AIエージェント1体あたり月3万円。あなたの会社、何体使う？

Microsoft executives have suggested that AI agents will require licensing fees just like human employees.

This might not resonate with everyone immediately, but the essence is simple: AI is shifting from being a ‘free subordinate’ to becoming an ‘outsourced entity that sends monthly invoices.’

Currently, the license for Microsoft 365 Copilot is $30 per user per month (approximately 4,500 yen). What happens if this fee is applied not to the ‘number of humans’ but to the ‘number of AI agents’? For sales, accounting, customer support, inventory management—if you operate five agents, that’s 150,000 yen a month; ten agents would cost 300,000 yen. That’s 3.6 million yen annually.

For a small business with ten employees, this amount is no laughing matter.

Moreover, this is not just a Microsoft issue. The entire SaaS industry is beginning to pivot towards ‘billing based on the value of AI.’ Motorola has raised the price of its wireless communication software by 50%. Salesforce has also introduced usage-based billing for its AI agent features.

Just when you thought ‘AI would improve operational efficiency,’ you might find that your IT costs have doubled. That future is closer than you think.

課金モデルの構造変化——何が起きているのか

First, let’s clarify what is happening now.

Until now, SaaS billing was based on ‘number of humans × monthly fee.’ If you have ten employees, you need ten licenses. It’s straightforward.

However, as AI agents begin to handle tasks, this structure breaks down. Even if you still have ten human employees, if you are operating twenty AI agents, you will incur charges for ‘thirty licenses.’ Alternatively, usage-based billing may be added based on the agents’ workload.

This change has three stages.

Stage 1: Feature Billing (Current)
You purchase additional licenses to use AI features like Copilot or Gemini. The market rate is around 3,000 to 5,000 yen per user per month.

Stage 2: Agent Billing (Upcoming)
A license will be required for each AI agent. This is the phase Microsoft is hinting at. Agents will be treated as ‘digital employees,’ each holding a seat (license) just like humans.

Stage 3: Performance Billing (Future)
Billing will be based on the results generated by AI—such as the number of inquiries processed, reports generated, and transactions automated. The more you use it, the higher the bill.

What is frightening for small and medium enterprises is the possibility that Stages 2 and 3 could arrive simultaneously. The double punch of agent licensing fees plus usage-based billing. Companies that previously managed with ‘50,000 yen in monthly SaaS costs’ could find themselves facing ‘300,000 yen’ after implementing AI. Such scenarios are becoming increasingly plausible.

モトローラの50%値上げが示す「本当の意味」

Seeing Motorola’s price increase as merely ‘a price revision by one company’ is naive.

Motorola raised the price of its wireless communication software for public safety. This is not hardware but software. In other words, the ‘usage fees for already implemented systems’ have increased.

This is a classic example of SaaS vendors raising prices for existing products under the pretext of ‘adding AI features.’

Consider this: if you received a notification that your business software now has ‘AI features added, and the monthly fee will increase by 50%,’ what would you do? Switch to an alternative software? What about the cost of data migration? Retraining the staff?

AI is being used as a pretext for price increases against locked-in customers. This is the essence of what is happening in the industry now.

In fact, according to a Gartner survey, SaaS spending is projected to increase by 20% year-on-year in 2024, with a significant portion of that increase attributed to additional charges related to AI. While the typical IT budget for small businesses is around 3-5% of revenue, some companies may see that ratio jump to 7-10%.

中小企業の防御策——「依存しない」という選択肢

So, what can small and medium enterprises do? ‘Let’s calculate costs carefully’ and ‘let’s implement strategically’—such textbook advice won’t ensure survival.

Here are three specific actions they can take immediately.

1. オープンソースLLMで「自前AI」を持つ

Llama 3, Mistral, Gemma—free open-source large language models are rapidly reaching practical levels.

For example, if you delegate internal inquiries to Copilot, you’ll incur a license fee of 4,500 yen multiplied by the number of users. However, if you use Ollama to run Llama 3 on an internal server, you only need to cover the initial server costs (around 100,000 to 200,000 yen). The monthly running cost would be just the electricity bill.

What about accuracy? Honestly, there are situations where it falls short of GPT-4o. However, for internal FAQ responses and generating standard reports, it is more than sufficient. Would you pay 300,000 yen a month for a perfect AI, or operate an 80-point AI for 5,000 yen? This decision is clear for small and medium enterprises.

2. 「マルチベンダー戦略」でロックインを避ける

Becoming dependent on a single platform leaves you powerless against price increases. The Motorola case exemplifies this.

Specifically, do the following:

• Use Microsoft 365 for email and calendar, but also utilize Claude API or Gemini API for AI assistance.
• Don’t lock your data in cloud storage; always keep it in an exportable format.
• Avoid making core business processes completely reliant on a specific SaaS.

Keeping switching costs low at all times will be your strongest bargaining chip against price increases.

3. 「AIエージェントの棚卸し」を四半期ごとにやる

As we enter the era of agent billing, ‘AI that is running for no particular reason’ will become the biggest source of cost leakage.

With human employees, you would notice if they weren’t working. However, AI agents will continue to incur charges even if they are only used three times a month.

Quarterly, inventory the usage frequency, processing volume, and cost-effectiveness of each agent. Disable agents that do not provide a satisfactory ROI. Just this could help many companies reduce annual AI-related costs by 20-30%.

本当の問いは「AIに払うか、人に払うか」ではない

After reading this far, some might wonder, ‘In the end, is AI going to be expensive?’

No, the framing of the question is incorrect.

The real question is ‘Whose AI will you use?’

If you use Microsoft or Google’s AI, you will be billed according to their rules. Price increases and feature limitations will be determined by their convenience.

However, if you customize an open-source model for your company, you can set the billing rules yourself. A small local business can build AI agents specialized for their operations with a five-person team. This can be operated for 10,000 yen a month, achieving the same or greater operational efficiency than a large company paying 1 million yen, but at less than one-tenth the cost.

This is not a pipe dream; it is already happening.

There are cases where local manufacturers are automatically generating quality inspection reports using Llama 3-based AI. Before implementation, outsourcing costs were 400,000 yen a month, but now it runs for less than 20,000 yen a month. That’s a 95% cost reduction.

The democratization of technology always works in favor of small and medium enterprises. However, this only applies to those companies that do not fall under the billing models of platform providers.

まとめ：3つの数字を覚えておけ

Finally, here are three numbers to remember from this article.

• $30 (approximately 4,500 yen): The current unit price for Copilot licenses. A future where this is multiplied by the number of agents is coming.
• 50%: The increase in Motorola’s prices. A precedent for price hikes under the pretext of adding AI features.
• 10,000 to 20,000 yen per month: The realistic running costs when operating open-source LLMs in-house.

The change in billing models among major SaaS vendors is unstoppable. However, whether you get swept up in that wave depends on your decisions now.

‘First, lay out all the details of your SaaS expenses.’

Let’s start from there.