Last week, the market witnessed Tesla’s new flying car model, while AI stocks sold off amid concerns over CoreWeave’s (CRWV) debt and massive spending on AI infrastructure.
AI stocks have since rebounded, but the sharp swings show how sensitive investors remain to news about AI spending and financing.
For the past few years, the AI boom has been a race to build powerful language models. Companies have spent billions of dollars on data centers, advanced chips, and computing infrastructure, betting that demand for AI will continue to grow.
But the next phase of AI may be less about building bigger models and more about how people interact with the technology in the physical world.
That’s where Ramin Hasani, co-founder and CEO of Liquid AI, sees potential.
In an exclusive interview with TheStreet at Qualcomm’s Snapdragon Summit in September, he discussed where he sees the next opportunities in AI; how AI could move beyond chatbots and data centers into cars, smartphones, and other everyday devices; and whether soaring AI valuations have gone too far.
AI isn’t just about chatbot models
Liquid AI was founded by a team from the Massachusetts Institute of Technology, and it develops AI models that run across various hardware types.
Unlike companies building consumer-facing chatbots, Liquid AI focuses on supplying technology to businesses that want to integrate AI into their own products.
Hasani said there is an important distinction between the consumer AI market, where companies compete to build apps and agents, and the enterprise market, where manufacturers need the underlying technology to make those products work.
Also read: CoreWeave stock plunges 12%, but Wall Street analyst says buy the dip
“If you just look at the model layer, you will see no differentiation,” Hasani said, referring to the growing number of AI models available from companies in the U.S. and China.
He argued that the real differentiation comes from combining models with the software systems needed to enable more sophisticated AI behavior.
For example, a carmaker that develops an intelligent assistant needs more than a language model. The system must understand users’ voice commands, process information from cameras, and access vehicle functions while responding quickly.
Liquid AI provides models and supporting software, developing a platform that enables businesses to customize and improve their AI systems. The idea originated from Hasani’s study of how neurons exchange information in small animal brains.
“Liquid means flexibility,” Hasani said, explaining that the goal is to build AI models that are adaptable after training.
Why the next AI battleground could be your car
Liquid AI is working with Mercedes-Benz on an AI-powered in-car assistant that Hasani expects to roll out in North America by the end of 2026.
Hasani said his company provides an AI system that runs inside a Mercedes vehicle on Qualcomm (QCOM) chips, allowing drivers to communicate with their cars more easily. Its software can access vehicle functions that traditional smartphone assistants cannot.
Hasani used Apple’s (AAPL) Siri as an example.
“What if you ask Siri to open my window?” he asked.
Related: Qualcomm CFO talks AI, data centers, and next big bet on robots
Siri generally cannot control a vehicle’s windows without the necessary integration yet.
“When we work directly with OEMs (Original Equipment Manufacturer), we also get all the access inside the car,” Hasani said. That access could enable thousands of vehicle functions through human language commands.
The auto industry is moving toward vehicles that can gain new features through software updates. That shift could create a growing market for chips and AI software designed to run inside cars.
Liquid AI is also partnering with Qualcomm (QCOM) to bring its AI models to Snapdragon-powered devices, allowing more AI tasks to run directly on the hardware rather than relying on cloud-based data centers.
Businesses might cut costs, not AI usage
Consumers may tolerate waiting several seconds for a chatbot to answer a complicated question. Businesses processing millions of requests often cannot.
Hasani pointed to Liquid AI’s work with Shopify (SHOP) as an example. Requests made through Shopify’s Shop app can be processed using a Liquid foundation model. The company needs to handle large volumes of requests while maintaining extremely low latency.
“You want to conclude a job that is coming in 20 milliseconds with an AI model…. There aren’t that many options in the world that can do that,” Hasani said.
Liquid AI’s approach involves optimizing models for the hardware on which they run.
For many businesses, a smaller model that responds quickly and costs less to operate may be more attractive than a larger model with stronger benchmark results. As AI becomes more deployed, efficiency could become as important as raw computing power.
Related: Micron stock tumbles despite analyst calls semiconductor selloff overblown
Hasani expects businesses to become more selective about AI spending as they look for better returns from the technology.
“There’s a gap between the capabilities of a model and the value that they’re creating,” he said.
He believes companies will find ways to reduce AI costs without necessarily using the technology less.
“Enterprises are getting smarter about the expenditure on AI, so they’re definitely gonna cut that cost, but that doesn’t mean that they’re gonna cut usage,” Hasani said.
How Liquid AI makes money
Liquid AI makes money primarily by licensing its models and software to manufacturers, generally charging an annual per-device fee in addition to development costs. It also charges enterprise customers based on token usage when customizing models.
Hasani has an ambitious target: “We want to be on 50% of all the devices on the planet.”
He said potential annual licensing revenue could range from less than $1 to hundreds of dollars per device, depending on the application, though the company has not disclosed revenue forecasts.
Liquid AI remains privately held, and Hasani said a potential IPO is on his mind, although the company is still too young to pursue one immediately. The company faces challenges including competition, hardware limitations, and the difficulty of securing large-scale commercial deployments.
There are senseless valuations in AI
As investors debate whether the AI boom resembles the dot-com bubble, Hasani sees a distinction between the technology’s potential and the prices investors pay for it.
“There are senseless kinds of valuations that are coming on,” he said. “I can totally see from a traditional economic sense, we would be in a bubble, like that’s how you would define a bubble. It exactly looks like that.”
But from a technological perspective, Hasani believes AI’s biggest breakthroughs are still ahead.
“The promise of this technology is much, much bigger than what it was before,” Hasani said.
One area he is particularly excited about is healthcare. Hasani pointed to recursive self-improvement, a process in which AI systems help develop more advanced versions of themselves.
He believes that could lead to breakthroughs in cancer research, longevity, and other areas of medicine.
“If you say that slow down the pace of innovation across the finding out how to cure cancer, I’m not gonna slow down because people are losing time,” Hasani said.
Beyond healthcare, Hasani sees opportunities for AI to help scientists answer fundamental questions about physics and the universe.
Still, Hasani pointed to some risks of developing “increasingly powerful” AI technology.
He said he has discussed AI safety with other foundation-model executives and found that researchers across the U.S. and China recognize the need for safeguards, even as competition between companies and countries intensifies.
He warned that some people could misuse AI and stressed the importance of keeping AI under human control. One priority, he said, is making AI outputs verifiable so that humans can understand and check the results.
“AI has to stay under control of humans… as long as we can make sure that’s the case, we should be fine,” he said.
Despite those concerns, Hasani remains optimistic about AI’s long-term potential.
“I am definitely worried,” Hasani said. “But you have to be cautiously optimistic.”
Related: Cathie Wood buys $11.8 million of tumbling tech stock