[INTERVIEW] SES AI eyes robotics boom with ‘vibe research’ tool for battery materials

The battery industry is bracing for a paradigm shift as the rise of physical artificial intelligence (AI) boosts demand beyond traditional applications such as electric vehicles (EVs) and energy storage systems (ESS) to include drones, humanoid robots and other robotic systems.

The transition forces battery makers to invest more heavily in developing new materials and capabilities for delivering the power output, safety and durability required for next-generation robotics.

SES AI, a Boston-based battery firm, believes “vibe research” will be at the center of this new paradigm. The AI-driven approach allows researchers to guide and orchestrate materials discovery through simple prompts, dramatically shortening battery development cycles “from years to weeks” while reducing research costs through the company’s Molecular Universe platform.

During an interview with The Korea Times, SES AI founder and CEO Qichao Hu emphasized that the race to produce humanoid robot batteries will be about how quickly companies can discover and commercialize new materials, as humanoid robots are expected to require batteries with greater capacity and longer operating hours.

“Humanoid robots mostly use 21700-type cylindrical cells (21 millimeters in diameter and 70mm in length)… The cylindrical cells have gone from 5 ampere-hour (Ah) cells to 7.2 Ah cells, so that robots can do more work and operate longer,” Hu said.

“What we do is provide the Molecular Universe platform (to battery makers) so they can quickly screen and come up with new materials that can achieve greater capacity in the same form factor.”

Founded in 2012, SES AI is a New York Stock Exchange-listed battery manufacturer and software developer. It operates three core business pillars: energy storage solutions for data centers, battery products for drones and materials discovery.

It has three bases: its headquarters in Boston and two manufacturing sites in Shanghai and Chungju, North Chungcheong Province.

The Chungju plant was originally established in 2021 as a joint venture with GM Defense to produce batteries for military electrification programs for the Joe Biden administration. After GM exited the project amid a strategic shift away from EV and military battery initiatives, SES AI took full ownership of the facility and converted it into a production base for drone batteries.

The plant plans to scale up its manufacturing capacity from 330,000 drone cells per year to 1 million.

Initially, the company gained attention as a startup for lithium metal batteries, a technology widely regarded as a successor to conventional lithium-ion batteries due to its higher energy density, drawing investments from Korean conglomerates including Hyundai Motor Group, SK Group and LG Group.

However, as the EV market — the largest source of demand for batteries — increasingly became dominated by a number of large manufacturers competing on production scale, cost and manufacturing efficiency, SES AI chose to step away from direct competition with the industry’s largest players. Instead, it shifted its focus toward drones, robotics and AI-powered materials discovery.

Battery companies globally spend 3 percent to 5 percent of their annual revenue on research and development, employing thousands of researchers and scientists. As the cycle of battery materials innovation accelerates, the cost of developing next-generation batteries is also rising.

Hu said SES AI aims to bring innovation to this process through its Molecular Universe platform. Traditionally, scientists have relied on a human-led trial and error processes to identify and test battery-relevant molecules. Through Molecular Universe, however, SES AI seeks to enable an AI agent to autonomously conduct much of the experimentation process based on prompts written in natural language by human researchers.

“It’s called ‘vibe research,’ and that’s when AI takes over,” he said. “AI actually does the research for you. … It’s still a human giving the prompt, but the platform itself will come up with ideas.”

A battery maker seeking to extend battery life at minus 20 degrees Celsius from 6,000 full charge and discharge cycles to 8,000 cycles would traditionally need several years of research, hundreds of scientists and extensive experimentation with different materials and manufacturing processes.

Through Molecular Universe, however, AI agents can autonomously conduct simulations and experiments through connected autonomous laboratories, continuously iterating until they identify a solution, according to Hu.

“With this, the research trend can end up with a single-person product team,” he said. “Instead of having a big team, you just have one product manager and this platform. … So this is a huge cost savings and time acceleration for product development.”

Hu said similar AI-driven “vibe research” platforms have already become profitable in the pharmaceutical industry over the past several years, noting that the battery industry is likely to follow a similar path. He added that some of the largest battery companies and car companies are already using Molecular Universe.

Since SES AI does not manufacture battery cells for the EV and ESS markets, it can support larger battery makers with strengths in mass production and manufacturing scale through its materials discovery platform. In return, the company can benefit from the success of those markets without directly competing in them.

Molecular Universe is currently available in version 3.0, whose most significant update is the introduction of an AI agent called StarSeeker. Unlike previous versions, which required researchers to manually execute individual tools, StarSeeker can orchestrate the platform’s various capabilities based on a simple natural language prompt.

For example, a user can enter a prompt such as, “Find the top 20 molecules for low-temperature sodium battery performance, generate 10 formulations for each, and rank them by performance.” StarSeeker then carries out the required tasks across the platform and returns the results.

The platform could become increasingly valuable as the rise of humanoid robots creates demand for new battery chemistries and materials capable of delivering lighter weight, higher power output and longer operating times.

The robotics sector is still in an early stage, leaving significant room for experimentation and optimization. This, according to Hu, makes rapid materials discovery a critical competitive advantage.

“(Tesla CEO) Elon Musk mentioned that there could be more humanoid robots built than the number of cars and also even the number of people, so that means the potential for this market is huge,” he said.

“So far, there’s no government subsidy or compliance requirement for humanoid robots, though it seems they will use similar batteries as EV. This basically means that whoever can make the best quality (product at) the lowest price can supply.”

SES AI posted $21 million in revenue last year, up roughly tenfold from a year earlier. The growth was driven primarily by its ESS business. Its operating loss narrowed to $82.61 million from $109.24 million over the same period.

Though SES AI has high expectations for Molecular Universe, Hu said it remains a cost center and requires long-term investment to realize its potential. To enable its sustainability, he said, he expects SES AI Korea, which operates the Chungju plant, to become a revenue generator through its drone battery business.

Hu added that a potential spin-out of Molecular Universe into a separate entity is under consideration, which could attract dedicated investment and potentially pursue an independent initial public offering.