The CFO Meets The Meter: AI Spending Will Not Sit Still

Boards that budgeted for artificial intelligence as a fixed line item are learning that its true cost scales with every query the CFO must now account for.

Sign up for my Substack daily AI newsletter here.

See my AI Training course portfolio for corporate Business Leaders here.

Article link: https://open.substack.com/pub/johanosteyn/p/the-cfo-meets-the-meter-ai-spending?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

Something has changed in the economics of enterprise artificial intelligence, and most boards approved the budget before it did. For years, technology was bought the way software has always been bought, as a fixed licence with a predictable annual figure a chief financial officer could plan around. The tools arriving now do not behave that way. They bill by consumption, and the meter runs faster the more the organisation leans on them. The question facing finance is no longer whether to invest in AI, but how to account for a cost that refuses to stand still.

CONTEXT AND BACKGROUND

The change is in the shape of the bill. Analysts describe how agentic AI moves the enterprise from fixed software and labour costs to variable compute use, where charges are levied per unit of input and output an AI model processes, so that an agent is no longer a licence but a metered operating cost. The same analysis notes that a query which once cost $0.04 can, once tools and reasoning steps are added, become a $1.20 orchestration, and that this pricing is tied to a constrained supply of chips, power and data centres, surfacing as usage caps, rate limits and unexpected repricing. Many providers are moving from flat subscriptions to consumption-based models, ending an era of predictable, and often subsidised, pricing. For a chief financial officer used to a stable annual figure, the past few months have been a shock to the budget.

INSIGHT AND ANALYSIS

This is why the AI line will not sit still. Because cost scales with consumption, and because agentic workloads consume far more than a simple chatbot, total spending climbs even as the price of each unit falls, and annual budgets can be exhausted long before the year is out. The discomfort deepens when the return is examined. Research from the Massachusetts Institute of Technology found that despite $30bn to $40bn in enterprise generative AI investment, roughly 95 per cent of organisations reported no measurable impact on profit and loss, with only about five per cent extracting significant value. Rising cost meeting absent return is the reckoning finance chiefs now face.

I have previously written about this, in a piece arguing that when AI hands staff a day a week of saved time, the failure to bank it is a management problem rather than a technology one, and that a board which cannot say what it received for that time is watching value evaporate. Paying for AI and capturing its return are not the same thing.

IMPLICATIONS

The response is a shift in who holds the pen. Finance is moving from permissive optimism to disciplined scrutiny, and the blank cheque is being withdrawn. One survey of finance leaders found that just 20 per cent of finance AI projects lean towards decision quality while 45 per cent chase productivity, leaving a gap between reported AI activity and the strategic value boards actually expect. Chief financial officers are answering by raising the bar. One has described requiring a business case with clear cost or efficiency impacts before approving material AI spend, observing that a blank cheque for AI makes prudent allocation very difficult. For South African boards the pressure is sharper still, because these tools are priced and metered in dollars while budgets are set in rand, so a weakening currency and a rising meter compound each other. Prudent financial control, a fiduciary duty here, now extends to a cost that moves every day.

CLOSING TAKEAWAY

None of this is an argument against artificial intelligence. It is an argument for governing it as the variable cost it has become. A board that approved AI as a fixed line item should now treat agent capacity as capital to be allocated, not a subscription to be renewed. In practice that means three disciplines: giving every AI initiative a value metric on the day it is approved, so spending is weighed against what it returns; setting spend ceilings and usage caps at the team and workflow level, so the bill is visible before the damage is done rather than after; and reviewing consumption monthly, because an annual cycle cannot keep pace with a cost that changes weekly. The meter is now part of the business. The task for leadership is not to fear it, but to read it, and to make every unit of spend earn its place.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net