JPY 500,000–1 million per month? Reading a Japanese business AI cost survey
What a LayerX survey reported by Nikkei says about spending levels and a possible decision point
Key takeaway
The high spending discussed earlier is not confined to the best-known cases. In a Japanese survey published by LayerX in June 2026, the largest group, 26.3%, reported monthly AI spending of JPY 500,000 to under JPY 1 million. 9.3% reported JPY 10 million or more, and over half expected costs to become a management issue within one year. For these respondents, AI spending is becoming a leadership question, not simply an IT expense.
AI cost is moving from an IT expense to a management issue
26.3%Share reporting JPY 500,000 to under JPY 1 million in monthly AI costs, the largest band
JPY 500,000–1 million per month annualizes to JPY 6–12 million, comparable to a full-time employee's total annual cost in some Japanese roles. This puts the salary-scale discussion in article 18 into context, without making the survey representative of every business.
What the survey distribution shows
LayerX surveyed 400 people who managed or understood their company's AI usage costs online on June 12–13, 2026. The results were reported by Nikkei and others.
About four in ten respondents reported at least JPY 1 million monthly, and roughly one in ten at least JPY 10 million. The latter annualizes to more than JPY 100 million. These figures show that large bills exist beyond a single anecdote, but they should not be treated as a population-wide rate for Japanese companies.
Costs are rising for many respondents, with agents cited as a factor
The direction of change is another important part of the survey.
- 46.5% reported costs had risen somewhat and 20.0% that they had risen substantially: 66.5%, about two thirds, in total. 23.8% reported little change.
- The spread of AI agents that carry out work autonomously was cited as a major factor in increased spending.
- 54.3% expected AI costs to become a management issue within one year, 19.8% within two to three years, and 19.0% said they already were: over 90% combined.
Remember who was surveyed: people who already managed or understood AI costs. The sample therefore reflects organizations with a degree of AI adoption, and may show higher costs than Japanese businesses overall. It can provide a useful reference for growing adoption, but not a national average.
Our interpretation is to compare this distribution with the demand scenarios in article 21. If spending continues to rise, some respondents may move into higher bands. The survey alone does not establish their growth rates or predict how many will move next year. When comparing your own position, examine both the current band and your measured trajectory over the coming year.
The decision point between metered spending and a fixed investment
Spending of JPY 500,000–1 million a month can be an important point to evaluate alternatives, because the payback period for purchased on-premises AI may become practical for suitable workloads.
Consider a simplified example: JPY 750,000 a month, the midpoint of that cost band, totals JPY 18 million over two years if it stays constant. The survey reports many respondents' costs rising, but does not predict this particular example. Purchased local inference instead starts with a known hardware investment, plus electricity, maintenance, and other operating costs. Compare both scenarios over a realistic period.
With an upfront-purchase system such as Sovereign GaiXer, expanding suitable local agent workloads does not add per-token charges. That can reduce exposure to a rising usage bill, though capacity expansion and operating costs remain. See article 3 and article 11 for estimation approaches.
Think of meters starting to turn across the company while future usage and tariff changes make the next bill difficult to forecast. The survey suggests many responsible staff are beginning to see that uncertainty as a management issue. It does not imply every provider changes prices every month.
Frequently asked questions
We spend under JPY 100,000 per month. Is this relevant yet?
That band represented 6.3% of this selected sample. Low current spending does not by itself establish safety or risk. Reviewing the cost structure while usage is still small can be easier than doing so after agent adoption accelerates.
How reliable is this survey?
It is a survey by a business, LayerX, of 400 people who understood their organizations' AI costs. It is not a representative census of all Japanese companies. Read it as evidence about the surveyed group, alongside its methodology and other sources.
Why did 10.5% answer that they did not know?
One possible explanation is fragmented purchasing: different departments subscribe separately, or employees expense personal subscriptions. The survey figure alone does not prove the cause. Consolidating costs is a useful starting point for token management, discussed in article 20.
Are the respondents spending JPY 10 million or more all large companies?
The article does not establish their company sizes. Large organizations may account for much of the spending, but the one-employee case in article 16 illustrates how workload intensity can also matter independently of headcount.
What should we do first as a management issue?
Establish total monthly spending and its breakdown, then its rate of change. Compare improving controls on metered usage with moving suitable workloads to a fixed hardware investment.
Does stopping metered local inference remove every concern?
No. It reduces local token-bill unpredictability, but electricity, maintenance, expansion, external services, and outcome measurement remain. The nature of the planning changes toward ordinary equipment investment and operations.
Summary
- In LayerX's June 2026 sample, 26.3% reported monthly AI spending of JPY 500,000 to under JPY 1 million, and 9.3% reported JPY 10 million or more.
- 66.5% reported year-on-year increases; agent adoption was cited as a major factor.
- Over 90% combined saw costs as already a management issue or expected them to become one within one to three years.
- JPY 500,000–1 million monthly is a staffing-scale amount in many Japanese contexts and a useful point to evaluate on-premises payback.
- Purchased local inference can reduce exposure to rising token bills, while full operating costs and capacity still need planning.
Related articles
The issue reached Japan too One employee, JPY 10 million in monthly AI usage
The US cases were not a distant problem. Lessons from a domestic case reported by Nikkei
What would it cost us?
Estimate costs using your own workload
Switching to lower-cost Chinese AI? Consider on-premises deployment too
The impact of a 1% price comparison, the structural issues that remain, and a third option
This article interprets published survey results and reporting. See LayerX's release for the methodology and detailed figures. Rounded percentages may not total 100%; chart bar lengths are approximate visualizations. Please contact us with factual corrections. The cited media, researchers, and companies do not endorse Sovereign GaiXer.
Company, product, and service names mentioned are trademarks or registered trademarks of their respective owners.