Lu Zhang: Finding Strategic ROI in Physical AI and Vertical Healthcare

Shift to physical data

Zhang said she has observed a fundamental shift in the AI landscape. “The narrative is changing, shifting from language model to world model to physical AI, and from the chat to agentic,” she remarked.

While standardized digital language data may be plentiful, the specialized information required for physical AI, encompassing robotics, sensing and real-world interactions, remains scarce. “We have an infrastructure of around 50 percent and compute 50 percent. The data is not even 10 percent.”

This deficit represents what she considers one of the premier innovation prospects in the current AI sector: the rise of firms focused on tactile sensors, novel data platforms and uniform pipelines designed for 3D real-world data.

Major corporations are beginning to appreciate the importance of this data. Zhang noted that among the 45 CTOs in her professional circle, which spans the logistics, semiconductor and manufacturing industries, there is growing demand for startups that can assist in harvesting and organizing their industrial data assets.

On the energy side, Zhang argued that the biggest burden in AI systems isn’t necessarily computation itself, but moving data around. “That part (of) energy consumption actually (is) 100 times more than compute itself,” she said.

This becomes particularly acute for three-dimensional physical data, which is far larger than text. In Zhang’s opinion, as physical AI and robotics scale, edge computing will become more essential. Without that shift, she doesn’t believe widespread deployment will be sustainable.

Where's the ROI?

The same practical lens applies when Zhang looks at AI in healthcare.

While most of the public discussion has focused on AI for drug discovery, she said the real budget — and therefore the real return on investment — sits in optimizing clinical trial results.

Her firm has already invested in a founder using AI to improve clinical trials, and in several companies building vertical AI models for specific therapeutic areas.

Zhang cited a company developing a vertical AI model for cell therapy, effectively building a digital twin of the human cell, as one example, commenting, “Just think about different indications — they can potentially just mimic the whole process of evolution of the human cell (under) different disease conditions.”

She said the company recently announced a major partnership with a large European pharmaceutical firm focused on Parkinson’s disease, involving assets “a couple of hundred million dollars” in size.

Zhang also pointed to other portfolio companies, including one using microglial cells and vertical AI to treat Parkinson’s disease and dementia, and another using high-density ultrasound in combination with AI to treat depression via a non-invasive, highly targeted approach.

For Zhang, the key development in healthcare is that the value chain is finally becoming complete. Early diagnostics, once under-rewarded by the healthcare system, are increasingly being linked to targeted treatment plans.

That, she believes, is why “it’s a great time for AI healthcare.”