Vertical AI and Purpose-Built Hardware Drive Next Wave of Technology Growth

The emerging frontier of artificial intelligence is not in general-purpose models but in technology tailored to specific industries, proprietary datasets and increasingly, hardware engineered for particular tasks. Companies showcased at the recent Bank of America Private Tech Trailblazers Conference in Palo Alto, California are targeting concrete problems in restaurants, construction sites, hospitals, cross-border finance and defense, where AI can deliver quantifiable value.

As foundation models grow more commoditized, the sources of differentiation have shifted. Competitive moats now depend on domain-specific data, vertical integration, customer trust and the capital required to scale manufacturing and infrastructure. Organizations building more of their own technology stack—from GPU systems and custom models to robotics and production facilities—are betting that durability will outweigh novelty.

The thing about AI is that specialized intelligence is a big story now. 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.

John Furrier, executive analyst at theCUBE Research

Nine Insights from the Conference

1. Bear Robotics Expands Humanoid Deployment Across Hospitality and Beyond

Bear Robotics Inc. operates approximately 16,000 autonomous mobile robots in the field with a backlog of roughly 4,000 units and annual revenue that doubles each year, according to Bren Pierce, the company's co-founder. Its primary customer base is restaurants in Japan and Korea, though it also serves care homes and casinos. An agreement with LG is opening opportunities in warehouses and factories. Pierce previously founded Kinisi, a humanoid robotics startup that Bear acquired; the humanoid systems now run on the same software platform and cloud infrastructure as the mobile robot fleet. Three technological advances have compressed development cycles from six months to days: foundation models trained on hundreds of examples, onboard processors like Nvidia Corp.'s Jetson Thor, and large language model-assisted coding. Tactile hands remain a bottleneck, with costs around $30,000 per unit.

2. Construction Automation Requires Semi-Structured Job Sites

Labor accounts for 55% to 60% of construction expenses. Address Robotics Ltd., operating under the name All3, is assembling an end-to-end value chain to reduce that proportion, explained Rodion Shishkov, the company's co-founder and chief executive officer. The company's software generates building designs from site surveys through permit-ready documentation. A production layer then directs industrial robots to fabricate custom elements at mass-production economics, while All3's Mantis mobile robot handles on-site assembly and finishing. By designing the entire process alongside the robots, All3 can restructure workflows to eliminate most edge cases rather than programming robots to handle every scenario. Following a seed funding round of approximately $25 million to $30 million, the company is preparing its first project: a six-story co-living structure on an irregularly shaped 11-sided plot.

3. Proprietary Models on Proprietary Data Outperform Generic LLMs

The model layer itself is becoming a commodity, with value migrating toward vertical applications in sectors like e-commerce and marketing, according to Raj De Datta, co-founder and CEO of Bloomreach Inc. The company operates roughly 100 models. Its Loomi AI engine, trained on 7 billion consumer profiles, generates models performing five to 10 times better than standard large language models. Nearly half of Bloomreach's customer base already deploys an AI agent; the cohort using four agents has grown 23-fold in a year. Revenue is accelerating without corresponding increases in staff. Through Loomi Connect, third-party agents can integrate directly with the platform, and these integrations are expanding at 83% month over month.

4. Electric Medium-Duty Vehicles Match Diesel Pricing While Cutting Fuel Costs

Harbinger Motors Inc. manufactures electric and hybrid medium-duty vehicle platforms priced competitively with diesel alternatives. A typical delivery truck operating in California saves approximately $30,000 annually on fuel after accounting for charging expenses, stated John Harris, Harbinger's co-founder and CEO. The company counts FedEx Corp. and Thor Industries Inc. among its customers. The battery technology Harbinger developed for delivery vehicles now powers Airstream travel trailers. The company manufactures delivery trucks, RV chassis, energy storage systems and Army autonomous ground vehicles on shared production lines, adding only businesses that require no additional capital investment. A collaboration with American Rheinmetall targets ground autonomy for defense applications, and Harbinger roughly doubles its production capacity annually.

5. Physics-Based AI Chip Reduces Power Consumption by 1,000x

Unconventional Inc., which operates as Unconventional AI, is co-designing hardware and software to slash AI power consumption by roughly 1,000 times, according to Ali Esfahani, the company's chief financial officer. Naveen Rao established the company a year ago. It has raised approximately $540 million, grown from 14 employees in January to about 60, and taped out a chip at Taiwan Semiconductor Manufacturing Co. on June 1. The chip's power characteristics came in orders of magnitude lower than conventional systems. The approach leverages physics-based dynamics on standard semiconductor processes, where the system's changing state serves as memory, analogous to neural pathways in biological brains. Esfahani argued that because the electrical grid can only expand linearly, today's architectures cannot sustain exponential growth.

6. Airwallex Achieves 60% Annual Growth Through Localized Global Banking

Airwallex Pty Ltd. operates a global financial platform enabling businesses to establish accounts, receive payments and distribute cards across the world's 80 to 100 largest economies as though operating locally, explained Irvin Sha, head of corporate development, capital markets and investor relations at Airwallex. The company was established in Melbourne in 2015 and invested years accumulating more than 90 licenses, banking relationships and card network partnerships. The platform now incorporates AI, allowing customers' agents to conduct transactions on their behalf. In agentic commerce, which Sha characterized as a land grab, interoperability and accuracy protecting all transaction participants will prove critical. Airwallex has raised approximately $960 million across Series F, G and H funding rounds at a valuation reaching $11 billion and maintains approximately $1.4 billion in annualized revenue.

7. Voice Agents Extend Healthcare Capacity Without Diagnosis or Prescription

Hippocratic AI Inc. develops voice agents managing appointment scheduling, pre-operative preparation, post-discharge follow-up and ongoing chronic disease management for health systems, insurance payers and life sciences organizations, all without performing diagnosis or prescribing treatment, according to Shubhra Jain, chief business officer of Hippocratic AI. The system prioritizes safety through 31 models: one conducts the conversation while 30 provide oversight. Six health systems that invested in the company contributed 6 million actual patient calls to refine the system. Since agents are not evaluated on call duration, they can develop relationships using memory that persists between interactions. Over three years, the company has secured more than 60 enterprise contracts, including five of the nation's largest insurance payers.

8. Capital Requirements Drive AI Companies Toward Public Markets at Larger Scale

The most significant market shift concerns the magnitude of capital that AI and robotics ventures require, pushing companies to reach public markets at substantially greater scale, noted JD Moriarty, vice chairman and managing director and global head of TMT equity capital markets at Bank of America Corp. The bank monitors what it terms AI tailwind companies. Several that faced difficulties two years ago are now considering public offerings. Public market investors prioritize durable, outsized growth above all else, and currently scale is outweighing differentiation, as CoreWeave Inc. demonstrated. In 2026, capital markets activity is tilting toward hardware and semiconductors rather than software, and the bank is developing integrated capital solutions extending beyond initial public offerings.

9. CloudWalk Achieves 99% Automation of Customer Support Using Proprietary GPU Infrastructure

CloudWalk Inc. serves more than 10 million active users through InfinitePay in Brazil, its new Pierre agent and JIM.com in the United States, and has surpassed $2 billion in revenue with hundreds of millions in profit, noted Luis Silva, CEO and founder of CloudWalk. Half of those users interact with its agents daily, and the agents operate on the company's own cluster of hundreds of Nvidia Blackwell GPUs. Because CloudWalk operates as a regulated financial institution, it embedded guardrails into its agent architecture to ensure compliance with owner instructions and minimize hallucinations during financial transactions. Agents now handle 99% of customer support, up from 65% 18 months prior, and the company generates $2.7 million in revenue per employee.

https://www.youtube.com/embed/videoseries?list=PLZvOz3_Xy7sM