AI’s ‘Cost’ is No Longer Determined by OpenAI

In the past week, three news items reveal a significant structural change:

• OpenAI proposed to give 5% of its equity to the U.S. government.
• Meta is preparing to sell AI computing resources through its cloud business.
• Microsoft invested $2.5 billion (approximately ¥375 billion) in a new company focused on AI deployment.

While these three developments may seem unrelated, they share a common theme: the power to determine the ‘price of AI’ is beginning to shift away from model developers.

This is a monumental development for small and medium-sized enterprises (SMEs). Why? Because the very structure of how monthly API fees are determined is changing.

—

5% Equity Equals $42.6 Billion: What OpenAI ‘Offered’ the Government

OpenAI CEO Sam Altman proposed giving the U.S. government 5% of the company’s equity. In the most recent funding round, OpenAI was valued at $852 billion (approximately ¥128 trillion). Thus, 5% amounts to about $42.6 billion, or approximately ¥6.4 trillion.

Why would they do this? On the surface, it appears to be a noble gesture of sharing the benefits of AI with the public. However, the reality is much more pragmatic.

OpenAI is currently transitioning into a for-profit entity. Originally established as a non-profit organization, it is now aiming to compete in a market worth hundreds of trillions of yen. What is the biggest risk in this transition? Regulation by the government.

By offering 5% of its equity, OpenAI is effectively making the government a ‘shareholder.’ Shareholders are not regulators; they are stakeholders in profit-sharing. This can be interpreted as a transaction to buy off regulatory risk with equity.

The essence of what is happening here is that the government will directly participate in determining the rules for AI development—specifically, what can be created and how it can be used.

What does this mean for SMEs? In the short term, the impact may not be immediately visible. However, thinking in the medium to long term, if the government becomes a shareholder in AI companies, it will also have a say in AI pricing policies. The future where ‘AI usage fees’ are determined politically, much like public utilities, is no longer a mere fantasy.

—

Meta Aiming to Become the ‘Electric Company’ of AI

Meta’s move is more direct.

Until now, Meta has made its AI models (the Llama series) available as open source. The models are free, but running them requires substantial computing resources, which can only be purchased from AWS, Google Cloud, or Microsoft Azure.

In other words, while the models are free, the cost of running them has been paid to other companies.

Meta is looking to change this. It plans to sell the computing power of its vast data centers to external clients—essentially entering the ‘AI cloud infrastructure’ business.

What does this mean? The cost structure of AI is about to change.

Currently, the majority of the fees for using AI tools are determined by the cost of computing resources. The API fees for OpenAI are largely based on the computing costs from Microsoft Azure. With Meta entering this computing resource market as a major player, competition will increase, leading to lower prices. It’s a straightforward equation.

As of now, the API fee for GPT-4o is $2.50 per million input tokens and $10.00 for output. Compared to the GPT-4 Turbo from a year ago, prices have already dropped to a fraction of what they were. If Meta initiates a price competition in computing resources, further decreases in API fees are guaranteed.

For SMEs, this is nothing but good news. The cost of utilizing AI, which was tens of thousands of yen per month until last year, could drop to several thousand yen or even a few hundred yen. That world is now closer than ever.

However, there is a caveat. Meta is an advertising company. In exchange for providing computing resources at a low cost, there is a significant possibility that it will leverage usage data for its advertising business. There is always a reason for ‘cheap.’ Users will need to assess whether they are comfortable with that reason before deciding to use the service.

—

Microsoft’s $2.5 Billion New Company—Separating ‘AI Creation’ from ‘AI Delivery’

Microsoft’s investment of $2.5 billion (approximately ¥375 billion) in the newly established AI deployment company exceeds the market capitalization of many regional banks in Japan.

Why did Microsoft choose to create a separate company instead of handling it internally?

This indicates a structural change: the processes of ‘creating’ AI and ‘delivering’ AI are beginning to emerge as separate businesses.

Previously, OpenAI created the models, Microsoft delivered them via Azure, and companies used them through APIs—a simple flow. However, the performance differences among models are narrowing. GPT-4o, Claude, Gemini, Llama—there is little difference in performance for many business tasks regardless of which one is used.

If performance differences among models are diminishing, what will differentiate them? ‘Delivery methods.’ How, in what form, and at what cost AI is delivered to various industries will become the key differentiators. Microsoft has created a separate company to specialize in that.

This means two things for SMEs:

First: The options for AI tools will explode. As more companies specializing in ‘delivery’ emerge, industry-specific and region-specific AI services will become available. For manufacturing, retail, construction, etc., the hassle of customizing general-purpose tools will decrease.

Second: The business of ‘delivering AI’ will present opportunities for SMEs. Creating a model may require hundreds of billions of yen, but establishing a system to deliver a model to a specific industry can be done for a few million yen. Local SMEs could become ‘regional AI deployment companies.’

—

Organizing Structural Changes—What SMEs Should Keep an Eye On

Let’s summarize the three movements from the perspective of SMEs:

The common thread among these three developments is that the power to determine the price of AI is shifting from model developers to infrastructure providers and the government.

This is reminiscent of the history of electricity. Initially, those who invented power generation technologies set the prices. However, over time, companies with transmission networks and regulatory governments gained control over pricing. AI is following a similar path.

—

So, What Should SMEs Do?

Three things:

1. Consider Current API Fees as ‘High.’
In six months to a year, they are almost certain to decrease. If you base your ROI calculations on current fees, you may make incorrect investment decisions. Think about whether it would still be viable if the fees were halved.

2. Design Systems That Are Not Dependent on Specific Models.
Building business systems solely around GPT is risky. Ensure that the design allows for interchangeable models. Right now, it may be OpenAI, in six months it could be Claude, and in a year it might be Llama. Minimizing switching costs will enhance your negotiating power.

3. Consider Becoming a ‘Delivery Side’ Business.
Your industry knowledge combined with AI tools can create services that can be sold to peers. There is no need to create a model. The ones who know the challenges in their industry best are not large corporations but the SMEs on the ground. Combine that knowledge with AI tools to package a solution. This is the only structure where local SMEs can compete with large corporations.

—

Transforming the Shift of Power into Opportunity

The power to determine the price of AI is shifting. While this may seem like a threat, it also means that the price of AI will continue to decrease due to political and infrastructure competition.

For SMEs, a decrease in technology costs is always a tailwind. When the costs of building websites and managing social media decreased, it was the agile SMEs that maximized the benefits.

The same applies here. A structural change is occurring that will lower the cost of AI. Instead of waiting, experiment on a small scale now. Prepare to scale up rapidly when prices drop by half.

While large corporations require board approval to respond to shifts in power, SMEs can start moving as early as next week. That speed is the only and greatest weapon they have.