Conclusion First: Relying Solely on ChatGPT is Now a Risk

Eighty percent of AI-related spending in the U.S. Congress is concentrated on OpenAI’s ChatGPT, according to recent research. Even the world’s largest democratic institution has narrowed its AI options down to just one company.

This is no laughing matter. What about your company?

Are your employees relying solely on ChatGPT for their AI tools? If five people share the $20 monthly personal plan, that totals $145 per month. With the Team plan, it would be $125 per month for five users (approximately ¥18,750). Annually, that amounts to about ¥220,000. You might think that’s cheap. But the real issue isn’t the cost itself. What if that one company raises its prices? What if it changes its API specifications? What if it shuts down its service? The real risk is having zero alternatives.

The True Danger of ‘Single-Vendor Dependency’ is the Loss of Negotiating Power

The case of the U.S. Congress illustrates a structural problem: concentrating on one vendor because it’s convenient can lead to a situation where you can’t move later.

There’s a term called vendor lock-in. Essentially, it refers to a state where the switching costs are so high that you have no choice but to accept the vendor’s terms. Large corporations may still have negotiating power, but small and medium-sized enterprises (SMEs) do not. They are seen as “customers who can be easily cut off,” so when prices go up, they have no choice but to comply.

In fact, OpenAI has changed its pricing structure multiple times over the past two years. The API for GPT-4 initially cost $0.03 per 1K tokens, but the pricing structure itself has changed with each model generation. This trend is likely to continue. There are reports that ChatGPT Plus may increase from $20 to $30 per month.

When you depend on a single vendor, you are at their mercy with every change. This is why the notion of “it’s cheap, so it’s fine” is not sufficient.

So, What Should You Do?—Options Have Dramatically Increased by 2025

Now, let’s get to the main point. I’m not saying to stop using ChatGPT. What I’m saying is to stop being in a state of only knowing ChatGPT.

As of 2025, the realistic options for SMEs to use AI have exploded. Moreover, costs have dramatically decreased. Let’s look at some specifics.

Option 1: Google Gemini—Already Quite Usable with a ‘Free Tier’

Google’s Gemini offers the Gemini 2.0 Flash even on its free plan. The Gemini Advanced plan costs ¥2,900 per month, but its integration with Google Workspace is powerful. For SMEs already using Gmail and spreadsheets, the switching costs are low. Just by using it alongside ChatGPT, the risk of single-vendor dependency is halved.

Option 2: Claude (Anthropic)—Superior in Long Text Processing and Accuracy

Anthropic’s Claude excels in reading and analyzing long texts with high accuracy. For tasks that require precision, such as reviewing contracts, summarizing manuals, and organizing meeting notes, there are scenarios where it is better suited than ChatGPT. The Pro plan is available for $20 per month (approximately ¥2,900).

Option 3: Local LLMs—A World of ‘Zero Monthly Fees’

This is where the structural change is most significant. Open-weight models like Llama 3.1, Mistral, and Phi-3 can run on your own PC or server. This means monthly fees can be zero.

You might think, “But don’t you need a high-performance PC to run them?” That was true in the past, but not anymore.

• Using a tool called Ollama, models with 7B to 13B parameters can run smoothly on a MacBook (with M2/M3 chips).
• The only initial cost is the purchase of the PC. If you already own an M2 MacBook, there are no additional costs.
• Even a 13B parameter model is sufficient for internal inquiries, drafting emails, and summarizing meeting notes.

Monthly costs of ¥2,900 × 10 employees = ¥29,000, totaling ¥348,000 annually. This could potentially drop to zero. For SMEs, this difference is significant.

Option 4: ‘Use Different Tools for Different Purposes’—This is the Most Cost-Effective Approach

In fact, the most realistic approach is not to try to do everything with one AI.

By selecting the optimal tool for each purpose, you can accomplish more for a monthly cost of ¥3,000 to ¥5,000 per person, compared to relying solely on ChatGPT Plus. Plus, you won’t be dependent on a single vendor. If one raises its prices, you have alternatives.

Is ‘In-House Development’ Necessary for SMEs?—The Answer is Almost No

There’s often talk about the need for companies to develop their own AI models, but to be honest, there’s almost no point for SMEs to develop their own LLM from scratch.

The training costs for a model comparable to GPT-4 are said to be in the billions of yen. What SMEs should focus on is not ‘development’ but ‘selection and combination.’

However, fine-tuning (adjusting existing models) is a different story. By using your company’s operational data, past proposals, and customer interaction logs, you can specialize open-weight models for your business. This can be done for tens of thousands to several hundred thousand yen. Cases where outsourcing can be done for under ¥500,000 are increasing.

In the past, “custom AI” was a privilege reserved for large corporations. Now, even companies with ten employees can access it. This is the real structural change.

Three Actions SMEs Should Take

Now that you understand the reasoning, what should you do starting tomorrow?

1. Take Inventory of Your Current AI Tools (Time Required: 30 Minutes)

Identify who in your company is using which AI tools and for what purposes. First, get a grasp of this. If you find that “ChatGPT is the only tool being used,” recognize that itself is a risk.

2. Add One Free Tool to Try (Time Required: 10 Minutes)

Try using one AI with a free plan, such as Gemini, Claude, or Perplexity. Ask the same question to two different AIs and compare the differences in their responses. Just doing this will help you experience the precariousness of single-vendor dependency.

3. Try Running a Local LLM on One Machine (Time Required: 1 Hour)

If you have a MacBook from the M2 series or later, try installing Ollama. With just two commands in the terminal, a local AI will start running. Experiencing, “Wait, this runs for free?” will lead to a reevaluation of your cost structure.

Are You Stopping Your Thinking Because It’s ‘Cheap’?

Eighty percent of AI spending in the U.S. Congress is concentrated on ChatGPT. This is not to say that “ChatGPT is bad.” The danger lies in the structure of concentrating on one vendor without considering alternatives.

For SMEs, AI costs may seem like a world of “a few thousand yen per month.” Therefore, the sense of urgency is low. However, the true costs are not just the fees. There are risks of price increases, service shutdowns, and costs associated with adapting to specification changes. When considering the “total cost” that includes these factors, single-vendor dependency is far from cheap.

In 2025, the options for AI are more abundant than ever, and costs are lower than ever. Shifting from “just ChatGPT” to “choosing based on purpose” is essential. SMEs that can make this switch will see their cost structures change.

Start today, whether it’s Gemini or Claude. I encourage you to ask the same question to an AI other than ChatGPT. That will be the first step toward breaking free from single-vendor dependency.