Conclusion

Let’s get straight to the point. AI is not万能 (all-powerful). There are instances of ‘expensive AI.’

An executive from Nvidia has acknowledged that the inference costs of AI have not yet fully decreased.

GPUs are in short supply, energy costs are skyrocketing, and training models can cost tens of millions to hundreds of millions of yen. Even using large language models via APIs can accumulate token charges. It’s common for companies to find that a monthly expense of several hundred thousand yen has unexpectedly turned into an annual cost of five million yen.

And now, there are three simultaneous phenomena occurring worldwide:

• The AI Bubble Collapse Theory—Investors are starting to ask, ‘Where are the returns?’
• The ‘Handmade Should Be the Norm’ Rebellion—Artists, craftsmen, and creators are pushing back against AI-generated content.
• The Situation in SMEs—An increasing number of businesses are questioning, ‘Do we really need AI?’

These three issues may seem unrelated, but they all stem from the same root. Everyone is beginning to confront the question: ‘Which is truly cheaper, the cost of AI or the cost of humans?’

This article will directly address that question.

—

Is ‘AI is Cheap’ True?—Breaking Down Cost Structures

First, let’s dispel a common misconception.

‘Implementing AI will reduce costs’—this is half true and half false.

To be precise, there are areas where AI is cheap and areas where AI is expensive. What SMEs often mistakenly do is apply examples meant for large corporations directly to their own situations.

Large companies can achieve results with AI because they handle vast amounts of data. In a call center processing one million inquiries per month, introducing AI can dramatically reduce the cost per inquiry. However, what happens when a small local business with only 100 inquiries per month implements the same system? The fixed costs are too high, and the cost per inquiry may actually become more expensive than using humans.

Let’s take a closer look.

AI Costs vs. Labor Costs: A Real Comparison for SMEs

The points are clear.

‘Tasks that are high in volume, standardized, and where mistakes are not fatal are cheaper with AI.’
‘Tasks that are low in volume, non-standardized, and require trust are cheaper with humans.’

That’s it. Whether you can make this distinction will drastically change the returns on AI investments.

—

What Nvidia Executives Really Wanted to Say

Nvidia’s CEO Jensen Huang has repeatedly stated that ‘the costs of AI are rapidly decreasing.’ Naturally, he would say that since he wants to sell GPUs.

However, at the same time, Nvidia executives have begun to say something else. The inference costs (the costs of actually running AI) are still high. While training costs are one-time expenses, inference costs accumulate every time you use it. Every day, every hour, every second.

There are estimates that OpenAI spends over $700,000 (about 100 million yen) per day to operate ChatGPT. Of course, this is a case on a global scale, but the structure is the same. AI becomes ‘more expensive the more you use it.’ Human salaries are fixed monthly, but AI inference costs are pay-per-use.

In other words, it’s not that ‘using AI extensively’ makes it cheaper, but rather that ‘using AI wisely and selectively’ makes it cheaper.

If this is misunderstood, SMEs will end up being financially burdened by AI.

—

What the ‘Handmade Rebellion’ Teaches Us

Creators around the world are raising their voices against AI-generated images, text, and music, declaring, ‘That is not creation.’

Some may dismiss this as ‘emotional resistance,’ but that is a shallow interpretation.

What is happening here is a market principle: ‘As the costs of AI have decreased, the value of AI-generated content has also decreased.’

Things that anyone can produce have no value. Who would pay 100,000 yen for a logo that can be generated in 30 seconds by AI? Who would bookmark blog posts that are churned out by AI?

When costs decrease, prices also decrease. This is basic economics.

Conversely, the fact that ‘humans did it’ is beginning to hold intrinsic value.

• Handwritten letters
• Products finished one by one by craftsmen
• Customer support provided by recognizable humans
• Information gathered by local individuals

These are not ‘expensive.’ They have become ‘rare.’

For local SMEs, this is nothing short of a tailwind.

Why? Large corporations win by scale. They mass-produce with AI, reduce costs, and compete on price. That’s how large companies operate.

However, ‘handmade,’ ‘done by humans,’ and ‘recognizable’—these do not scale. Because they do not scale, large corporations cannot imitate them. This is a weapon that only SMEs can possess.

—

So, What Should SMEs Specifically Do?

I have three recommendations.

1. Cut Back Office Tasks with AI Thoroughly

Invoice processing, data entry, standard email replies, meeting minutes, and schedule adjustments. SMEs that are using human time for these tasks should stop immediately.

With tools costing a few thousand to tens of thousands of yen per month, you can save over one million yen in annual labor costs. This has been tested. In one of our clients, automating invoice processing alone resulted in an annual cost reduction of 800,000 yen. The initial setup cost was 50,000 yen. The ROI was 16 times.

2. Keep ‘Humans’ in Customer Interactions. In fact, Strengthen Them

Phone support, complaint handling, proposal sales, and follow-ups. If you think you’re ‘streamlining’ by introducing AI here, customers will leave.

Especially in local B2B businesses, the relationship of ‘we order because that person is there’ is the foundation of sales. The moment you replace this with an AI chatbot, you lose customers to competitors.

Use the time saved by AI to focus on customer interactions. That is the right approach.

3. Clearly Showcase ‘Humans Are Doing It’

Being handmade, having humans involved, and being locally produced—don’t hide this. Instead, put it front and center.

In an era overflowing with AI-generated content, saying ‘we have humans writing,’ ‘craftsmen inspect each item,’ and ‘local staff will visit directly’—this is the differentiation that is already upon us.

—

Summary: AI is a Tool. Don’t Misuse It

Whether the AI bubble will burst is uncertain. However, one thing is clear.

In areas where AI costs have decreased, the value of AI-generated content also decreases.
In areas where AI costs are still high, humans are cheaper and of higher quality.
And the very fact that ‘humans did it’ is beginning to create new value.

If you understand these three structural changes, the strategies for SMEs become clear.

• Automate the backend with AI to cut costs
• Retain humans on the front end to increase value
• Use ‘humans are doing it’ as a weapon to compete on a different playing field than large corporations

Before investing 3 million yen in an AI system, first try automating invoice processing with a tool costing 5,000 yen per month. Then use the time saved to make one phone call to a customer.

That will likely yield a much higher increase in sales.