Conclusion

To put it simply, the “price disruption” in AI development has begun.

AI has started creating AI.

This is not just a technically interesting story; it represents a structural change that fundamentally alters the management decisions of small and medium-sized enterprises (SMEs).

Until now, the introduction of AI meant initial development costs ranging from 3 to 10 million yen, with monthly operational costs in the tens of thousands of yen. For local SMEs, this was a matter of “not relevant to us.” However, now, technology that allows AI to improve its own algorithms, miniaturize, and enhance efficiency automatically has reached a practical stage. What lies ahead?

A scenario where for just 50,000 yen a month, SMEs can acquire processing capabilities equivalent to AI systems built by large corporations at costs of tens of millions is becoming increasingly plausible.

In this article, we will clarify what exactly is happening when we say “AI creates AI,” how this changes the cost structure, and what SMEs should do now, supported by numbers and examples.

—

What Happens When “AI Creates AI”

There is a term called “recursive self-improvement.” This refers to a mechanism where AI evaluates its own model structure and algorithms and automatically generates more efficient versions.

Traditionally, improving AI models required human engineers to spend weeks to months on tuning. Adjusting hyperparameters, redesigning architectures, and preprocessing data—all of these tasks required specialized human labor. In other words, “labor costs” accounted for the majority of AI development costs.

Now, AI itself has begun to intervene. In research by Google’s AutoML and Meta’s LLM, there are reports of AI exploring model structures on its own and achieving accuracy that surpasses human-designed models. IEEE Spectrum has also featured the “mechanism by which AI builds better AI,” demonstrating that this trend is not just a passing fad.

The key point is that “the labor hours required for engineers will drastically decrease.” Tuning tasks that previously cost 1 million yen a month will be automated through AI self-improvement. Humans will only need to define “what they want to solve.” The structure of development costs itself will change.

—

Another Revolution — The Era When AI Runs on “Personal Computers”

There is another factor changing the cost structure: the miniaturization of AI models.

Noteworthy is the practical application of Ternary Neural Networks (TNNs). While traditional AI models calculated using 32-bit or 16-bit floating points, TNNs compress parameters to three values: “-1, 0, +1.” This dramatically reduces model size, allowing them to run efficiently on consumer-grade CPUs.

Let’s look at some specific numbers:

• Litespark (a new inference system): Achieves up to 52 times the throughput compared to traditional AI frameworks. This means it can perform the same processing with just 1/52 of the computational resources.
• HCInfer (an efficient inference engine): Capable of executing high-accuracy inference even on devices with 4GB of memory.

What this means is that AI processing can now run on existing PCs within companies without the need to rent expensive cloud GPU servers.

Breaking down the traditional AI operational costs looks like this:

In rough terms, this is less than one-tenth of the previous costs. This is the basis for the scenario of “AI comparable to that of large corporations for just 50,000 yen a month.”

—

What Values Increase and Decrease with This Change

Now we get to the main point. The technical details are fine, but the real question is what changes for SMEs with this shift.

Values That Decrease

The value of “just having AI” itself decreases.

If the cost of AI implementation drops to 50,000 yen a month, any company can adopt it. This means that saying “we use AI” will no longer be a differentiator. If a demand forecasting AI that large corporations built for tens of millions can be obtained for 50,000 yen, the difference in investment will not provide any advantage.

This is good news for SMEs. The disparity in financial power will become irrelevant in this domain.

Values That Increase

The value of the ability to decide “what to make AI do” increases.

As tools become cheaper, the competition will be determined by “how they are used.” Specifically:

• The ability to identify where to integrate AI into one’s own business processes.
• The ability to articulate customer challenges and translate them into instructions for AI.
• The ability to translate AI output into usable forms on-site.

In some cases, SMEs may have an advantage over large corporations in this regard. Why? Because decision-making is faster.

While large corporations take three months to circulate approval documents for AI implementation, SMEs can say, “Let’s try it first,” and test it the following week. Given the rapid pace of AI evolution, this agility will become a decisive advantage.

—

What Specifically Will Happen — Three Scenarios

Scenario 1: The Resolution of Individualization Happens “Automatically”

In small manufacturing enterprises, the issue of tacit knowledge being individualized among veteran craftsmen. Previously, it would cost hundreds of thousands of yen in consulting fees to create manuals.

Using AI self-improvement and miniature models, it will become possible to film the work of veterans, have AI automatically analyze the steps, and generate manuals with text and images—all for under 50,000 yen a month. Moreover, as AI continues to self-improve, the accuracy of the manuals will increase. Humans will only need to check, “Is this correct?”

Scenario 2: A “24-Hour Analysis Team” for 50,000 Yen a Month

In local retail, an AI that integrates sales data, weather data, and social media reviews to conduct demand forecasting will run on the company’s PC 24/7. Previously, hiring a dedicated data analyst would cost 5 to 7 million yen annually. Now, an AI for 50,000 yen a month can continuously perform the same analysis automatically.

Humans will only need to look at the AI’s forecast results and make the final judgment on inventory levels. If a 10% to 30% optimization of inventory is achieved, it could lead to cost savings of several hundred thousand yen a month just from reducing food waste. The return on investment could be realized in the first month.

Scenario 3: A “Solo CEO” Has an AI Team

Freelancers or solo CEOs can delegate tasks such as creating proposals, conducting market research, providing first responses to customer inquiries, and bookkeeping to AI. If a 50,000 yen AI can replace outsourced tasks that previously cost 200,000 to 300,000 yen a month, that amount can be redirected to core business activities like sales or product development.

One person can now do the work of “five people.” This is the reversal structure for SMEs.

—

So, What Should We Do?

“Amazing technology is coming. So let’s prepare” — many articles say this, but they don’t specify what to do concretely. Let’s be clear.

There are three things you should do today.

1. List Three “High-Cost Tasks” in Your Company

AI is not omnipotent. However, it excels in tasks that are “repetitive,” have “clear criteria for judgment,” and “take a long time when done by humans.” First, list three tasks in your company that meet these criteria.

2. Try One AI Tool That Costs 50,000 Yen a Month

There’s no need to immediately implement self-improving AI. However, experimenting with integrating tools like ChatGPT (3,000 yen a month) or Claude (3,000 yen a month) into your operations can start today. Building the “muscle to use AI” now will determine how quickly you can absorb the core technology when it arrives.

3. Articulate Your Company’s Strengths That Cannot Be Replaced by AI

As AI becomes cheaper, the value of what AI can do will decrease. Conversely, the value of things that AI cannot do — such as building trust with local customers, understanding nuances that can only be grasped on-site, and long-standing relationships with suppliers — will increase relatively. Articulating these strengths will become the starting point for management strategies in the AI era.

—

Summary: The Era of the “Have-Nots” Has Arrived

The era when AI creates AI breaks the structure where “those with large amounts of capital win.”

When development costs drop to one-tenth, the “barrier of upfront investment” for large corporations becomes meaningless. SMEs will acquire AI capabilities equivalent to those that large corporations spent tens of millions on for just 50,000 yen a month. At that point, the differentiating factor will not be financial power, but rather “the speed at which decisions are made about how to use it.”

For local SMEs, this is not a threat. It is a chance.

While large corporations are debating in boardrooms, SMEs can test, fail, and improve on the ground. This cycle of agility will become the greatest weapon for SMEs.

If AI is improving itself, we must continue to improve our own methods.