The price of a single AI token dropped by more than 90% from 2023 to 2026. That should have made AI cheaper for businesses, right? But here is the thing: enterprise AI spending has more than doubled in the same period. So why are companies paying more if tokens cost less?
The answer lies in how organizations use AI. When tokens get cheaper, businesses run more AI agents. They automate more workflows and generate more code. This means they consume far more tokens overall. Costs don’t depend on token price alone. Instead, thousands of small decisions about which AI models to use drive expenses.
Jean-Michel Lemieux, former Shopify CTO now at Spellbook, explains the challenge clearly. He says, “I’m in these board meetings, and it feels like I’m a coach of a pro team, but I can’t see them play the game.” He means executives struggle to grasp how AI costs add up on a daily basis.
Reza Khadjavi, CEO of Motion, adds, “If I don’t go into the weeds on this stuff, the company might die.” He builds tools his team can actually use and understands the value of controlling AI spending tightly. For him, hands-on work is essential.
Why AI Costs Spiral Out of Control
AI cost problems surface on monthly invoices, but the real drivers hide in daily choices. Each time someone picks a model or decides to automate a process, costs rise. Arun Shastri, global AI Leader at ZS, points out that organizations embed AI in more workflows. This means more usage and more costs.
Manos Koukoumidis, CEO of Oumi AI, warns against using the most powerful, expensive AI models for simple tasks. He calls this “widely irrational.” Choosing the right tool for the job matters a lot. Using a big, costly model for a small problem wastes money.
Robert Sweeney, former Meta engineering manager and now at Citizen Health, asks, “How can a director or a senior manager even hold the engineering team accountable at all if he has no concept of what it takes to build the software?” Without understanding, leaders can’t control costs effectively.
Fintech and Software Companies Navigate AI Growth
Several fintech and software firms show how AI and software engineering intersect today. Relevant Software reported over 200 projects for 200+ clients in 2026, with 92% senior staff and 96% retention. Their clients rate them 4.9 on Clutch and 9.8 NPS scores, with 98% satisfaction.
Topflight Apps generated over $200 million in client net worth through its products. The Software House completed over 30 fintech projects, focusing on payment infrastructure. PixelPlex worked on more than 80 crypto projects. Software Mind now has over 1,000 team members modernizing enterprise banking. EffectiveSoft delivered 1,800+ projects in capital markets and open banking.
BNP Paribas’ GOmobile app passed 1 million downloads, showing fintech’s reach. These companies rely on AI but must balance costs carefully. Their success depends on smart, efficient use of AI across many workflows.
Jean-Michel Lemieux sums it up well: “Our bet is the company could be five times better than it would be without doing this.” The potential of AI is huge. But managing its costs means understanding the small choices behind the scenes.
Based on
- Platform Engineering ROI: What it costs to build your own platform — thenewstack.io
- Why tech leaders are writing code again | Business Insider Africa — africa.businessinsider.com
- Relevant Software Releases 2026 Fintech Infrastructure Market Report Analyzing Top Engineering Partners | Currency News | Financial and Business News | Markets Insider — markets.businessinsider.com
- How to Keep Innovating Without Burning Through Your AI Budget – Business Insider — businessinsider.com
- Meet the People Carving Out a Slice of the AI Cost-Saving Business – Business Insider — businessinsider.com
