Your AI costs more than you

Every engineering team I know has quietly crossed a threshold: the total cost of AI tools per developer now rivals a meaningful chunk of a junior engineer's salary. And almost nobody is tracking it. The line items are scattered. GitHub Copilot here, a Claude Pro subscription there, some OpenAI API usage on this credit card, cloud compute for fine-tuning on that one. Add in the team-wide ChatGPT Enterprise license that half the org forgot they had, and you're looking at a number that would make any CFO uncomfortable, if they could actually find it.

The back-of-napkin math

Let's walk through what a single developer's AI stack might cost in 2026. GitHub Copilot Pro runs $10/month, though many power users opt for Pro+ at $39/month. Claude Pro is $20/month. ChatGPT Plus is another $20/month. Then there's API usage for internal tools, prototyping, and side projects, easily $50 to $200/month depending on the team. Layer on cloud compute for any fine-tuning or inference workloads, and you're adding another $100 to $500/month. That's $200 to $780 per developer, per month, before we even count the more specialized tools: Cursor, Midjourney, Notion AI, Perplexity Pro, or whatever the team picked up last quarter. A DX newsletter survey of engineering leaders found that over 20% of organizations are already spending more than $500 per developer per year on AI tools alone, and nearly half have dedicated 1-3% of their total engineering budgets to AI tooling. Those numbers are climbing fast. At the high end, a well-equipped developer's AI bill can easily clear $1,000/month, or $12,000/year. Multiply that across a 50-person engineering team and you're looking at $600,000 annually, a figure that starts to compete with two or three junior engineer salaries.

Nobody owns this line item

The real problem isn't the spending itself. It's that no single person or team is responsible for tracking it. GitHub Copilot might sit under engineering ops. ChatGPT Enterprise gets billed to the IT department. Individual Claude and API subscriptions show up on personal expense reports. Cloud compute for AI workloads is buried in the infrastructure budget. The result is a cost center that's growing rapidly but has no owner, no dashboard, and no review cycle. This is a pattern we've seen before.

The SaaS bloat parallel

A decade ago, companies went through the same thing with SaaS subscriptions. Teams adopted tools independently, nobody tracked overlap, and by the time finance caught on, the average organization was running 275 SaaS applications with 25-30% of spending going to unused or underutilized licenses. Zylo's 2025 SaaS Management Index found that the average organization wastes $21 million a year on unused SaaS licenses. Even companies with fewer than 500 employees waste an estimated $4.2 million annually. The SaaS sprawl era taught companies an expensive lesson about decentralized purchasing. It took years of audits, consolidation projects, and purpose-built management platforms to get spending under control. And now, AI subscriptions are layering on top of that barely-contained stack. The irony is hard to miss: tools sold as cost-saving measures are themselves becoming a significant, unmanaged cost center.

The ROI question nobody wants to answer

The uncomfortable truth is that most organizations can't tell you which AI spending is actually producing returns. Morgan Stanley Research estimates that nearly $3 trillion in AI-related infrastructure investment will flow through the global economy by 2028, with more than 80% of that spending still ahead. But the gap between investment and measurable results remains enormous. Forbes reported earlier this year that fewer than 10% of enterprises report measurable ROI from their AI investments. IDC research found that 88% of AI proofs of concept never make it to production. MIT's NANDA initiative puts the figure even higher, estimating that 95% of enterprise AI pilots stall before delivering measurable returns. The enterprises that do reach production deployment report strong results, with Morgan Stanley citing an average 171% return on investment for companies running agentic AI in production. But only 11% of enterprises that adopt AI ever get there. The rest are stuck in what analysts call the "pilot plateau," spending on tools and experiments without a clear path to sustained value. This dynamic plays out at the individual level too. A developer with five AI subscriptions might use one of them 90% of the time. The other four are insurance policies, costing $40-80/month each for the occasional use case that could probably be handled by the primary tool.

Smart spending, not less spending

None of this means AI tools aren't worth paying for. The productivity gains from a well-chosen coding assistant or a capable LLM are real and well-documented. The question isn't whether to use AI, it's which AI spending actually produces value. A few principles that help: Audit what you're actually using. Most developers and teams have accumulated subscriptions through experimentation. A quarterly review of what's active versus what's collecting dust can easily cut 30-40% of per-developer AI costs without any loss in productivity. Consolidate where tools overlap. GitHub Copilot, Claude, and ChatGPT all have significant functional overlap for common tasks. Picking a primary tool and using free tiers or pay-as-you-go for edge cases beats paying full price for three competing subscriptions. Track AI spending as its own category. If your organization doesn't have a line item for "AI tools per employee," create one. The SaaS era proved that invisible spending grows unchecked. Visibility alone changes behavior. Measure outcomes, not seats. Morgan Stanley's research suggests that a 1-2% margin uplift from AI can justify substantial investment, but only if you can actually measure that uplift. Companies that tie AI spending to specific productivity metrics are far more likely to sustain and scale their investments.

The prediction

Within the next 18 months, "AI cost audit" will become a real job title, or at least a real line item in consulting engagements. The same way FinOps emerged to manage cloud spending and SaaS management platforms emerged to tame subscription sprawl, a new discipline will form around understanding, measuring, and optimizing per-employee AI costs. The companies that get ahead of this won't be the ones spending the least on AI. They'll be the ones who actually know what they're spending, and can prove it's working.

References