> ## Content Index
> Fetch the complete content index at: https://research.porbital.io/llms.txt
> Use this file to discover other available public pages before exploring further.

# How Industrial AI Can Raise Productivity Without Raising Producer Profit
- URL: https://research.porbital.io/how-industrial-ai-can-raise-productivity-without-raising-producer-profit/
- Published: 2026-09-17T12:09:10.000Z
- Updated: 2026-10-17T07:49:26.000Z
- Description: AI can lower a factory’s costs without raising its profit. Who keeps the savings depends on competition, supply-chain coordination, and access to customers.
- Author: Roverzon
- Tags: AI and Productivity, #i18n-industrial-ai, Industrial AI, Manufacturing, Supply Chains, Platform Economics

## Quick read

- **Higher productivity does not guarantee higher profit.** When rivals can adopt the same AI tools, competition may pass the savings to buyers.
- **The factory is only part of the system.** Faster production does not automatically improve coordination across orders, components, financing, and delivery.
- **Who keeps the gains depends on scarce resources.** As AI becomes widely available, customer access, orders, and supply-chain coordination may be harder to replicate than the model itself.

---

*AI can make a factory more productive without making it more profitable. The question is who gets to keep the savings.*

Suppose a factory sells a component for $10\. It costs $8 to make, leaving $2 in profit. An AI tool reduces the cost to $6\. At the same selling price, profit doubles.

That looks like a good investment. Then the factory's competitor buys the same tool.

Both producers now have a $6 cost. If buyers regard their components as interchangeable, they can play one supplier against the other. Should competition bring the price down to $8, both factories are back to earning $2 per unit.

The productivity gain is real. So is the possibility that the buyer, rather than the producer, captures it.

This is a simplified example, and prices don't have to behave this way. It does get at the question I think matters most in industrial AI: once the technology is widely available, what lets a company keep the value it creates?

## Lower Costs, Unchanged Profits

Go back to the moment before the second factory adopts AI. The first factory has a clear advantage: it can earn more at the existing price, cut its price to win orders, or do some of both. That advantage comes from a difference its competitor can't yet match.

Once the competitor catches up, the difference is gone. Both factories are more efficient, and neither has any more bargaining power than before; buyers can use the restored competition to negotiate away the savings. The factory didn't lose its efficiency. It lost the gap.

The assumptions matter, of course. The products have to be interchangeable, and buyers need a credible choice of suppliers; nobody's price falls to a particular number just because everyone adopted AI. Change the assumptions and the outcome changes: if the first factory uses AI to build process knowledge the second can't reproduce, it keeps an advantage, and the same goes for better quality, unique data, distribution, and customer relationships. Two businesses using the same model are not the same business. What the example shows is that access to the model, by itself, is a weak explanation for why one of them should earn more.

So there are two numbers a producer should care about: how much the tool saves, and how quickly competitors can match the result. The first measures the productivity gain. The second determines how much of it the producer keeps.

## The Factory Is Only Part of the System

For roughly a decade I have looked at industrial intelligence through China's manufacturing development, and my starting point has always been how production is organized, not how sophisticated any one tool is. I think of that development as layers added in succession: standardization, informatization, automation, intelligence. Over the same period production itself spread from family workshops to factories, corporate groups, and industrial clusters, and the bigger the scale and the finer the specialization, the more the work depended on different firms making compatible decisions.

None of this is new with AI. The State Council's [Guiding Opinions on Actively Promoting the “Internet Plus” Action](https://www.cac.gov.cn/2015-07/04/c%5F1115815010.htm?ref=research.porbital.io), dated July 1, 2015, already named collaborative manufacturing as a priority, with Internet-enabled coordination across industrial chains on the agenda.

And yet from 2017–2018 onward, some of the manufacturers I followed, along with the first wave of vision-oriented AI vendors, did not grow stronger as digital tools spread; by 2026, in my observation, they look weaker. That is an observation about those firms, not an industry statistic, and it certainly doesn't show that AI caused their difficulties. It does raise a question: why didn't smarter technology produce stronger businesses?

One answer, I think, is that we treated two different problems as one: a firm making better decisions, and several firms working better together.

A factory knows its processes, equipment, and capacity. A component supplier knows what it can deliver and when. A buyer knows its budget, quality requirements, and expected demand. Each holds information the others need, and nobody holds all of it.

Production means fitting those pieces together. Orders have to match component availability; process choices have to satisfy both cost and quality; when demand changes, financing, schedules, location, and delivery may all need to change with it.

So even if every firm uses AI to sharpen its own analysis, the chain as a whole can still be stuck on coordination. A faster decision in one part of the chain isn't worth much if another part is the bottleneck.

This doesn't make a supply chain a free market, where everyone simply negotiates and adjusts. Contracts, technical standards, dominant buyers, and organizational authority all shape what each firm can choose. AI can help firms coordinate; the question is whether it actually gets into those relationships.

The argument has a boundary. A vertically integrated factory with a bounded, repeatable process can capture a lot from adopting AI internally. Coordination across firms matters most when production spans different organizations, and their inability to work together is what limits the result.

## A Reusable Model Is Not a Reusable Deployment

Industrial AI vendors have a related problem: the technology may be reusable when the work of delivering it is not. A large customer has its own processes, equipment, data, organizational rules, and supplier relationships, and making an AI system useful means adapting it to each of them. Then the next customer shows up with a different set.

[SenseTime's 2024 Sustainability Report](https://www.sensetime.com/cn/csr%5Fcorporate?ref=research.porbital.io) describes customized model training, incremental model training, and inference deployment as parts of its MaaS offering to industry partners. Handing over the model, in other words, is only part of the job.

Smaller factories are a different problem. They often lack the capital, the technical staff, or the expected return to buy and implement even a standardized product. So the vendor is stuck between large contracts that take heavy integration work and a broad market that can't readily buy; a technology can be valuable and still be a hard business to scale.

The number to watch is what happens to the cost of delivery as customers are added. If each deployment needs extensive new work, a reusable model doesn't give you the economics of reusable software.

(I look at the platform side of this in [AI Had Its iPhone Moment. It Won't Get the iPhone Economy.](https://research.porbital.io/ai-had-its-iphone-moment-it-wont-get-the-iphone-economy/): ongoing inference costs, and suppliers that compete with their customers, limit what independent builders get to keep.)

## Who Connects the Firms?

Xiaomi shows another way to organize the work: the company managing the supply chain brings the participants into a shared system, instead of leaving each supplier to improve by itself.

Its [2022 ESG Report](https://ir.mi.com/system/files-encrypted/nasdaq%5Fkms/assets/2023/04/26/10-13-04/2022%20ESG%20Report%5FENG.pdf?ref=research.porbital.io) describes a digital management system connecting procurement, inventory, production, partner management, and stages of R&D. By year-end it covered all primary smartphone production suppliers and some secondary ones; the report also describes visibility across overseas supply-chain operations, with timely warnings.

I want to be precise about what this shows. It shows an organizational basis for coordination: the firms aren't just analyzing their own data, they can plan procurement, inventory, and production together. It doesn't show that AI raised Xiaomi's profit or changed supplier margins; digitalization and AI deployment are different claims. (I'd put Foxconn's industrial internet in the same category, and comparing the two would take evidence on deployment and operating results.)

What these examples point to is the strategic question: when comparable AI capability gets easier to obtain, what stays scarce? My answer is orders, standards, financing, supplier relationships, coordination systems, and access to customers.

A company that controls those assets gets to shape which suppliers enter the network, what they must deliver, and how the work is allocated. That puts it in a much better position to influence how the resulting value is divided.

An independent AI vendor can improve an important task and still be just another supplier to the organization running the chain. It can make one step better without getting to decide what that improvement is worth.

My expectation that value moves toward large firms rests on two conditions: AI capability becomes widely available, and these complementary assets stay concentrated. If both hold, the firms that can keep investing and reach customers directly are the ones positioned to keep the gains.

This is how I read JD Industrial and Alibaba's Cainiao moving into industrial and logistics networks: as attempts to occupy that position, by combining AI with orders, capital, distribution, and business relationships. That is my interpretation of their strategy, not a finding about their returns.

## Better Economics, Greater Dependence

None of this means suppliers lose.

Steadier orders, faster payment, joint development, and access to technology or distribution can all make a supplier's business better; joining a well-organized supply chain is often more attractive than hunting for customers alone.

What comes with it is dependence. The company managing the chain still controls entry and access to customers, so a supplier can earn more without gaining any bargaining power.

A similar tension appears in payments. My analysis of [Circle’s move toward owning the payment last mile](https://research.porbital.io/circles-last-mile-gamble/) examines why local execution and customer access can matter more than the underlying settlement technology, and how owning that access changes a network’s relationship with its partners.

Weak consumer demand makes the dependence deeper: reliable orders get more valuable, and firms outside the network have a harder time funding investment or reaching customers.

The outcome I expect, then, isn't suppliers declining together. It is a wider gap between the firms inside the network and the firms left out of it, with even the winners more dependent than they were.

## What to Ask About an Industrial AI Business

When I look at an industrial AI business, I start with three questions.

**Can competitors reproduce the capability?** If rivals can use the same tools to do the same work, what preserves the company's advantage: process knowledge, data, quality, or customer relationships?

**Does adoption improve a task or improve coordination?** Are orders, components, standards, and production decisions better aligned? If the bottleneck remains elsewhere, a faster task may not increase total output.

**Can the company retain the gain?** Beyond the model, does it have the orders, capital, supplier relationships, or customer access needed to influence pricing and profit allocation?

These questions are also how to prove me wrong. If manufacturers and AI vendors can secure financing, coordinate suppliers, and reach customers on their own, value doesn't have to concentrate in a few firms, and my prediction should weaken accordingly.

(The retail version of this argument is in [my analysis of PDD and the limits of AI efficiency](https://research.porbital.io/when-efficiency-runs-out-what-pdd-got-right/): when merchants and buyers face their own constraints, how much of an efficient marketplace's value can the platform keep?)

Go back to the two factories. Cutting unit cost from $8 to $6 is a real improvement. Whether profit rises depends on what competitors can match, what buyers can substitute, and who gets to negotiate the price. Widen the lens to the whole production network and the answer is the same: if AI spreads while orders, capital, coordination, and customer access stay concentrated, the firms holding those are the ones that keep the new value.

Industrial AI deserves to be judged by what it improves. The next question is whose income that improvement becomes.

That income question extends beyond the factory. In [my analysis of AI, layoffs and long-term interest rates](https://research.porbital.io/ai-layoffs-demographics-higher-interest-rates/), I examine whether productivity gains become wages and taxable income, and what that means for public finances.