The AI Bill Is the Wrong Question

Op-ed by Erwan Simon, CEO and co-founder of GENIAL

While France worries about the price per token, OpenAI and Anthropic have just invested $5.5 billion in a completely different problem: the human cost of AI. That shift is what French business leaders need to understand before they keep making decisions about their bill blindly.

A two-sided paradox

On Tuesday, May 12, Les Échos (France's leading business daily) ran a headline about companies breaking into a "cold sweat" over their AI bills. The observation is accurate, but it misses the point.

On the same bill, a French business leader today sees two opposite overruns. On one side, professional licenses at €20 or €200 per seat that go unused: a visible cost, invisible value. On the other, tokens that skyrocket, €4,000 for the month when €800 was budgeted: a visible cost, value that's hard to measure. Opposite conclusions, same cause: nobody measures usage. You look at a bill, look for someone to blame, and cut by default.

Yet the price per token drops tenfold every year. Building your AI strategy on unit cost is building on sand.

A confidence gap, not a technology gap

EY's AI Sentiment Study, published in March 2026 (18,000 people, 23 countries), ranked France among the furthest behind, alongside Japan, the United Kingdom, Canada and Australia. Meanwhile, in India, China, the UAE, Brazil, Mexico and South Korea, the new pioneers, 94% of the population uses AI and nearly one person in four lets an AI act on their behalf.

Our lag isn't technological, financial or human. It's a confidence gap, fed by a French debate focused on costs and risks, and unable to ask the right question.

The number that should have made the headlines

On May 12, 2026, the very day of the Les Échos article, Arthur Mensch, co-founder of Mistral AI, gave a figure before a French National Assembly committee of inquiry that nobody picked up on: "Our AI consumption for our employees is 10% of our payroll. If you extrapolate, three or four years from now, 10% of payroll in Europe is roughly a trillion euros."

Ten percent of payroll. That's no longer a software expense. It's a budget category in its own right, comparable to training, temp staffing or outsourcing.

What the vendors just made official

The French debate is still talking about the price per token. Meanwhile, the vendors themselves have just shifted $5.5 billion toward a completely different issue: the human rollout of AI inside the customer's organization.

On May 4, 2026, Anthropic announced a $1.5B joint venture with Blackstone, Hellman & Friedman and Goldman Sachs to embed engineers directly in its customers' operations. Jon Gray, president of Blackstone, named the bottleneck: "One of the main bottlenecks to AI adoption is the scarcity of engineers who can deploy at high speed."

On May 11, OpenAI responded with $4B backed by 19 funds led by TPG, and acquired Tomoro, a Scottish firm of 150 Forward Deployed Engineers: hybrid profiles, half engineer, half business operator, popularized by Palantir, whose job is to embed in the customer's teams to turn a technical capability into business value.

The signal couldn't be clearer. Fortune sums it up in one number: for every dollar spent on software, companies spend six dollars on services. That's the ratio the two American joint ventures have just positioned themselves to capture.

This turn reverses thirty years of enterprise IT. Before: you bought software, paid for a turnkey integration project, and waited for a productivity gain. Today: you buy raw capacity (tokens, agents), and the business integration layer has to be built inside the company. That layer is human. An investment that doesn't produce an immediate gain, but the capacity to produce gains.

Leaders are behind their own teams

There's one last gap that should worry leaders as much as the bill. While executive teams debate strategy, two out of three French employees already use generative AI, and half of them use it weekly.

AI is no longer a transformation topic for tomorrow. It's already in the workflows, but invisible to leadership. When a leader discovers the bill at the end of the month, what they're really discovering is that they didn't know what their own teams were doing with the tools they had, or hadn't, approved.

The asymmetry is even sharper in SMBs and mid-market companies, largely absent from the public debate. No CIO, no procurement department, no FinOps. Teams adopt through Shadow AI, the leader takes the hit on the bill, and cuts by default. A strategic disaster, at the precise moment when it should be managed.

AI spending is an HR expense

If Mensch is right, and if 10% of European payroll spent on AI is no longer a projection but a trajectory, then the budget line to watch is no longer SaaS licenses. It's a fraction of payroll itself, halfway between software spending, training spending and outsourcing spending.

That 10% also fits the trend of companies able to multiply revenue per employee by ten, as shown by companies such as Gamma, Lovable and Manus.

For French leaders who want to close the gap EY pointed out, the urgent task isn't negotiating better prices with Anthropic or OpenAI (dependence on these vendors is the other side of the issue, covered in AI sovereignty: when unplugging an AI becomes a 2027 presidential election issue). It's to stop thinking in terms of "IT cost" and start thinking in terms of "human investment relative to algorithmic consumption." In practice:

• How much of my teams' time goes into turning these tokens into value?
• Which profiles am I missing?
• Am I ready to invest in teams that won't produce an immediate gain, as the Americans just did to the tune of $5.5 billion in a single week?

The question is no longer "how much does AI cost?" It's: "what share of my people budget am I willing to reallocate so this technology creates value for the company, and how will I know?"

Until French leaders start answering that question, they'll keep making blind decisions on a subject where their teams are already moving ahead in the open.

---

Erwan Simon is CEO and co-founder of GENIAL, a Bordeaux-based company that specializes in the operational rollout of generative AI in SMBs and mid-market companies. He is an accredited AI Expert with Bpifrance (France's public investment bank) and an ambassador of "Osez l'IA," the French government's program to help businesses adopt AI.