I have been interested in a possibility that sits awkwardly beside the excitement around frontier AI: some valuable productivity improvements may come from applying relatively ordinary technology in places where it has not spread.
The intuition includes emerging-market businesses, physical infrastructure, and fragmented industries. A basic operating system, better equipment scheduling, or a reliable financial pipeline could matter more to a particular company than a new model capability. That is a hypothesis about adoption and organization, not evidence that any broad market is cheap.
My finance and technical experience make me want to examine both sides of the transaction: what a tool can improve and what prevents a business from capturing that improvement.
A capability gap is not yet an opportunity
Suppose two facilities use the same equipment differently. One has a reliable maintenance process and the other does not. It is tempting to infer that transferring the process will create a predictable gain.
The visible difference might instead reflect power reliability, financing, labor arrangements, supplier access, or the cost of downtime. A tool designed for the first facility may assume precisely the infrastructure the second lacks. Calling this “low-hanging fruit” can hide the hard complementary work.
The useful question is which constraint binds. If the problem is missing information, a sensor or workflow can help. If the problem is an unavailable replacement part, a more accurate diagnosis may simply document the delay. If the problem is working capital, a software subscription can worsen the immediate cash problem even when the eventual efficiency gain is real.
Technical savings must reach the cash flow
An engineering model can estimate energy saved under an intervention. An operating model asks when the change can be installed, whether workers use it, and whether production is interrupted. A financial model asks who pays, when savings arrive, and who can retain them.
A schematic project value would subtract installation, downtime, maintenance, financing, and organizational costs from the discounted operating improvement. Every term can depend on the local institution. A small measured technical gain can be commercially meaningful; a large theoretical gain can be unfinanceable.
This is one reason I find physical AI interesting from both research and investment perspectives. A model that chooses the right observation or intervention could reduce implementation uncertainty. It still needs a path through procurement, operations, and financing.
Roll-ups do not remove the mechanism
Combining fragmented businesses could share expertise, purchasing, software, and capital. It could also add integration costs and remove useful local knowledge. The investment thesis should specify which repeated task becomes better or cheaper and why common ownership is necessary to obtain that improvement.
A software vendor, a specialist service, or a financing arrangement might deliver the same operating change without acquisition. Those are real counterfactuals. An attractive acquisition model does not prove that the ownership change caused the productivity gain.
The comparison should also include the businesses that were difficult to integrate. Selecting only the successful installations would make a standardized intervention look more portable than it is.
The evidence I would want
Start with one recurring operating problem and a small set of comparable sites. Record the initial constraint, total implementation work, adoption, cash timing, and measured operating outcome. Compare the intervention with the best practical alternative available to those sites.
Then ask what transfers to the next location. If every deployment requires extensive bespoke work, that can still support a service business, but it weakens the claim of cheap technology diffusion. If a reusable process survives local differences, the evidence becomes more interesting.
I do not have a completed empirical study behind this broader thesis. I want it to remain in the set of questions I investigate because it forces a useful discipline: explain why an existing capability has not spread, and show that the proposed institution can overcome that reason. The technological frontier and the adoption frontier can be far apart. Neither distance alone tells us where the return is.