Enterprise AI spending is hitting a breaking point, and it's forcing companies to rethink how they build their AI infrastructure. When Canva Inc., a design platform generating over $900 million quarterly, cut its 2026 revenue-growth forecast from 30% to 20%, the culprit wasn't market demand or competition. It was the cost of running AI features built on expensive third-party frontier models.

The company's experience reveals a deeper economic reality that's reshaping how enterprises approach artificial intelligence. Canva didn't abandon AI. Instead, it rebuilt its entire stack with in-house models and task-level routing, reportedly cutting the cost of an AI task by roughly 90%. Its video and image models became 17 and 30 times cheaper than frontier alternatives.

Why Are Enterprise AI Costs Spiraling Out of Control?

The problem isn't that AI doesn't work. Companies like Uber Technologies Inc., Microsoft Corp., and Lindy (registered as Crivello Corp.) all deployed advanced AI models successfully. The issue emerged when they received their invoices. Uber consumed an entire year's AI budget in a single quarter. Lindy found that Anthropic PBC had become its largest expense, bigger than payroll. Microsoft, despite its partnership with Anthropic, began building more of its own model capability while openly stating it wants to reduce and ultimately eliminate the cost of paying Anthropic.

These aren't companies rejecting AI because of ideology. They're making hard financial decisions because the economics no longer work. The industry has been measuring progress in tokens, model calls, and usage metrics, but those are largely vendor-revenue metrics, not enterprise-value metrics.

How Are Enterprises Taking Control of Their AI Economics?

• Routing Workloads Strategically: Companies are resetting defaults and routing workloads toward lower-cost models instead of defaulting to expensive frontier options for every task.
• Building In-House Capabilities: Organizations are developing their own models and deploying them on company-controlled infrastructure rather than relying entirely on third-party APIs.
• Switching Providers When Margins Compress: Enterprises are actively moving traffic between model providers to reduce costs while maintaining or improving performance on core use cases.

The shift reflects what industry analysts call "financial sovereignty," a concept that goes beyond just data residency or self-hosting. It's about controlling the economic terms under which AI operates. A sovereign enterprise controls its data, evaluations, policies, routing, cost telemetry, and exit paths. It decides what to own, what to rent, and when frontier capability creates a genuine advantage.

Palantir Technologies Inc. Chief Executive Alex Karp has attacked what he calls "tokenmaxxing," which is optimizing the vendor's bill rather than the value the enterprise retains. The alternative, according to analysts, is "sovereign alpha," retaining more of the value created from your data, workflows, and domain expertise because you control the cost curve and preserve the ability to move.

What Does Financial Sovereignty Actually Mean for Enterprises?

Financial sovereignty isn't a binary choice. It exists on a spectrum, and every organization must decide where to place its boundary conditions. Microsoft's approach illustrates this nuance. The company offers Microsoft Foundry, a platform that includes Microsoft models, partner models, and customer-controlled deployments. Its use of Anthropic can be understood as a developer-focused decision: engineers may prefer Claude for certain tasks. But once that preference breaches the company's pain tolerance for margin degradation, it creates a financial trigger to impose controls and route work elsewhere or use lower-cost internal and open-weight models.

Uber's situation was more straightforward. The company used state-of-the-art models for the highest and best use of its engineering talent, and that strategy worked well until it ran into a financial wall. When you burn through an entire year's budget in roughly four months, there's no way to confidently budget two or three years out the way a chief financial officer is required to do.

Lindy and Canva discovered that owning their financial alpha was more strategic than claiming state-of-the-art capability. At the end of the day, businesses run on margins. If your financial livelihood relies on a vendor whose economics can compress your margins, the model becomes unsustainable.

Even Switzerland's sovereign AI initiative, Apertus, developed by ETH Zurich, EPFL, and the Swiss National Supercomputing Centre, illustrates the complexity. The project trained a fully open model on the Alps supercomputer across more than 1,000 languages, representing a meaningful sovereign asset. However, owning a model and training infrastructure doesn't automatically make every downstream deployment sovereign. Swiss organizations may still choose commercial cloud capacity, including Microsoft's locally hosted services, when that's easier or cheaper to operationalize.

The takeaway is clear: capability can be rented, but control must be architected. No vendor can confer sovereignty. Vendors provide components. Only the enterprise can define and enforce its sovereignty boundary. As more companies face the reality of AI cost escalation, the question is no longer whether to use AI, but how to structure your AI economics so that you capture the value your organization creates.