Enterprise AI budgets are flowing in two directions at once, and only one of them builds an asset the buyer actually controls.
Gartner's latest forecast puts worldwide AI spending at $2.7 trillion in 2026, but the composition of that figure matters far more than the total.
Executives who read only the headline will miss the tension between what they are buying and what they actually own.
The Number Behind the Number
Gartner projects 2026 AI spending is a 49.5 percent increase over 2025, with $3.6 trillion forecast for 2027.
AI infrastructure accounts for $1.5 trillion of this year's total, more than half.
Gartner's John-David Lovelock calls the data center buildout the largest infrastructure project humanity has undertaken, and describes demand as inelastic even as memory prices climb.
Compute Concentration is a Sovereignty Issue
Gartner expects hyperscalers and service providers buying AI-optimized servers to remain the largest single area of spending.
For an enterprise, that means the compute layer beneath its AI strategy is being built, owned, and priced by a handful of providers. That could result in a high-risk bet that will be difficult to manage.
Sovereign Intelligence does not require owning an AI data center; it requires retaining the ability to choose, shift, and negotiate. Which of your AI workloads could move elsewhere, and at what cost, if the terms change?
Embedded AI Quietly Erodes Optionality
AI software spending is forecast at $462 billion in 2026, that's up 60 percent from $288 billion a year earlier.
Much of that is Agentic AI capability embedded in incumbent vendor products. Gartner notes that buyers are adopting these proprietary features despite known risks around vendor lock-in, data sovereignty, and potential runaway costs.
After three decades following enterprise technology investments, I recognize the pattern: CIO convenience wins the first purchase, and the CFO's switching cost arrives at contract renewal.
The Custom-Build Counter-Current
Gartner raised its 2026 growth outlook for AI application development platforms from 28 percent to 39 percent, and for Generative AI models from 110 percent to 117 percent.
The stated drivers are enterprise demand for tailored applications and cost-efficient domain-specific language models. This segment is small next to embedded spending, yet it is where proprietary advantage is actually created.
Buyers who want enterprise AI shaped by their own data, workflows, and business outcome economics are voting with their IT budgets.
The Underfunded AI Foundation
AI data spending is forecast at just $3.1 billion in 2026, against $1.5 trillion for infrastructure.
Even at roughly 278 percent growth, it remains a rounding error. A sovereign application is only as defensible as the governed, proprietary data behind it, yet data governance is routinely treated as plumbing in budget reviews.
AI cybersecurity spending, nearly doubling to $51 billion, is a more encouraging sign that buyers are purposefully grasping the protection requirement.
Executive Outlook for AI Applications Growth
With Generative AI in the Trough of Disillusionment, Gartner observes enterprises turning to simpler embedded features from incumbent vendors rather than sweeping transformation programs.
That pragmatism is sensible, but it should not become the whole strategy. CIOs and CFOs should apply a portability test to every embedded AI contract, asking what happens to their data, prompts, and workflows at exit.
At the Applied-AI application layer, sovereignty is less about where a model runs and more about who accumulates intelligence learning. Every workflow refined inside a vendor's embedded AI agent risks compounding the vendor's advantage rather than yours, unless contracts and architecture say otherwise.
The practical path to Sovereign Intelligence is to redirect incremental budget toward assets that compound: governed data, domain-specific models, and application layers the enterprise owns outright.
Infrastructure should be treated as a sourcing decision with exit options, not a default. With spending on track for $3.6 trillion in 2027, the window to choose deliberately to avoid vendor lock-in is narrowing.
So here is the question for your next board discussion. As your AI spending grows, will you own more intelligence, or merely rent more of it? Choose wisely, because the cost and security implications are significant.