The day an AI bill exceeds a human salary
AI spending is moving from a tool budget toward a staffing-scale decision. A new yardstick for reported costs and forecasts
Key takeaway
US reports have described USD 40,000, roughly JPY 6 million, in monthly token costs for one engineer, and USD 1.3 million, roughly JPY 200 million, for a three-person team. A research forecast also suggested AI coding costs could exceed developer salaries by 2028. At that scale, AI spending belongs alongside staffing costs, not merely small software expenses.
AI spending is moving from incidental expense to staffing-scale cost
USD 40,000/monthA reported monthly token cost for one engineer, approximately JPY 6 million
A manager reportedly said they did not know whether to discourage or encourage this consumption. Should an employee delivering results be stopped because of AI costs? The answer requires evidence about value.
Reported figures: AI bills on the scale of payroll
| Reported example | Amount | A labor-cost comparison |
|---|---|---|
| Monthly tokens for one engineer | USD 40,000/month (about JPY 6 million) | An annualized JPY 72 million, comparable to a very highly paid employee |
| Monthly tokens for a three-person development team | Over USD 1.3 million/month (about JPY 200 million) | Comparable to the annual revenue of a small company |
| Typical coding user in article 15 | USD 150–250/month | Still in the range often treated as a tool expense |
| Heavy coding user in article 15 | USD 500–2,000/month | Can approach a part-time worker's monthly pay, depending on location |
Amount
- Monthly tokens for one engineer
- USD 40,000/month (about JPY 6 million)
- Monthly tokens for a three-person development team
- Over USD 1.3 million/month (about JPY 200 million)
- Typical coding user in article 15
- USD 150–250/month
- Heavy coding user in article 15
- USD 500–2,000/month
A labor-cost comparison
- Monthly tokens for one engineer
- An annualized JPY 72 million, comparable to a very highly paid employee
- Monthly tokens for a three-person development team
- Comparable to the annual revenue of a small company
- Typical coding user in article 15
- Still in the range often treated as a tool expense
- Heavy coding user in article 15
- Can approach a part-time worker's monthly pay, depending on location
Reports also included a claim that a company without usage caps spent around USD 500 million on AI in one month. This should be treated as a second-hand claim, not a independently established company figure.
The case for spending more: Jensen Huang's reasoning
Some see high AI use positively. NVIDIA CEO Jensen Huang was reported to suggest that an engineer earning USD 500,000 a year could reasonably use USD 250,000 worth of tokens. The argument is that spending half a salary on AI can pay off if productivity doubles.
A similar approach has been reported in Japan. Accounting software company freee reportedly allocates token budgets by department and compares labor hours with token expenditure to assess ROI. Comparing another employee with more AI capacity puts AI spending on the same decision scale as labor.
Could AI costs exceed salaries in 2028?
A Gartner forecast suggested that AI coding-tool costs could exceed developer salaries by 2028. Two factors underpin the argument.
- Agents multiply consumption. One instruction can trigger hundreds of model calls, as discussed in articles 16 and 17.
- One AI per person can become several agents per person, running in parallel like a team of assistants.
Lower model prices or better efficiency could of course change the forecast; article 21 examines unit prices versus total spending. Still, the direction of the management question is clear: AI costs increasingly deserve attention alongside labor costs.
Thinking of AI as a fixed-cost resource
Do you keep paying AI like an external consultant by the hour, or bring it in as a fixed-cost resource? An upfront-purchase on-premises system such as Sovereign GaiXer follows the latter model.
After the initial hardware purchase, useful work can run around the clock within the machine's capacity without additional local token charges. Electricity, operations, and applicable maintenance remain. Instead of higher use automatically increasing a token bill, more useful work can spread the fixed investment across more tasks. For organizations with several heavy users spending USD 500–2,000 monthly, evaluating payback may be worthwhile. See article 3 and article 11, and use your own workload and total costs.
Imagine AI as a highly capable hourly consultant. You want to give it more work because it is useful, but billed hours keep accumulating. Unlike a person, it can operate around the clock, so spending can exceed a salary unless the organization actively considers value and limits.
Frequently asked questions
What work could consume USD 40,000 or USD 1.3 million a month?
Reports pointed to development workflows with several agents running in parallel for long periods. Repeated reasoning and rereading of context can consume far more tokens than an ordinary human chat.
Could AI costs really exceed salaries?
It is a forecast, not a certainty, and lower unit prices could moderate it. The rapid usage growth reported in cases such as Uber shows why organizations are considering that possibility.
Should we cap employees' AI usage?
For metered services, establish appropriate limits or alerts. Set them alongside outcome expectations so that controls do not blindly stop high-value work.
Should we take Huang's USD 250,000 suggestion literally?
His position as the leader of an AI infrastructure supplier is relevant context. The more general idea, comparing AI investment with labor costs and resulting productivity, is useful even if the specific amount is inappropriate for your organization.
We do not have a development department. Is this relevant?
Coding is an early example. Agents are also being applied to writing, accounting, and inquiry handling, so multiple AI workers per person could become relevant across office work.
What are the disadvantages of treating on-premises AI as a fixed-cost employee?
Hardware has finite processing capacity and may not suit tasks requiring the latest very large models. Match the model and deployment to the work; see article 5.
Summary
- Reports cited salary-scale AI spending, including USD 40,000 monthly for one engineer and USD 1.3 million for a three-person team.
- A Gartner forecast suggested AI coding costs could exceed developer salaries by 2028.
- Huang's pro-spending argument and Japanese examples show a move toward comparing token costs with labor and ROI.
- AI spending is moving from a tool expense to a staffing-scale decision. The evaluation framework needs to evolve.
- Upfront-purchase AI provides a resource without additional local token charges. Payback depends on useful workload, capacity, and full operating costs.
Related articles
A year's AI budget, gone in four months Uber's challenge of successful adoption
Adoption reaching 84% upset the budget assumptions. Lessons from figures shared by the people involved
Token demand could grow 24-fold Why lower unit prices may not shrink the bill
Cheaper units do not guarantee lower totals. Forecasts and reported consumption explain the pattern
What would it cost us?
Estimate costs using your own workload
This article is based on public reporting, without independent interviews. Amounts reflect the reporting dates; second-hand claims are identified as such. Yen conversions are approximate, at around USD 1 = JPY 150. Please contact us with factual corrections. The media, researchers, and companies discussed do not endorse Sovereign GaiXer.
Company, product, and service names mentioned are trademarks or registered trademarks of their respective owners.