NASDAQ Loses $2 Trillion — How Will AI Costs for SMEs Change Amid the ‘AI Bubble Burst’?

Approximately $2 trillion vanished from the NASDAQ in a single day, marking a 5.5% drop. This is the most significant crash since the Trump tariff shock in April 2025.

The cause is concerns over increased regulations on AI data centers. Hedge funds have collectively sold off AI-related stocks, and the term “AI bubble burst” is circulating widely in the media.

However, this raises an important question: When a bubble bursts, what prices fall? And who stands to benefit?

To conclude, this could be the “biggest window for SMEs to start using AI at lower costs.”

—

What’s Happening — $720 Billion in Investments at Stake

Amazon, Microsoft, Alphabet, and Meta. These four companies alone are projected to invest over $720 billion in AI infrastructure by 2025, equivalent to about 108 trillion yen. This is a level of investment comparable to a national budget being poured into data centers and GPUs.

However, the market is starting to ask, “So, what are the returns?”

Microsoft has postponed some of its data center construction plans. Meta’s AI contributions to advertising revenue have also fallen short of analyst expectations. As voices expressing concerns over the “lack of recovery prospects” spread, hedge funds have begun to offload their AI-related positions.

This is not just a stock price issue; it marks a turning point in the very structure of AI service supply.

—

Three Price Changes SMEs Should Watch

When a bubble bursts, the excess supply that has built up loses its outlets. As a result, what becomes cheaper? Let’s take a closer look.

#### 1. LLM API Costs — Price Destruction Has Already Begun

In fact, the decline in API costs has been ongoing without waiting for the bubble to burst.

As of 2023, OpenAI’s GPT-4 API was priced at about $30 per 1M tokens. By 2025, GPT-4o is expected to be just $2.50 under the same conditions, a reduction of approximately 1/12 in just two years. Google’s Gemini 2.5 Flash has even reached $0.15 per 1M tokens, which is 1/200 of the initial price of GPT-4.

Moreover, the entry of Chinese companies is accelerating price competition. DeepSeek’s API is priced at a fraction of that of U.S. competitors for similar performance.

What happens when the bubble bursts? Major cloud vendors will start to sell off excess capacity at discounted rates to recover their investments. Already, spot prices for GPU instances on AWS and Azure are trending downward. It’s only a matter of time before this affects API pricing further.

Implications for SMEs: With monthly API costs of just tens of thousands of yen, tasks that once cost hundreds of thousands of yen per month can now be handled. The bubble burst will further accelerate this trend. The perception that “AI is too expensive to consider” needs to be updated.

#### 2. GPUs and Computational Resources — The Used Market and Excess Capacity

NVIDIA’s H100 was initially traded at about $40,000 (approximately 6 million yen) each. It was difficult to obtain due to high demand.

However, if investments in data centers cool down, the situation will change. GPUs will be released into the market from postponed or canceled projects. Reports indicate that prices for used H100s have already begun to drop to the $20,000 range in some markets.

That said, whether SMEs need to own GPU servers is another matter. The important point is the structural change that leads to lower unit costs for computational resources. Prices for cloud GPU instances and pay-as-you-go inference APIs will all decrease in tandem.

Implications for SMEs: It’s not about buying GPUs for in-house use. It means that “the cost of using AI via the cloud is decreasing.” The range of what can be operated with a budget of 50,000 yen per month may double in six months.

#### 3. AI Talent — The Exodus from Big Tech Begins

This is perhaps the most overlooked point.

Big tech companies have rapidly expanded their AI departments over the past two years. However, when investments cool down, the first to be cut will be personnel from projects that “are not yet directly tied to revenue.”

In fact, reports of layoffs at Google, Meta, and Amazon have begun to surface in late 2024. AI-related engineers and data scientists are starting to enter the market.

That said, this doesn’t mean local SMEs will hire engineers from Silicon Valley. What’s noteworthy is the indirect effects.

• The rates for freelance AI engineers will decrease.
• Access to AI talent through side jobs and contracts will become easier.
• The contract prices for AI development companies and system integrators will drop.

If talent that was previously secured by large companies at salaries of 20 million yen returns to the market, SMEs may have the opportunity to collaborate with skilled individuals on contracts ranging from 300,000 to 500,000 yen per month.

Implications for SMEs: There’s no need to insist on hiring full-time employees. An environment is being established where SMEs can work with levels of talent that were previously out of reach through contracts and spot agreements.

—

Don’t Wait for the ‘Bubble Burst’ — What to Do Now

After reading this, some may think, “Well, if I wait a little longer, it might get even cheaper.”

That line of thinking is dangerous.

Why? The decline in AI prices is not due to the bubble bursting, but rather because technological advancements and competition are structurally driving costs down. The collapse of the bubble will only temporarily accelerate this trend, making a strategy of “waiting for the bottom” meaningless.

What’s more important is to start small experiments at the current cost levels.

What specifically should you do?

1. Identify repetitive, large-scale, and simple judgment tasks in your business. Tasks like assisting in creating estimates, classifying inquiry emails, and summarizing daily reports are precisely where AI excels.

1. Try hitting APIs with a budget of 10,000 to 30,000 yen per month. ChatGPT API, Claude API, Gemini API. All of these are pay-as-you-go and can be started small. Even if you fail, it’s just a few tens of thousands of yen in tuition.

1. Experiment to determine the boundary between “work that should be done by humans” and “work that can be delegated to AI.” The right answer varies by industry and company. You won’t know until you try.

—

The Real Question Is Not Whether It’s a Bubble

Regardless of whether the NASDAQ crashes or recovers, there’s one thing that SME owners should consider.

“In a world where AI costs have dramatically decreased, will we change the way we work or not?”

Two years ago, using the GPT-4 API for business required a budget of several hundred thousand yen per month. Now, equivalent or better processing can be done for just a few thousand to tens of thousands of yen per month. Will you continue to say this change is “still irrelevant”?

The bursting of the bubble will bring AI closer to being “an infrastructure that anyone can use” rather than just “the domain of a few large companies.” Just as electricity and the internet did.

As large corporations tighten their investments, stock prices fall, and the media clamors about an “AI winter,” it will be the SMEs quietly hitting APIs and systematizing their operations that emerge as the next winners.

Instead of watching stock price charts, it’s time to reassess your business flow. The answers lie there.