CEO Column | C4 | Next-Generation IR: Translating Your Moat into a Language AI Can Cite

Not “we are strong,” but whether AI can say for you that “they really are strong.”

Two weeks ago I met a large Japanese company in Tokyo. They didn’t want me to talk strategy or show a video; they just said: “Please give me your Moat File — I want to feed it into AI for comparison.” At that moment I suddenly understood: next-generation IR (Investor Relations) is not about persuading people but about teaching AI how to talk about you. You think you are speaking to investors, but you are actually speaking to their Agents, and an Agent is not moved by your story; it asks only one thing: “Is your moat standardized, verifiable and citable?”



01 | A Moat Is Not “We Have Technology” — That’s a Slogan

A future Moat
must meet three standards:

1. It can be broken down into metrics (Metrics)
2. It can be cited (Referenceable)
3. It can be compared (Comparable)

If your moat cannot be read by machines, it doesn’t exist.
Examples:
“Our product is two years ahead”
  → AI: Evidence? Version differences? Test data?

“Our team is world-class”
  → AI: Please provide structured résumés, patents, achievements.

“Our model is stronger than others”
  → AI: Benchmark? Test results? Inference cost?

An Agent doesn’t read adjectives. It reads the chain of evidence.



02 | The Harshest Review PSF Has Received

When working on a cooperation project in Brno, Czech Republic, the other side’s AI Agent, after reading our materials, threw back one line:
“Moat unclear. Metrics insufficient.” (The moat is not clear; metrics are insufficient.)
Not because we weren’t strong enough, but because we hadn’t presented ourselves “in a way AI can understand.” At that moment I became completely certain of one thing: a future moat is not built; it is “certified” by machines.

03 | So What Is a Moat That AI Can Cite?

The following is the most valuable knowledge you will get today. Moats that AI can cite usually involve four kinds of data structures:

1. Technical Moat

AI-readable formats include:
* Model Card
* Performance metrics (Latency / Throughput / Accuracy)
* Energy Curve
* RPS capacity (Requests Per Second)
* Inference Cost Model
* Version History

One line is worth more than saying it 100 times: “Latency P90 = 38ms (validated in environment X)”



2. Data Moat

Not “we have a lot of data”
but:
* Volume
* Source
* Permissions
* Cleaning process
* Degree of structuring
* Retraining cycle
* Pipeline evidence chain

What AI wants is “Data Lineage.” Have this, and you’re already halfway ahead of others.



3. Workflow Moat

What you can do and your competitors cannot is often not the model but the process. You must let an Agent read and understand:
* SOPs
* SLAs
* Human-AI collaboration boundaries
* Risk matrix
* Rollback strategy
* Tiered permissions
* Cross-department nodes

What AI wants is “whether your company runs predictably.”


4. Capital & Supply Moat

In the era of the compute war this matters even more:
* GPU exclusivity
* Scarce supply volume
* Cross-border energy agreements
* Government cooperation
* Financing structure
* Depreciation model
* Cash-flow buffer

When AI analyzes you it runs a sensitivity test; how many shocks you can withstand, AI sees even more clearly than people do.



04 | What Will Investors Buy? Not Stories but “AI Citability”

Let me give you a realistic fact first: it is not that investors don’t believe you; their Agents won’t let them believe you. Because an Agent scores directly:
* Moat completeness
* Moat comparability
* Moat verifiability
* Moat defensibility
* Moat operationality

Once your documents are uploaded, AI does three things:
1. Parse (Extract)
2. Compare (Benchmark)
3. Recommend (Recommend / Reject)

And AI’s recommendation matters more than your passion.


05 | How Do We Do IR at PSF? (You Can Do the Same)

In Japan, the Czech Republic, Singapore and the UAE we now use one method:

1. Break every moat into “metric + evidence”
Moat = Metric + Evidence + Process

2. Make every metric citable
For example:
* GPU utilization API
* Delivery success rate API
* Automatic PUE reporting
* Contract rollback mechanism JSON
* SLA attainment database

3. Make every narrative cross-verifiable
We don’t say “very good”; we state:
* Source
* Assumptions
* Sensitivity
* Method of verification

06 | Conclusion: A Moat in the AI Era Is a “Strength Understood by Machines”

You tell stories; AI wants specifications.
You talk about vision; AI wants models.
You talk about a moat; what AI wants is: “Moat validated. Competitive advantage confirmed.”

This is the IR of the future — not for people to read, but for machines to read.