Companies that can control rapidly rising artificial intelligence costs may gain an advantage as AI models become increasingly commoditized, according to PwC.
The professional services firm said AI cost-control tools are becoming widespread and standardized, making them necessary to compete but less useful as a differentiator. Disciplined spending could also free capital for additional AI initiatives and create a compounding advantage.
Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram
One global technology company reportedly cut the cost of each AI run by 65% to 80%, allowing it to run three to five times as much AI on the same budget.
Why AI Spending Keeps Rising
Token prices are falling, but total AI spending continues to increase as lower unit costs encourage broader deployment. More workflows can also mean more calls, retries and system dependencies.
“Everyone tries to use AI everywhere, even if it just makes workflows more complex and expensive,” PwC said, noting that access to the same underlying models limits the competitive value of higher spending.
Companies also often lack visibility into token consumption and where waste occurs.
Hidden Costs Add Up
AI expenses can accumulate across planning, tool use, retrieval, reasoning, orchestration, safeguards, logging and review. Indirect infrastructure costs are also often excluded from initial budgets.
Agent-based systems can increase spending further by creating plans, delegating tasks, retrieving information or repeating processes when results fall short.
Model costs vary sharply, with PwC estimating that one million tokens can cost anywhere from pennies to $50. Choosing the cheapest model is not necessarily the best option because weaker systems can create additional work, poor decisions or compliance problems.
Financial Discipline Can Reduce Waste
PwC recommends examining three sources of AI cost overruns: rates, such as supplier price changes; volume, including excessive calls and retries; and mix, meaning the wrong model tier for a task.
Its operating model calls for assessing cost and value before development, redesigning systems to eliminate waste, linking spending to business outcomes and reinvesting savings in additional AI projects.
Companies can reduce costs by limiting unnecessary context, combining tasks into fewer calls, setting spending limits and routing work to the least expensive suitable model. PwC said these controls should be built into AI systems through budget limits, routing rules, workflow thresholds and audit trails.
Human Oversight Still Matters
Automated controls do not replace human oversight. PwC said technology should flag decisions for review and provide the information needed to align actions with business priorities.
In the technology company case study, the approach cut average runtime from 12 hours to four hours while maintaining output quality. PwC recommends tracking the cost of each AI workflow against its business outcome, putting AI spending on the CFO’s agenda and preparing for more outcome-based vendor pricing.
Discipline May Define The Next AI Advantage
PwC said companies should start with their most valuable AI applications, where better cost management and governance can deliver the greatest returns.
“The next round of AI advantage won’t go to whoever runs the most powerful models,” PwC said, noting that many companies will use the same underlying systems.
“Advantage will likely go to whoever runs them with more discipline,” the firm concluded.
