Conclusion

Let’s get straight to the point. The costs of AI are rising. And the burden will fall on small and medium-sized enterprises (SMEs).

The White House has begun addressing the electricity issues faced by AI data centers. Goldman Sachs has warned of inflation risks stemming from AI. While some API fees from OpenAI have decreased over the past two years, the cost per use for high-performance models has skyrocketed.

These may seem like separate news items, but they are all interconnected.

Rising electricity costs → Increased cloud and API production costs → Passed on to consumers → Higher monthly costs for SMEs.

The question is not whether costs will rise, but rather “when and by how much?” When that wave hits, the divide between companies that are prepared and those that are not will become stark.

AI Data Center Electricity Consumption Reaches “Another Country” Level

The White House has intensified discussions with utility companies and data center operators in late 2024, driven by a simple statistic.

According to estimates from the International Energy Agency (IEA), global electricity consumption by data centers could rise from approximately 460 TWh in 2022 to over 1,000 TWh by 2026. This 1,000 TWh is nearly equivalent to Japan’s total annual electricity generation (around 1,000 TWh). We are entering an era where AI learning and inference consume the equivalent of an entire country’s electricity.

Even within the United States, multiple forecasts suggest that the electricity demand from AI data centers will triple or quadruple by 2028. The problem is that supply cannot keep up. Building new power plants takes 5 to 10 years, and significant investments are required to enhance the power grid. When demand exceeds supply, electricity prices will rise. This is basic economics.

In fact, in Northern Virginia (the world’s largest data center hub), the utility company Dominion Energy is gradually raising industrial electricity rates. In Texas, the political issue of balancing data center attraction with rising electricity prices has begun to surface.

This rise in electricity costs will directly impact the cost structures of major cloud providers like AWS, Azure, and Google Cloud. And these cloud giants are not running a charity. When costs rise, they will pass those increases onto consumers.

Goldman’s Warning—The Risk of AI Becoming an “Inflation Device” Instead of a “Deflation Device”

Goldman Sachs’ research team has pointed out that the surge in AI-related investments could lead to inflationary pressures in the medium term.

There are three key points:

1. Rising Hardware Costs. The prices of NVIDIA GPUs (H100/B200) remain high due to excessive demand. The price of high-bandwidth memory (HBM), essential for AI learning, is expected to continue rising even after 2025, with some forecasts predicting increases of 20-30% by 2026.

2. Structural Increase in Electricity Costs. As mentioned earlier, this will push up prices across the entire cloud service sector, not just AI.

3. Soaring Labor Costs. AI engineers in the U.S. earn an average of $200,000 to $400,000 annually. These labor costs will be reflected in service prices.

In other words, AI is not a “magical tool that gets cheaper the more you use it.” When infrastructure costs rise, the burden on users also increases. The productivity gains and cost reductions from AI can be offset by rising AI infrastructure costs—at times, the cost increases may prevail.

This is the reality of “AI inflation.”

What Will Happen to SMEs’ API Costs?—Estimating with Three Scenarios

So, what will happen to the API and cloud costs that local SMEs pay each month?

At present, let’s assume that the typical monthly cost for SMEs incorporating AI into their operations is between 30,000 to 100,000 yen. This is based on their experiences with using ChatGPT API for customer inquiries, image generation, meeting minutes summarization, and data analysis.

Scenario A: Gradual Increase (Annual Increase of 10-15%)

A pattern where cloud companies gradually pass on the rising electricity costs. This is the most likely scenario.

• Current: 50,000 yen/month → 1 year later: 55,000 to 58,000 yen/month → 2 years later: 60,000 to 67,000 yen/month
• An annual increase of 60,000 to 80,000 yen. This is the kind of situation where one might say, “I didn’t notice it was getting more expensive.”

While this may not seem significant for a single company, if we consider that not only API fees but also cloud storage and SaaS monthly fees will rise simultaneously, an overall increase of 200,000 to 300,000 yen in IT-related costs annually is quite plausible.

Scenario B: Sudden Spike (Price Revision of Specific Services)

A scenario where OpenAI or Google suddenly revises the prices of high-performance models. In fact, OpenAI significantly changed the token price upon the launch of GPT-4o. There is no guarantee that a price increase won’t happen in the opposite direction.

• Current: 50,000 yen/month → After revision: 80,000 to 100,000 yen/month (60-100% increase)
• A sudden doubling could be fatal for SMEs with limited monthly budgets.

Scenario C: Switching to Cheaper Models (Cost Optimization Works)

A scenario where improvements in open-source models (like Llama and Mistral) enable businesses to operate without using expensive APIs.

• Current: 50,000 yen/month → After optimization: 10,000 to 20,000 yen/month
• However, this requires a level of technical literacy to determine “which model is sufficient.”

Realistically, a combination of A and C will become the survival strategy for SMEs. While accepting price increases from major APIs, they will switch to open-source or lightweight models where possible. The ability to make this judgment will create cost differences.

So, What Should We Do?

“AI inflation” is not a distant issue for SMEs. However, there are ways to address it.

1. Take stock of current API usage.
Are you aware of what you are paying for each month? Surprisingly, many SMEs have accumulated SaaS contracts that they have signed without much thought. Start by laying out all current invoices. That is the starting point.

2. Reassess whether “GPT-4 is really necessary for this task.”
If a process that costs 50,000 yen with GPT-4o can achieve sufficient quality with GPT-4o-mini or open-source models, costs could drop to 1/5 to 1/10. In fact, some companies that have tried it found that 80% of customer inquiries could be adequately handled by GPT-4o-mini.

3. Negotiate annual contracts and volume discounts.
Continuing to use services under monthly billing without any thought is the most expensive option. If usage is predictable, aim for a 10-20% discount through an annual contract. The same logic applies to reserved instances from cloud providers.

4. Consider the option of “not using AI.”
It is not necessary to do everything with AI. If AI costs are rising, reevaluate the line between “what can be left to AI” and “what should be done by humans.” When the API cost rises from 30,000 yen to 80,000 yen for automated tasks, there may be instances where it is cheaper to hire a part-time worker. Cost comparisons should always be updated.

5. Stay informed.
Changes in API fees, cloud pricing, and electricity cost trends can be announced suddenly. Ignorance is not an excuse. Establish a system to regularly check for updates on pricing changes for major services.

AI Inflation Arrives as “Invisible Price Increases”

Even if electricity costs rise, small business owners will not receive notifications saying, “Electricity prices have gone up.” Instead, they will receive notices of “API price revisions” and “cloud usage invoices.”

When the increase is a few thousand to tens of thousands of yen per month, it may be brushed off with a “well, it can’t be helped.” However, when these increases accumulate over the course of a year and happen simultaneously across multiple services, IT-related costs could unexpectedly become 1.5 times higher—this is a structurally visible future.

Large corporations can absorb these costs through economies of scale. SMEs cannot do the same. That is why it is essential to understand both “how to use AI” and “the cost structure of AI” now.

Technological evolution will not stop. Cost increases will not stop either. Continuously choosing the optimal use for your company while keeping an eye on both will be the survival strategy for SMEs in the era of AI inflation.