The automobile did not enter our lives as a necessity. It began as a mechanical wonder, became a convenience, and eventually changed where we lived, worked, shopped, and worshiped. Before long, American life was organized around the assumption that a family would own a car.

The purchase price was only the beginning. The owner also inherited gasoline, insurance, maintenance, repairs, taxes, parking, and eventual replacement. Freedom of movement came with a permanent place in the household budget.

The telephone traveled a similar road. What began as a fixed instrument attached to a wall became a mobile phone, then a smartphone, and finally a pocket computer that many Americans can scarcely function without. As its usefulness grew, so did our dependence and the monthly bill.

Artificial intelligence appears to be traveling the same road.

I subscribe to ChatGPT and Claude. The arrangement seemed familiar enough: pay a monthly fee and use the service. Then Claude informed me that use beyond my plan could be billed through usage credits, charged by the token.

A token is not a coin but a small unit of information processed by an AI system. Providers often price usage by the million tokens. Yet few consumers can predict what an ordinary question, a long document, repeated revisions, or an AI agent completing a complicated assignment will cost before the meter starts running.

Welcome to the AI cost maze.

At the top of that maze are sums beyond ordinary comprehension. Technology companies are pouring money into chips, data centers, electricity, cooling, engineers, and model development. Goldman Sachs estimates that roughly $7.6 trillion may be invested globally in computing capacity, power, and data centers between 2026 and 2031. Wall Street is already trying to trade computing power somewhat like oil. Yet computing capacity is not oil. It cannot be stored in a tank, and each new chip generation can quickly erode the value of the last.

At the other end of the maze, the price of using AI keeps sliding. Chinese AI company DeepSeek recently released a coding model charging about 28 cents for a million output tokens, compared with $25 for Anthropic’s Claude Opus. Whatever the technical gap between the two, the price trend is unmistakable.

That produces the great AI cost paradox: Artificial intelligence may become cheaper by the unit while becoming more expensive as a habit.

Families may pay directly through subscriptions, premium plans, usage credits, or specialized applications. Even Americans who never subscribe to a chatbot may pay indirectly for AI embedded in banking, insurance, health care, retail, education, government, and the devices they already own. The cost may be buried in another bill rather than labeled “artificial intelligence.”

Money, however, is only part of the expense.

AI also costs time. A system may draft a letter in seconds, but someone must still read it, correct it, verify its claims, and accept responsibility for the result. An inexpensive model that repeatedly errs may cost more than a premium model that gets it right the first time.

GitHub Copilot users recently received a lesson in that reality. Complex AI work began consuming their monthly credits far faster than many expected. Some users turned to less expensive models, only to discover that correcting inferior results required additional time and additional usage.

There is also the cost of error. A poor restaurant recommendation is inconvenient. Bad advice involving medicine, taxes, investments, contracts, or employment can inflict real harm. The more consequential the decision, the more expensive blind trust becomes.

Then there is privacy. AI often becomes more useful as users entrust it with more of their work, purchases, correspondence, health, finances, preferences, and relationships. A 2025 consumer survey of 1,008 adults found that 61% had tried generative AI, with many using it to compare prices and find deals. Yet consumers also expressed concerns about privacy, unwanted subscriptions, overspending, misinformation, and poor product choices.

The proper question, therefore, is not, “How many tokens did I buy?”

We do not buy gasoline because we desire gallons; we buy the ability to reach a destination. Cellular data works the same way — gigabytes mean little by themselves; the communication and access they deliver mean everything.

The same principle should govern AI. Did the service provide a reliable answer, save meaningful time, reduce an expense, improve the quality of work, or make possible something the user could not otherwise accomplish? When the answer is no, even inexpensive AI is overpriced.

Families should total the annual cost of every AI subscription rather than viewing each monthly charge alone. They should subtract the time spent checking and correcting a tool’s output from the time it supposedly saves, cap automatic overage charges, and cancel overlapping services that do not provide distinct value.

Businesses face the same test. As McKinsey warns, additional capacity is not automatically more valuable. An hour saved matters only if it produces more work, better service, lower costs, or less risk. Otherwise, the claimed efficiency exists mainly on a presentation slide.

Christians have another reason to approach this technology soberly. Paul told the Corinthians that “it is required of stewards that they be found faithful” (1 Corinthians 4:2, ESV). Faithfulness with a subscription bill is a small thing next to the matters Paul had in mind, but a family that never examines what AI costs in money, time, privacy, and judgment cannot know whether the exchange is wise.

Christian stewardship asks more than whether an AI service is affordable. It asks whether the tool helps us fulfill our responsibilities, pursue truth, and exercise judgment rather than surrender it — and whether the time it frees up returns to family and neighbor, or simply fills our time with more digital distraction.

The automobile brought freedom, but also expense and dependence. The smartphone brought extraordinary access, but also claimed our attention and reshaped daily life. AI may prove even more consequential because it reaches beyond transportation and communication into work, thought, judgment, and decision-making.

AI may become our next necessity. Before it does, Americans should know what they are paying, what they are receiving, and what they may be surrendering. A useful tool should remain a servant. Whatever its price in dollars, AI costs too much when we permit it to become the master.