In recent months, the debate around artificial intelligence has stopped being a purely technological question and has become a question of systemic risk. Reports of pressure on rating agencies to keep AI-related debt in the investment grade category are not a detail for insiders: they are the symptom of a structural tension that affects every company now using generative AI tools in its processes.
The picture is more complex than a simple "bubble." It is a system in which real costs are offloaded onto actors other than those who generate them, and in which service continuity depends on increasingly fragile financial equilibria.
Why AI Debt Is a Wake-Up Call for AI Users, Not Just for Those Who Finance It
Most major tech companies have historically funded AI through cash generated by their core business. That phase is over. In 2026, AI-related bond issuance reached record levels, with hyperscalers turning to the markets to finance infrastructure worth hundreds of billions of dollars.
The figure that should make anyone using AI services think twice is this: more than 16% of the US corporate bond index is now tied to artificial intelligence. It is no longer a niche. It is a structural component of the credit market.
The uncomfortable question: if a significant portion of AI debt were downgraded, what would the consequences be for the availability and pricing of the services your company uses every day?
The Two Scenarios Every Company Must Consider
Scenario 1: Service Disruption
The concentration of the AI market is extreme. In the first quarter of 2026, the four leading AI companies raised 80% of all global venture capital. This concentration means that a single financial distress event can have immediate knock-on effects.
The Sora case is illuminating: despite seemingly unlimited resources, the product was shut down when its unit economics proved unsustainable — a daily operating cost of 15 million dollars against total revenues of just over 2 million.
For a company that has built critical processes on a single provider, disruption is not a remote hypothesis. It is a concrete operational risk. The Cloud Security Alliance has formally recommended treating the financial condition of frontier AI providers as a separate line item in the third-party risk register.
Scenario 2: Soaring Prices
The efficiency paradox is brutal: the cost per token has collapsed by roughly a thousand times in two years, yet total spending has increased because volumes grow faster than price declines. This phenomenon is known as the Jevons effect.
The real problem is not the unit price, but the cost of inference as a recurring bill. While training is a one-time cost, inference in production is an operating expense that grows with adoption. If providers were to reprice contracts to reflect the true costs of infrastructure, the impact on corporate budgets could be significant.
The Layoff Dilemma
The InvestCloud case in Marghera is not isolated. A fintech company laid off 37 qualified employees, justifying the decision with a "new organizational model based on systems integrated with artificial intelligence." The ruling by the Tribunal of Rome on a graphic designer fired after her duties were "progressively absorbed" by generative AI tools confirms that this dynamic is already the subject of litigation.
The problem is that many of these decisions were made in a context of artificially low AI prices. If the scenario changes, companies that replaced staff with AI find themselves in an uncomfortable position: they have eliminated internal skills that are difficult to rebuild, and now depend on external providers for critical functions.
Emergency Plan: What to Do Concretely
1. Map Your Real Dependency
It is not enough to know which AI tools you use. You need to understand which critical processes depend on a single provider and what the impact would be of a disruption or a 300-500% price increase.
The CSA recommends identifying workflows that generate revenue or are critical to security and that depend on a single frontier model without a validated fallback.
2. Build a Multi-Provider Fallback
Portability is not a luxury. It is a necessity. The Model Context Protocol and open-model-based architectures make it possible to shift competitive value from model weights to a company's internal data.
Concrete action: test a second provider for every critical function. Do not wait until it is operationally necessary to do so.
3. Renegotiate Contracts with Continuity Clauses
Current contracts rarely include protections for provider distress scenarios. The CSA recommends negotiating:
- Advance notice of material product discontinuation
- Data portability commitments
- Audit rights
- Consumption corridors and predefined terms for overages
4. Recalculate AI ROI Under Stress Scenarios
If token costs increased by 500%, which of your AI use cases would remain economically sustainable? This question should be part of any AI business case.
Gartner places AI in the "trough of disillusionment" for 2026, predicting that ROI predictability must improve before companies truly scale.
5. Rebuild Critical Skills, Not Just Tools
If you have laid off staff and replaced them with AI, assess which skills have been irreversibly lost. The ability to handle exceptions, make decisions in ambiguous contexts, maintain customer relationships — these are skills that AI does not easily replicate.
6. Include Provider Financial Risk in Continuous Monitoring
Do not limit yourself to evaluating the technical quality of the model. Monitor:
- Rating actions on the provider's debt
- Funding rounds and terms
- Changes in pricing models
- Signs of financial distress
The CSA recommends a recurring cadence, not just at contract renewal.
The Question No One Wants to Ask
The cost of AI for a company today is a fraction of the real cost that providers are bearing. This divergence is not sustainable indefinitely.
The question is not whether there will be an adjustment, but when and how. Companies that prepare now — building fallbacks, negotiating contractual protections, retaining critical skills — will be the ones that survive the adjustment without devastating operational disruptions.
The others will discover, at the worst possible moment, that the AI that seemed free had a hidden cost that someone else was paying. And that someone has stopped paying.
And do you have an emergency plan?
Start with your website: ditch the bloatware and embrace good old WordPress—we’re here to help 🙂