The extensive energy demands of artificial intelligence are causing a bottleneck in data centers, and Hammerhead AI Inc. wants to do something about it after raising $10 million in seed funding today.

The oversubscribed round was led by Buoyant Ventures and saw participation from SE Ventures, which is backed by the French energy management giant Schneider Electric SE, as well as AINA Climate AI Ventures, MCJ Collective, Wovenearth Ventures, Bombellii Ventures, Clearvision Ventures, Stepchange and Acclimate Ventures.

The AI energy crisis is no joke. Big data center operators such as Microsoft Corp., Amazon Web Services Inc., Google LLC and Oracle Corp. are all desperate for energy as they scramble to build out their infrastructure to support the AI revolution. With data centers springing up across the country, there simply isn’t enough electricity to power them all, as grid queues in some locations can stretch for years.

The result is that many so-called AI factories, which are packed full of powerful graphics processing units, typically run at only 30% to 50% utilization, the startup says. That means thousands of GPUs, amounting to megawatts of computational power, are left sitting on the sidelines.

Hammerhead’s solution to AI’s energy conundrum is to use AI agents to optimize the token processing capacity and power utilization of individual GPUs. By doing this, the startup reckons, it’s able to boost output by as much as 30%. It says its Orchestrated RL Control Agents dynamically orchestrate AI workloads within existing energy budgets, which enables data center operators to deploy much faster than otherwise and utilize much more of their available computational capacity.

Whereas other solutions are focused on increasing power utilization efficiency, or PUE, the ORCA agents target system-wide token productivity. They optimize every aspect of the data center – so not just the GPUs, racks and servers, but also related equipment such as the chillers and cooling distribution units, and sometimes even the actual AI workload. By doing this, they can increase the token throughput of each GPU, which means they can speed up processing as well as deployment times.

“With ORCA, we’re enabling AI factory operators to achieve greater output from existing resources,” said Hammerhead co-founder and Chief Executive Rahul Kar. “Our investors recognize the potential for AI factories to transform their business models and move beyond the limits of grid infrastructure.”

Kar isn’t joking when he promises to “transform” the business models of data center operators, for the ORCA agents can deliver some serious efficiency gains, potentially freeing up megawatts of idle computing capacity. And in constrained markets, that additional processing power can generate revenue in the region of $20 million to $50 million per year, he said.

He should know. Kar has spent years working on energy optimization, having previously founded and scaled up a startup called AutoGrid, which pioneered novel power orchestration technologies and was later acquired by Schneider Electric. Alongside co-founder and Chief Technology Officer Rajeev Singh, he has helped to orchestrate more than 8,000 megawatts of energy assets across more than 20 countries.

Holger Mueller of Constellation Research told SiliconANGLE that Hammerhead AI provides yet more evidence that AI is the solution to many of AI’s problems — in this case, its almost unquenchable energy requirements. “It doesn’t provide more energy, but it does allow AI to run much more efficiently than before, orchestrating workloads in coordination with user’s cost budgets,” the analyst said. “Hammerhead will go up against single AI infrastructure players, and it’s task now is to prove how valuable it can be to enterprises running AI factories.”

Hammerhead has a bit of a head start in this thanks to its close relationship with Schneider Electric, having graduated from SE Venture’s first accelerator program. Its ORCA agents have already been deployed in some of the energy giant’s own facilities.

SE Ventures Global Head and Managing Partner Amit Chaturvedy said Hammerhead’s ORCA agents immediately stood out for its focus on optimizing infrastructure to enhance token processing for AI. “There is an opportunity to launch reference designs for future AI factories,” he said. “It’s the only company positioned to boost data center operators’ revenue through its unique token maximization approach.”

Hammerhead said it will use today’s funding to accelerate the development of its Orca agents and deepen its ties with AI factory operators as it bids to reshape data center economics.

Image: SiliconANGLE/Meta AI