Conclusion

Let’s get straight to the point: “The cost of AI computation has begun to change structurally.”

You can’t buy a Mac mini.

Tim Cook of Apple himself acknowledged during the earnings call in May 2025 that “the demand for AI is rising more rapidly than expected.” The Mac mini and Mac Studio are being purchased by companies as platforms to run AI agents locally.

“There is such a demand for AI that Apple is experiencing shortages”—this might sound like a story about large corporations. However, that’s not the essence of the matter.

The real question is: “Which is cheaper, continuously paying for cloud APIs or having a Mac mini on hand?”

For small and medium-sized enterprises with 10 to 50 employees, the answer to this question is starting to reverse. This is the true meaning behind the shortage of Mac minis.

—

Facing the Reality of “How Much Are We Paying for Cloud APIs Every Month?”

First, let’s look at the actual numbers.

When small and medium-sized enterprises use cloud APIs like ChatGPT, Claude, or Gemini for their operations, the typical cost structure looks like this:

• ChatGPT Team Plan: Approximately $25 (around 3,750 yen) per person per month. For 30 people, that’s about 112,500 yen per month, or 1.35 million yen annually.
• Using API on a Pay-As-You-Go Basis: Depending on usage, for tasks like creating meeting minutes, drafting emails, or handling internal FAQs, costs for a team of 30 typically range from 30,000 to 80,000 yen per month.
• Annual Costs: 360,000 to 960,000 yen; Over 3 Years: 1,080,000 to 2,880,000 yen.

You might think, “Well, that’s about right.” But what you need to pay attention to is that this is an ongoing charge that will never stop. As long as you continue to use it, you will keep paying. Moreover, there’s a risk that API prices will increase in response to rising demand.

—

How Much Would It Cost to Do the Same Thing with Mac mini and Local LLM?

Now, let’s calculate the costs if you were to implement Mac mini as a local AI server.

It’s important to clarify a key assumption here: You don’t need to buy 30 Mac minis for a company of 30 people. For a local LLM server, 1 to 3 units are sufficient. They would be placed on the internal network, allowing everyone to access them via a browser.

Hardware Costs

Running Costs (Monthly)

Total Cost Comparison Over 3 Years

The difference is as much as 3 million yen or more. For small and medium-sized enterprises, 3 million yen is the amount needed to hire one additional employee.

—

Answering the Question: “But Isn’t the Performance of Local LLMs Low?”

This is where things have changed dramatically compared to a year ago.

Until 2024, locally running open-source LLMs were honestly close to being “toys.” However, as of 2025, the situation has completely transformed.

• Open models like Llama 3.1 70B, Qwen2.5 72B, and Mistral Large are now available for free with performance comparable to GPT-4.
• With the Mac mini’s M4 Pro chip (48GB unified memory), 70B parameter models can run at practical speeds.
• Tools like Ollama allow installation and operation in less than 30 minutes.

Summarizing meeting minutes, drafting emails, searching internal manuals, and checking estimates—80% of the AI tasks used in the daily operations of small and medium-sized enterprises can now be sufficiently handled by local LLMs.

For the remaining 20%, such as image generation or research using the latest information, you can use cloud APIs in conjunction. There’s no need to do everything locally. A hybrid structure of “80% local, 20% cloud” becomes the optimal cost solution.

—

The “Real Benefits” for Small and Medium-Sized Enterprises Go Beyond Cost

Cost reduction is just the entry point. The structural changes that local AI brings to small and medium-sized enterprises can be summarized in three points.

1. Data Doesn’t Leave the Company

As long as you use cloud APIs, customer information and internal know-how are all sent to external servers. The concerns of business owners about “Is our data safe?” are valid. With local solutions, data never leaves your company’s Mac mini.

For regional small and medium-sized enterprises, trust with customers is paramount. The value of being able to say, “We do not send your data outside” cannot be quantified in monetary terms.

2. No Downtime

Cloud services can experience outages. In 2024 alone, OpenAI’s API faced multiple outages. When an outage occurs, business operations come to a halt. With local solutions, as long as your internal network is operational, you can keep running.

3. Cultivating “AI Talent” Internally

This is the biggest advantage. If you only use cloud APIs, employees will only become “users of AI services.” However, by operating local LLMs in-house, experiences such as “choosing models,” “designing prompts,” and “tuning for business needs” will accumulate internally.

For small and medium-sized enterprises to differentiate themselves with AI, having “people who understand their business and can work with AI” in-house is critically important. This becomes a competitive advantage that cannot be obtained through outsourcing.

—

Take the Claims of “Brain-like Chips Reducing Energy Consumption” with a Grain of Salt

The original news source also mentions brain-like chips (neuromorphic chips). While research is indeed progressing, it will be at least 3 to 5 years before these technologies reach the field of small and medium-sized enterprises.

There’s no need to factor in “it might become cheaper in the future” into current decision-making. At this very moment, you can create an AI environment with two Mac minis and free open-source LLMs at less than one-fifth the cost of cloud APIs. This fact alone provides ample reason to act.

—

So, What Should You Do?

Step 1: Buy one unit and try it out.

The Mac mini M4 Pro (48GB) costs about 220,000 yen. Install Ollama and load Llama 3.1 or Qwen2.5. This takes about an hour.

Step 2: Use it for “one specific task” within the company.

It could be summarizing meeting minutes or automatically checking daily reports. Just run one task for two weeks.

Step 3: Measure the costs and effects.

Compare it with the monthly costs of cloud APIs. If there are no quality issues, bring in a second unit and roll it out to the entire team.

The important thing is to “just start using it.” The 3 million yen difference has already accumulated while holding three discussion meetings.

—

What the Shortage of Mac minis Indicates

The reason for the shortage of Mac minis is not due to Apple’s supply issues. It’s because companies have realized that “having a computer on hand is cheaper than continuously paying for the cloud.”

This trend is irreversible. The performance of open-source LLMs is improving every month, while hardware prices are decreasing every year. The price advantage of cloud APIs will continue to structurally diminish.

This is good news for regional small and medium-sized enterprises. Even without the budget for cloud services running into the hundreds of thousands of yen per month like large corporations, having a 220,000 yen Mac mini allows them to compete on the same level.

The notion that “AI is expensive” is coming to an end. That turning point is now.