive delays and rework.

Designing advanced AI chips has become so complex that even getting a design to “tape-out”—the point where it’s finalized for manufacturing—is increasingly prone to costly failure. Modern AI chips, which pack in tens of billions of transistors to support today’s frontier models, can cost more than $500 million to develop before a single unit ships.

Normal, founded in 2022 by former engineers and scientists from Google Brain, Google X, and Palantir, is also using its chip design software internally to build its own experimental AI hardware. It has already taped out a prototype chip using the company’s “thermodynamic” approach, which uses the inherent randomness of physical systems to compute more efficiently than traditional GPUs. It’s an early step in a longer-term effort to significantly reduce the energy demands of AI.

“The mission of the company is to go after this so-called AI energy crisis,” said Sbahi. “Data centers are expected to hit an energy wall around 2030, and most of the strategy now is to find new ways to acquire more energy—but our position is to solve the problem in terms of the hardware that we’re using.”

Seeking alternatives to existing AI hardware

Normal Computing is part of a growing group of startups exploring alternatives to conventional AI hardware, including Unconventional AI, led by former Intel AI chief Naveen Rao, which raised a $475 million seed round in January led by Andreessen Horowitz and Lightspeed Ventures. Another is Extropic, which is developing probabilistic AI chips based on a different technical approach.

Sbahi said the company chose the name “Normal Computing” to reflect its view that its approach is closer to how computation should naturally work. “We think this is the more normal way of computing,” he said, pointing to how the company’s software and hardware are designed to align with the underlying physics. “The software really matches the hardware.”

While building energy-efficient AI chips is the company’s long-term goal—initially focused on inference workloads for generative AI—the current fundraise will focus on scaling Normal’s commercial software business.

“Hopefully someday we’ll be integrated into mainstream semiconductor design manufacturing,” said Sbahi. He added that the semiconductor industry’s high costs and complexity make it difficult for new approaches to break in, which is why Normal has focused on working with existing chipmakers rather than trying to disrupt the system from the outside.

“It’s very expensive to make mistakes,” he said.