Conclusion First: The “Price” of AI Has Started to Break Down

Until just six months ago, if you wanted to use a decent AI model through an API for business, you had to be prepared to spend around 50,000 yen a month. Incorporating a GPT-4 class model into an in-house chatbot or document generation could easily exceed 100,000 yen a month just in token charges.

Now, it’s becoming free.

Chinese AI labs are continuously releasing open-weight models, leading to a proliferation of AI tools that can be used for free without limits. Models like DeepSeek, Qwen, and GLM are being released under commercially usable licenses, achieving benchmark results that rival or even surpass those of GPT-4o and Claude 3.5.

Even more shocking is that the coding-specialized model “DeepSWE” has recorded scores surpassing GPT-4.5 on SWE-bench (a software engineering practical benchmark). Moreover, the API usage costs are just a fraction of what they used to be. The conventional wisdom that “high cost equals high performance” has been completely overturned.

What I want to convey in this article is not just that “amazing AI has emerged.” It’s about how the collapse of AI costs will change the competitive landscape for small and medium enterprises. I will focus solely on that.

What’s Happening—A Look at the “Price Collapse” Through Numbers

First, let’s present some concrete numbers.

When calculated annually, small and medium enterprises that were spending 600,000 to 1,200,000 yen on AI tools can now obtain equivalent or better functionality for almost zero cost.

Why is this happening? There are three reasons.

1. The “Open Strategy” of Chinese AI Companies
DeepSeek and Alibaba (Qwen) are expanding their ecosystems by releasing models as open-weight, recovering costs through cloud infrastructure and API charges. In other words, the models themselves are “bait.” This is the opposite approach to OpenAI’s strategy of locking users in with a $20 monthly subscription, resulting in the price of the models available to users approaching zero.

2. The Plummeting Learning Costs
The reported learning cost for DeepSeek V3 is about $5.6 million. Considering that the learning cost for GPT-4 is estimated to be around $100 million, that’s about one-twentieth. As learning costs decrease, barriers to entry lower. With more entrants, price competition arises. This is a textbook scenario.

3. The Reality of Local Execution
Advancements in model lightweighting technology (quantization) mean that distilled versions like Qwen2.5-7B and DeepSeek-R1 can run on gaming PCs costing around 100,000 yen. There’s no longer a need to incur cloud charges. Monthly costs can be zero, with only electricity bills to consider.

Is “Waiting for Lower Prices” True?—Half True, Half Wrong

Here, let’s dispel a common misconception.

“If AI is getting cheaper, wouldn’t it be better to wait for implementation?”

This is half true and half wrong.

The true part: The prices of tools are indeed continuing to drop. What costs 50,000 yen this month might be free next month. There’s no need to rush into a costly annual contract. Especially if you’re getting estimates in the hundreds of thousands of yen for “AI tool implementation” from external vendors, waiting just three months could dramatically change the situation.

The wrong part: Even if tools become cheaper, the “ability to utilize them” won’t develop just by waiting. This is critically important.

When the price of AI tools reaches zero, where will the competitive advantage shift? The answer is clear: the ability to design how to integrate AI into one’s business and the experience of the team executing it. This cannot be accumulated without actual use.

There’s a manufacturing company in a rural area that has incorporated AI image recognition into its inspection process. The tool itself is open-source and free. However, it took three months to figure out how to create training data tailored to their products, how to integrate it into the production line, and where to set the threshold for defect judgments—this “last mile” took three months. This experience from those three months is something that competitors will find hard to replicate later.

In other words, waiting to save on tool costs is correct. But postponing “hands-on experience” is incorrect. If free tools are available, you should start using them today.

The “Reversal Structure” for Small and Medium Enterprises

Now, let’s get to the main point. What happens when AI costs approach zero?

The systems that large corporations have spent hundreds of millions of yen to build—equivalent performance can now be obtained by small and medium enterprises for zero monthly cost. This is not just about “catching up.” It’s about the structural advantage that small and medium enterprises will gain.

The reason is simple: large corporations’ existing AI investments become “sunk costs.” Making the decision to replace systems built at the cost of hundreds of millions of yen with free tools is organizationally extremely difficult. Approval processes, security reviews, contracts with vendors—these slow down decision-making.

On the other hand, small and medium enterprises have nothing to lose. If the president says, “Let’s use this,” they can act the next day. This speed of decision-making will be the greatest weapon in the zero-cost era.

To be specific:

• Customer Support: Outsourcing a GPT-4o-based chatbot would cost an initial fee of 1 million yen plus a monthly fee of 50,000 yen. Now, if built in-house based on DeepSeek, it can be effectively free. If it can absorb the work of two out of five phone support staff, that’s an annual personnel cost saving of over 6 million yen.
• Sales Materials and Proposal Creation: Creating a proposal that used to take three hours per item can now be done in 30 minutes with AI. If sales representatives can handle 20 proposals a month, the impact on sales is immeasurable.
• Accounting and Invoice Processing: Automatic reading and categorization of invoices using OCR + LLM. If you were using a SaaS costing 30,000 yen a month, switching to an open model for in-house operation would save 360,000 yen annually.

None of these are about “what AI can do.” They are about “how much money can be saved each month” and “how many jobs can be freed up.”

So, What Should We Do?

I’ll say three things.

1. Make High-Cost AI Contracts Short-Term
There’s no guarantee that current prices will remain six months from now. In fact, they will certainly drop. Avoid SaaS that ties you into annual contracts and choose monthly contracts or pay-per-use options. If you’re lured by “20% off for annual contracts” and a free alternative tool comes out six months later, you’ll be in trouble.

2. Start Using Free Tools “Today”
DeepSeek’s chat (chat.deepseek.com) is available for free right now. Google’s “Gemini” also has a free tier. First, try it out in your own work. Summarizing meeting notes, drafting emails, organizing data—anything will do. Only those who have tried can judge, “This is usable” or “This is still not feasible.”

3. “Train AI Talent” Instead of “Hiring AI Talent”
The cost of hiring AI talent is skyrocketing. In Tokyo, salaries range from 8 million to 12 million yen. This is not a competitive arena for small and medium enterprises in rural areas. However, you can have your current employees use AI tools and gain experience integrating them into their work. If the tools are free, the education costs are almost zero. Instead of hiring from outside, cultivate from within. This is the realistic solution for small and medium enterprises.

What Lies Ahead After the Price of AI Has Crashed

The collapse of AI prices means that the differences in AI itself will diminish.

Everyone can use the same free tools. So, what will create the differences? “How deeply do you understand your own business?” and “How specifically can you translate on-site challenges into AI?”—in the end, the competition returns to “on-site capabilities.”

This is good news for small and medium enterprises in rural areas. They know the field. They see their customers’ faces. Their decision-making is swift. Companies that have these three elements will become the biggest beneficiaries in the zero-cost AI era.

Conversely, companies that continue to say, “AI seems difficult” or “It’s still too early for us” will realize it when their competitors have armed themselves with free tools. By the time they notice, the difference will not be “the presence or absence of tools” but “the difference in experience.” This is something that cannot be bought with money.

The tools are already lying around for free. The only question is whether to pick them up.