Vertical AI Emerges as Growth Engine at Bank of America Private Tech Conference

The shift is gaining importance as foundation models become more interchangeable. Competitive advantage is moving toward domain-specific data, vertical integration, trust and the capital required to build at scale, according to the report. Companies that control more of the stack, from GPU infrastructure and proprietary models to robotics and manufacturing, are betting that durability will matter more than novelty.

"The thing about AI is that specialized intelligence is a big story now," said John Furrier, executive analyst at theCUBE Research. "It was a sleeper story about a year ago, but domain-specific data, these verticals have the necessary requirements because the GPUs want data, the systems want data, the intelligence comes out of the data, and that is the key." He spoke with industry experts during the conference. The interviews covered humanoid and construction robotics, vertical AI agents in commerce, healthcare and finance, low-power AI chips, electric industrial vehicles and the capital markets funding them.

Bear Robotics Inc. has about 16,000 autonomous mobile robots in the field, a backlog of about 4,000 and revenue that doubles every year, according to co-founder Bren Pierce. Its core market is restaurants in Japan and Korea. It also serves care homes and casinos, and a partnership with LG is taking it into warehouses and factories. Pierce sold his humanoid startup, Kinisi, back to Bear, and the humanoids run on the same software base and cloud infrastructure as the mobile robots. Foundation models that learn from a few hundred examples, onboard compute such as Nvidia Corp.'s Jetson Thor and coding with large language models have cut work that once took six months down to days. Tactile hands remain the missing piece, at about $30,000 per hand.

Labor makes up 55% to 60% of construction costs, and Address Robotics Ltd., which operates as All3, is building an integrated value chain to shrink that share, said co-founder and CEO Rodion Shishkov. Its software designs a building from the plot through permit-ready documents. A production layer then directs industrial robots to make one-of-a-kind elements at mass-production cost, and the company's Mantis mobile robot assembles and finishes them on site. Because All3 designs the process as well as the robot, it can redesign the process to remove most corner cases instead of teaching robots to handle every one. After a seed round of about $25 million to $30 million, its first site is being prepared for a six-story co-living building on an odd, 11-sided plot.

The model layer is turning into a commodity, and value is moving to vertical domains such as e-commerce and marketing, said Raj De Datta, co-founder and CEO of Bloomreach Inc. Bloomreach uses about 100 models. Its Loomi AI engine, trained on 7 billion consumer profiles, produces models that perform five to 10 times better than out-of-the-box LLMs, according to the report. Almost half of its customers already use an AI agent, the number using four agents has grown 23-fold in a year, and revenue is speeding up without any growth in headcount. Through its Loomi Connect offering, third-party agents can call the platform directly, and those calls are growing 83% month over month.

Harbinger Motors Inc. sells electric and hybrid medium-duty vehicle platforms for the same price as diesel, and a typical parcel truck in California saves about $30,000 a year on fuel after charging costs, said co-founder and CEO John Harris. Its customers include FedEx Corp. and Thor Industries Inc., and the battery system it built for delivery trucks now powers Airstream travel trailers. Harbinger builds delivery trucks, RV chassis, energy storage systems and Army autonomous ground vehicles on the same production lines, and it only adds businesses that need no new capital spending.