What AI Lacks Is Not Models, but "People Who Can Put Models to Work"

On August 31, MIIT issued a notice launching a special initiative to cultivate AI application service providers.
Several numbers are particularly noteworthy:
By the end of 2026, the service provider resource pool should exceed 2,000; by the end of 2027, no fewer than 3,000.
And what is even more interesting is that the notice is not just about large models.
It includes Agents, Token procurement, delivery and implementation, operations management, and even Forward Deployed Engineers.

What does this show?

MIIT has already seen the real challenge of AI deployment:What is lacking now may not be models, but "people who can put models to work."

You see, there are more and more models.
Enterprises increasingly want to use AI.
But when it actually reaches the factory floor, problems arise immediately:
Where does the data connect?
How should business processes be changed?
How is enterprise knowledge fed in?
What exactly does the Agent do?
Who is responsible when something goes wrong?
Who operates it after go-live?

Model companies solve "can AI do it," but enterprises care more about "can AI actually get work done."

So this policy is actually filling a previously overlooked link in the AI industry chain:
application service providers.

And its requirements for service providers are particularly interesting:
They must understand industry pain points,
understand technical mechanisms,
know security risks,
and be good at delivery and operations.
It even calls for building FDEs – Forward Deployed Engineers – to genuinely go on-site and deploy AI for users.

This means:

AI competition is moving from "model capability competition" to "application delivery capability competition."

01The Opportunity for Industrial Software Companies

And within this, industrial software companies actually have an opportunity.
Why?
Because although industrial software companies may not train large models themselves,
what they face every day is:
production processes, equipment data, business systems, industry knowledge, and on-site delivery.
These are exactly what AI needs most once it truly enters the enterprise.

So an interesting change may emerge in the future:
AI companies do not all have to build their own models.
Some handle models,
some handle Agents,
some handle data,
some handle industry applications,
and those who truly combine these things and deliver them to the enterprise site
may become an increasingly important new role:
AI application service providers.

02The Real Signal Released by the Policy

So the real signal MIIT is releasing this time is not:
"Enterprises should buy more Tokens in the future."
Rather, it is:

AI is transforming from a technology product into a productivity service that can be continuously delivered, continuously operated, and continuously consumed.

This also means
that in the next stage of AI competition, what matters may not be who has the largest model, but who best understands an industry and can truly drive AI into the business.

For industrial software companies,
this may be the next round of opportunity.