5 Billion for Anthropic, 60 Billion for Cursor — The ‘Flow of Money’ in AI Has Changed. Which Layer Should SMEs Ride On?

5 billion for Anthropic. 60 billion for Cursor. 40 million for NeoCognition.

When these three figures are lined up, a certain structure emerges. The ‘flow of money’ in AI has clearly changed.

It’s not the companies creating models that are receiving astronomical valuations, but rather companies that use models to get work done.

For small and medium-sized enterprises (SMEs), this is not a distant issue. The choice of ‘which layer to ride on’ will significantly affect future return on investment.

—

What Happened — Reading Three News Stories Through Structure

First, let’s organize the facts.

1. Amazon Adds 5 Billion to Its Investment in Anthropic
Anthropic is an AI model company that develops Claude. With this investment, it will procure up to 5 gigawatts of AI chips from Amazon, strengthening its model training and operational infrastructure. While the total investment amount is enormous, this is an investment in the ‘model layer’ — the domain of creating the brain of AI itself.

2. Reports Indicate SpaceX is Considering Acquiring Cursor for Approximately 60 Billion
Cursor is an AI-powered automatic programming tool that dramatically increases the speed at which engineers write code. Notably, the 60 billion valuation is placed not on a company creating models, but on a ‘tool that enhances developer productivity using models.’ This reflects the valuation of the ‘tool layer.’

3. NeoCognition Raises 40 Million
This company develops AI agents that learn autonomously like humans and perform specific tasks. Although the amount is small compared to the previous two, it signals that funding is beginning to flow into the ‘agent layer’ — the domain where AI makes decisions and executes tasks on behalf of humans.

—

Reading the Structure: Money is Flowing from ‘Upstream’ to ‘Downstream’

When dividing the AI industry into three layers, it looks like this:

The model layer is akin to a ‘power company.’ It requires massive capital and computational resources to enter. As symbolized by Anthropic’s 5 billion, this is a battlefield for giants like Amazon, Google, and Microsoft, leaving little room for SMEs.

The tool layer is closer to ‘appliance manufacturers.’ They create products that make human life convenient using electricity. The 60 billion valuation for Cursor is evidence that the market is beginning to judge that ‘what can be done using models is more valuable than the models themselves.’

And the agent layer is like a ‘housekeeping service.’ It not only uses appliances but finishes household tasks on behalf of humans. The investment in NeoCognition indicates that this area is finally entering a practical phase.

The flow of money is clearly shifting from ‘upstream (models)’ to ‘downstream (tools → agents).’

This is actually good news for SMEs.

—

Why This is Good News for SMEs

The reason is simple. As the model layer becomes commoditized, the prices of downstream tools and agents will decrease.

Two years ago, using a GPT-4 class model for business could cost tens of thousands of yen per month just for API usage. How about now? As a result of competition among Claude, GPT-4o, and Gemini, comparable performance can now be accessed via subscriptions costing 20,000 to 30,000 yen per month. The unit price of APIs has also halved within a year.

As models become cheaper, the tools built on top of them also become cheaper. Cursor can be used for just 20 dollars (about 3,000 yen) per month. If an engineer’s coding speed can be increased by 2 to 3 times for just 3,000 yen, the return on investment is extraordinarily high.

In other words, SMEs do not need to ‘create models.’ They only need to think about how to incorporate the tools and agents that ride on top of the models into their operations.

This is where the competitive edge lies.

—

Three Steps SMEs Should Take Starting Today

Step 1: Invest 10,000 to 30,000 yen in the Tool Layer (Right Now)

First, introduce AI tools that are already at a practical stage into your operations.

• Claude Pro / ChatGPT Plus (about 3,000 yen per month): Summarizing meeting minutes, drafting emails, streamlining research tasks
• Cursor (about 3,000 yen per month): Developing internal tools, modifying existing systems
• Notion AI / Google Gemini (from about 2,000 yen per month): Document management, information organization

Within a budget of 10,000 to 30,000 yen, start by transforming the work of one person to be ‘AI-centric.’ The key here is not to implement it company-wide but to experiment with 1 person × 1 task.

For example, if a sales administrative staff member can reduce the time spent creating estimates and responding to standard emails from two hours a day to 30 minutes using Claude Pro, that results in a saving of 40 hours a month. That’s a labor cost saving of 60,000 to 80,000 yen at an hourly rate. For an investment of just 3,000 yen a month.

Step 2: Small Trials in the Agent Layer (Within 3 to 6 Months)

Once you feel the effectiveness of the tool layer, the next step is the agent layer.

An agent is a system where ‘once instructed, the AI thinks for itself and executes multiple steps.’

Here are some specific examples:

• Initial Customer Inquiry Response: An AI agent assesses inquiries received via email or chat, automatically replies if it’s a standard response, or escalates complex cases to a human.
• Automating Accounting Tasks: Reading invoices in PDF format and automatically entering them into accounting software, with AI performing reconciliation checks.
• Recruitment Screening: Reading application documents and automatically listing candidates that meet the criteria.

While the pricing for specialized agents like NeoCognition is still unclear, even now, using no-code/low-code agent-building tools like n8n, Make, and Dify, you can create simple agents for a budget of several thousand to tens of thousands of yen per month.

With a budget of 50,000 to 100,000 yen, can you get one business process to operate ‘without human involvement’? This will be the next experimental theme.

Step 3: Systematize with a Combination of ‘Tools + Agents’ (6 Months to 1 Year)

Combining the tool layer and the agent layer accelerates the ‘systematization’ of operations.

For example, the flow could look like this:

1. An inquiry email arrives from a customer (trigger)
2. The AI agent categorizes the content and searches past response history (agent layer)
3. Claude generates a response (tool layer)
4. A human checks and sends it with one click (human judgment)
5. The response history is automatically recorded in the CRM (agent layer)

The cost of building this entire flow, including tool usage fees and the labor for agent construction, can be achieved for around 100,000 to 200,000 yen per month.

Traditionally, trying to do the same thing manually would require the cost of one dedicated staff member — around 3 to 4 million yen annually. Now, it can be done for 1.2 to 2.4 million yen per year. Moreover, it provides 24/7 support without becoming dependent on specific individuals.

300,000 yen becomes 120,000 yen. And the quality stabilizes. This is why SMEs should ride on the tool and agent layers of AI.

—

Clearly Define Layers Not to Enter

Conversely, the layer SMEs should avoid is the ‘model layer.’

There may be proposals from vendors to create a custom AI model from scratch or fine-tune a proprietary LLM, with initial costs of 3 million yen and monthly fees of 500,000 yen.

I can assert that there are almost no cases where this is necessary for companies with fewer than 100 employees.

It makes no sense for SMEs to attempt to replicate what Anthropic is doing with 5 billion for just a few hundred thousand yen. Models are meant to be ‘used,’ not ‘created.’ At least not for today’s SMEs.

Don’t be misled by the term ‘custom AI.’ Simply calling existing models via API and combining them with your business data is sufficient to perform ‘custom’ work.

—

So, What Should You Do?

To summarize:

1. Do not enter the model layer. This is a battlefield for giants. It’s not a place for SMEs to compete.
2. Invest 10,000 to 30,000 yen in the tool layer starting today. Claude, Cursor, Notion AI. Change the work of one person first.
3. Try the agent layer on a small scale within 3 to 6 months. With a budget of 50,000 to 100,000 yen, get one business process to be in a state of ‘running on its own.’
4. Systematize within 6 months to 1 year. Combine tools and agents to eliminate dependency on individuals and create reproducible business flows.

The shift of the flow of money from ‘models’ to ‘tools’ and ‘agents’ means that investment is finally beginning to gather in areas where SMEs can benefit.

The cost of models continues to decrease. Tools can be used for 3,000 yen per month. The cost of building agents is also rapidly declining.

The question is simple. ‘From which business will your company start?’

You don’t need 5 billion dollars to find the answer. Just 3,000 yen and an hour of experimentation will suffice.