Cloud spend has become a strategic liability in the age of AI. This article explores why CFOs must take charge of cloud economics, from forecasting and cost visibility to governance and board reporting.
Over the last 18 months, I’ve watched cloud move from an operational line item to a top-three boardroom issue. It’s now second only to payroll for most venture-backed companies but, unlike headcount, it’s volatile, unpredictable, and increasingly tied to the most strategic decisions leaders are making today.
The shift has been accelerated, even distorted, by the rapid adoption of AI. This is not just a cost spike. It’s a new economic model. While traditional SaaS businesses often enjoy 75–80% gross margins, AI-native software companies are reporting figures closer to 50–60%, with some dropping below that. It’s not a temporary inefficiency. It’s structural.
These aren’t growing pains. They’re a new baseline.
Unlike traditional software, where infrastructure costs taper as the business scales, AI expenses grow almost linearly with usage. Every inference call, every chatbot session, every API hit – all incur a marginal cost. And when you can’t flatten cost against adoption, you no longer have the same gross margin profile or unit economics. That changes everything, especially if you’re on a path to IPO or preparing for acquisition.
Boards have caught up fast. Glossy AI roadmaps aren’t enough. They want to know how you’re defending margin. They expect you to prove that your AI strategy won’t cannibalise long-term profitability. That means forecasting infrastructure costs with real precision and articulating how each dollar of cloud spend ties back to business value.
If that weren’t enough, AWS removed a long-standing financial safety net in June 2025 by ending cross-customer discount pooling. That decision exposed CFOs to far more direct commitment risk, right when usage volatility has never been higher. The timing couldn’t be worse.
All of this has brought us to a new reality: cloud has evolved from a flexible cost centre into a strategic liability, and one that must be actively managed with the same rigour as debt or equity risk.
Many finance leaders still rely on static cloud forecasts, point estimates based on the prior quarter’s usage or “growth plus 10%” assumptions. But AI doesn’t work that way. Usage can double overnight. New features can go viral. And underestimating cloud consumption by even 10–20% can create massive variance between reported and expected margins.
The shift we’re seeing among more mature teams is toward driver-based forecasting: cloud cost models that are anchored to actual business inputs – user sessions, inference calls, API volumes. These aren’t just engineering metrics anymore, they’re critical finance levers
At Cloud Capital, we’ve seen finance and engineering teams start sharing these assumptions monthly, not just in annual planning meetings. They’re holding joint quarterly reviews and setting clear targets: +5% forecast variance. Anything beyond that is now considered a financial control failure.
One of the most urgent problems in cloud economics today is how AI infrastructure costs are accounted for or more accurately, how they’re not. Too many companies lump AI expenses into general “cloud infrastructure”, blurring the financial impact of their most strategic (and expensive) technical investments.
That’s a mistake. Inference costs, the ongoing expense of serving live AI experiences to users, now represent a significant share of cost of goods sold (COGS) for AI-native software. And yet very few companies can articulate what an AI feature actually costs to operate.
The leading teams are changing that. The best-run finance teams are introducing sub-ledgers for AI infrastructure, clearly separating model training (one-off) from inference (recurring) and calculating metrics like cost per AI transaction or cost per 1,000 inferences. That kind of granularity unlocks efficiency levers like model distillation, better GPU utilisation, and smarter routing but more importantly, it brings AI into the domain of gross margin management.
Boards are already asking for this visibility. Investors won’t be far behind.
What used to be a back-office optimisation task, managing Reserved Instances or Savings Plans, has become an active area of financial risk. With the end of cross-customer pooling, companies are now individually responsible for any commitment missteps.
Overcommit and you pay for capacity you can’t use. Under-commit and you pay on-demand premiums that destroy margin.
Sophisticated finance teams are beginning to treat cloud contracts like a bond portfolio, laddering commitment terms across 12, 24, and 36 months to smooth renewal risk, tracking utilisation rates obsessively, and modelling downside scenarios to stress-test their exposure.
The goal here is not just cost savings. It’s governance. It’s being able to walk into a boardroom and say, “We’re managing our cloud portfolio with the same sophistication we’d apply to capital structure or FX exposure.”
Of course, none of this works if you can’t track where spend is going. And the unglamorous truth is that many cloud environments are still poorly tagged, meaning infrastructure usage is effectively untraceable.
This isn’t just an operational inconvenience. It’s a financial blind spot. Without reliable tags for things like team, environment, or product line, you can’t create accurate forecasts, build unit economics, or hold business units accountable.
The gold standard here is full enforcement at the CI/CD level, tagging policies that prevent untagged resources from being deployed in the first place. With that foundation in place, CFOs can produce dashboards that show cost per customer, cost per feature, and even cost per AI interaction. These aren’t just helpful, they’re diligence-ready metrics that inspire investor confidence.
Beyond cost, there’s a growing awareness that cloud infrastructure carries broader forms of risk – compliance, ESG exposure, and audit readiness. And yet many companies treat cloud governance as an afterthought.
The best finance leaders are changing that. They’re adding dedicated budget lines for cloud compliance and FinOps, factoring in higher costs for regulated workloads (FedRAMP, HIPAA, GDPR), and forecasting for governance tooling across the organisation. Some are even tracking Scope 3 emissions tied to cloud usage and optimising region selection to reduce carbon intensity per dollar of revenue.
It may seem like overhead, but this is fast becoming table stakes, not just for enterprise sales, but for investor-grade operational maturity.
What we’re witnessing is a shift in the role of finance. In the age of AI, cloud isn’t just a tech issue. It’s an economic one. And it’s increasingly visible in investor conversations, margin analysis, and valuation models.
The companies that win won’t be the ones with the most compute, they’ll be the ones with the clearest handle on how compute drives business value. That means budgeting with precision, isolating AI costs, managing commitments like financial instruments, enforcing discipline through tagging, and treating governance as core infrastructure.
The days of delegating cloud to engineering are over. The next era of capital-efficient growth belongs to financial leaders who take full ownership of cloud economics and treat it as a board-level priority.
By: Ed Barrow, Co-Founder and CEO, Cloud Capital