Conclusion First

Don’t bet on “AI itself.” Bet on “what becomes cheaper with AI.”

Cursor’s valuation has reached $5 billion (approximately 750 billion yen). The company creates an AI code editor, essentially a “tool for developers to write code.” It is used by developers worldwide through a subscription model priced at $20 per month.

On the other hand, the CEO and CFO of the AI nuclear power startup Fermi Energy have suddenly resigned. This company had a grand vision of meeting the enormous power demands of AI data centers.

When you juxtapose these two news items, a clear picture emerges.

“AI infrastructure” and “AI tools” are entirely different games. And small and medium-sized enterprises (SMEs) should almost certainly invest in the latter.

The Real Reason Cursor is Winning—”Cost Disruption” is Happening

We must not conclude the story of Cursor with just “a great startup has emerged.” The essence lies elsewhere.

Engineer productivity is increasing by 2 to 3 times. This means that development that previously required three people can now be managed by one. With a $20 monthly tool, labor costs could potentially be reduced by hundreds of thousands of yen per month.

In fact, development teams that have adopted Cursor report that “code review time has been halved” and “the time spent writing boilerplate code has become almost zero.” The annual recurring revenue (ARR) is said to have exceeded $300 million, more than ten times what it was just two years ago.

Why has it been able to grow so rapidly? Because its revenue model is clear.

• Low unit price (starting at $20/month) → Decision-making can be completed at the operational level.
• Value can be felt immediately → Users realize “it’s fast” the moment they use it.
• Customers increase organically → Once one person starts using it, it spreads to the entire team.

This follows the same pattern as how Slack and Notion spread. It permeates organizations from the bottom up, and before you know it, it becomes company-wide. SME presidents don’t even need to go through formal approval processes. Engineers on the ground can sign up with a credit card and see productivity improvements the very next day.

“Self-sustaining automation.” This is the strongest form.

Why Fermi is Struggling—”Cost Disruption” is Too Far Away

The concept behind Fermi Energy was logical. The power consumption of AI data centers is increasing explosively. Data centers operated by OpenAI and Google consume as much power as a medium-sized city. The idea was to supply that power using small modular reactors (SMRs).

However, numerous problems were piling up.

1. The distance to monetization is despairingly far. Building a nuclear power plant takes 5 to 10 years just for permits. Construction costs are in the hundreds of billions of yen. During that time, revenue is zero. This is on a completely different timeline than Cursor, which can generate revenue immediately at $20 per month.

2. Dependence on a few customers. The buyers of electricity are limited to major AI data center operators. If Google, Microsoft, or Amazon change their policies, the business plan could collapse entirely. In fact, Microsoft’s contract for the Three Mile Island nuclear plant’s restart plan is currently in turmoil due to cost reviews. A business that relies on just a handful of customers can be mortally wounded by the withdrawal of one.

3. The technical risks are on a different scale. Software bugs can be fixed with updates. Design flaws in a reactor cannot be corrected. Approval from regulatory authorities, meeting safety standards, and gaining community consent—if any one of these is lacking, the project stops. The simultaneous resignation of the CEO and CFO likely reflects these structural difficulties.

In summary, here’s the comparison:

Lessons for SMEs—”Where You Spend Money Will Determine Your Fate”

Now we get to the main point.

The world of AI can be roughly divided into three layers.

1. Infrastructure Layer (semiconductors, power, data centers) → The world of NVIDIA and Fermi
2. Model Layer (foundation models, LLMs) → The world of OpenAI, Google, and Anthropic
3. Application Layer (tools, business software) → The world of Cursor and ChatGPT

SMEs should overwhelmingly invest in the application layer. There are three reasons for this.

① Because it is an area where costs dramatically decrease. Tools like Cursor can be used for just a few thousand yen per month. If some of the coding that was outsourced can be handled internally, it can lead to annual cost savings of hundreds of thousands of yen. Translation, meeting minutes, data analysis—tasks that previously cost 500,000 yen to outsource can now be completed in-house with AI tools costing just a few thousand yen per month. This experience of “300,000 yen becoming 50,000 yen” is happening routinely in the application layer.

② Because failing doesn’t hurt. If a $20 tool doesn’t work out, you can cancel it the next month. This is a different story from investing in a nuclear power plant. The strength of SMEs lies in their speed of decision-making. “Try it first. If it doesn’t work, stop.” This is possible in the application layer.

③ Because it eliminates dependency on individuals. The greatest value of AI tools is transforming tasks that could only be done by capable individuals into tasks that anyone can perform. By training AI with the tacit knowledge of veteran employees, even newcomers can produce outputs of the same quality. AI tools are the ultimate weapon for systematization.

“So, what should we do in the end?”

Here are three specific actions to take.

1. Immediately implement one AI tool costing 2,000 to 3,000 yen per month. ChatGPT Plus ($20/month), Claude Pro ($20/month), Cursor Pro ($20/month). Any of these will do. Have one person start using it. If it proves effective, it will spread organically.

2. List the “outsourced tasks” you are currently doing. Translation, design, coding, data entry, meeting minutes. There will definitely be tasks among these that can be more than 50% replaced by AI tools. Compare your annual outsourcing costs with the monthly tool fees. The difference should be significant.

3. Don’t be swayed by discussions about the infrastructure or model layers. “You should invest in AI semiconductors” or “You should build your own LLM”—these are irrelevant for SMEs. Whether to buy NVIDIA stock is a personal choice, but venturing into the infrastructure layer as a business is suicidal. You could end up like Fermi.

The Structural Changes Indicated by This News

The rapid growth of Cursor and the management turmoil at Fermi did not occur coincidentally. The entire AI industry is experiencing a shift in the “center of value.”

Infrastructure is monopolized by large tech companies. The model layer is converging into competition among a few companies. However, the application layer is infinitely expansive. This is because there are countless needs arising from the combinations of industry, business, and region.

Inventory management for local manufacturing. Property descriptions for local real estate companies. Social media management for individually owned restaurants. These are tasks that neither OpenAI nor Google will undertake. Large corporations also avoid these areas. It is precisely because they are SMEs that they can optimize AI utilization for their specific situations. This is the structure of reversal.

Large corporations take a year to implement a “company-wide unified AI infrastructure.” SMEs can start using it next week. This speed difference is a decisive advantage in today’s era.

Cursor’s valuation of 5 trillion yen signifies the “victory of tools.” Fermi’s management turmoil illustrates the “trap of infrastructure.”

What SME leaders should do is clear.

Don’t get drunk on the dream of infrastructure. Earn with the reality of tools.