How to monetize AI in the battery sector

Stanford spinout Aionics thinks it has figured out how to make money with artificial intelligence and machine learning in the battery industry.

The seed-stage startup, which was founded in 2020, was built on the thesis that there’s only one step in the battery material development process that has the potential to get exponentially better: the compute step. 

According to CEO and cofounder Austin Sendek, that’s the moment where the cost savings can add up over time. “The cost to build the factory is going to be the same; the cost to run experiments is going to be the same,” he said. “But the cost to run a simulation or make a computational prediction is becoming better. It's also becoming cheaper, and doing so rapidly.”

With that assumption in mind, Aionics developed models to predict and understand molecular properties, applying deep learning techniques originally developed for tasks like image recognition. The company trains those models on both the large public datasets used in chemistry and biology, as well as on internal Aionics data.

The goal? As co-founder Venkat Viswanathan explained, Aionics is aiming to use these models to identify specific molecules that can have a disproportionate impact on the performance of a mixture, such as for a battery electrolyte. Today the company develops those mixtures itself, but that’s not for lack of trying other, ultimately less-profitable, avenues to monetizing AI for battery material development.

The abandoned software path

In 2020, Aionics started out as a software developer, licensing its models to battery companies that wanted to identify and design new, customized electrolytes. Among its early customers was Cuberg, the Bay Area lithium-metal battery company that was later acquired by Northvolt.

But Aionics quickly came up against a major problem with that business model: it’s hard to quantify the value of accelerating research.

“In essence, it’s uncertain,” Sendek said. “There’s a throughline to value, but it’s often a long one thorough years of iteration.” That was a tough early learning for the company, Sendek added, and likely one that other startups will have to learn as well.

“Just because you can perform a task better doesn’t necessarily mean that you’re going to make money doing it,” he said. “We’re seeing a glut of AI for materials companies selling a tool or a platform, and not seeing a lot of uptake in the industry of those sorts of tools.” That’s made especially challenging by the fact that major companies like Microsoft and Google are already using AI for similar material exploration purposes. 

Even within Aionics’ existing customer pipeline, the finances didn’t look great in early 2022, two years post-launch.

“We were having these conversations with customers where they were saying ‘if you can actually deliver this insight to us, we’re going to raise a $100 million round and we’ll pay you $100,000 for your platform,’” Sendek said. “But that doesn’t really make sense for us.” So, the company pivoted.

Making AI profitable

In late 2022, Aionics decided to rewire its business model, moving from the traditional Silicon Valley software approach to something more common in the biotech industry. The company decided to use its own models to develop electrolyte formulations, both for battery manufacturers, and as part of its own ongoing research.

The models first predict the properties of individual molecules, such as boiling points or stability, explained Viswanathan. The second, key piece, he said, is predicting how molecules behave when mixed together; with that prediction, the company can pinpoint, predicting the properties of a new electrolyte mixture.

Aionics is developing those models to identify specific molecules that can have a disproportionate impact on the performance of a mixture, even when added in small quantities, he added.

The company’s bet, said Sendek, is that its value to a customer is much clearer if Aionics can deliver an electrolyte that drops into a battery, as opposed to delivering the tools to help a battery maker identify that electrolyte. The model mirrors that of the partnership between Pfizer and BioNTech to co-develop a COVID-19 vaccine in 2020.

“We often make that comparison in the sense that BioNTech developed the vaccine but didn’t manufacture it,” he said. “Pfizer was the manufacturing partner, but in doing the discovery and the prototyping BioNTech was able to bring this new solution to market, and it was a much more successful outcome for them than if they had an mRNA software platform that they sold to Pfizer.”

Broadly, Sendek said, Aionics’ shift from a software play to electrolyte development has allowed the company to grow more rapidly in the two years since. “The idea of vertically integrating and selling products rather than software tools has already attracted much more commercial interest,” he said. 

And the model presents Aionics with “a very clear pathway to scalable revenue,” said Viswanathan. “Whereas if you sell a platform, you’re making maybe $50,000 per customer, so in order to build a multi-million dollar company you need hundreds of customers.” 

Aionics, in contrast, is generating revenue from up-front payments on partnership deals but also hopes to generate ongoing revenue as batteries are developed using its formulas. That’s a milestone they expect to hit within the next year or so. The company’s first publicly announced customer for this model is Porsche battery manufacturing subsidiary Cellforce Group.