Nine in 10 Companies Will Miss AI-First. Gartner Just Published the Reason.

Gartner’s analysts stood up at the firm’s Data & Analytics Summit in Sydney on June 16, 2026, and handed data chiefs a number that sounds like ambition and reads like a warning: more than one in 10 enterprises will be AI-first by 2030.

Flip it, and it's nearly nine in 10 that will not be AI-first.

Gartner frames its six data and analytics trends as things to weigh over the next two years. They work better as a diagnosis. Each one names something broken in the average enterprise, and each one carries a date.

Six trends, one diagnosis

Start with the geopolitics, because no CDO asked for it. Gartner treats sovereign AI as an external reality that lands on the roadmap whether you invited it or not. Nation-states want control over their own AI capability and less dependence on foreign suppliers, and localizing data and analytics control is part of how they get it. This produces questions the board was not asking the CDO two years ago. Where does the training corpus physically sit? Whose courts can compel access to it? What happens to your model strategy when a cloud provider’s regional footprint becomes somebody’s foreign policy instrument? Data residency used to be a checkbox that legal handled. It’s an architecture decision now.

Then comes the trend most enterprises will skip and most will need. Gartner predicts that by 2029, explicitly modeled business decisions will be five times more trusted and 80% faster than ungoverned ones. Most enterprise decisions are not modeled anywhere; they live in a pricing analyst’s spreadsheet, a stored procedure written in 2019, and a threshold someone hardcoded before the acquisition and never documented. That was survivable when a human clicked approve. It stops being survivable when an agent executes 40,000 of those decisions before lunch, and nobody can reconstruct why.

The plumbing trend is the challenging one. Adoption of data streaming for agentic AI sat under 15% in 2025. Gartner expects it to pass 60% by 2028. That is a backbone rebuild on a three-year clock, and plenty of shops still run the nightly batch job that has quietly defined their business since before the pandemic. An agent quoting a customer from inventory counted 14 hours ago is not intelligent. It’s confidently wrong, at machine speed, in writing.

Gartner’s answer to rising data complexity is more agents: point them at the pipelines to profile, reconcile and flag drift before a human notices. The circularity is obvious, and the advice is probably still right, because no headcount plan closes that gap. Budget for one caveat. An agent that does not understand your data will produce tidy garbage faster than your team ever produced messy garbage, which makes continuous monitoring the actual control.

The governance platform trend is the least glamorous and the easiest to defend to a CFO. Regulatory complexity is multiplying across jurisdictions while autonomous agents multiply inside the company. Assurance by spreadsheet does not survive that collision.

The last trend is a quiet confession. Gartner expects 40% of enterprises to be using GraphRAG by 2029, pairing knowledge graphs with large language models. Standard retrieval augmented generation cannot handle complex, context-rich queries, which is exactly the kind enterprises keep asking. The fix is the semantic modeling work that got defunded three years ago to pay for a chatbot pilot. Ontology is back, wearing a better jacket.

What the 10% will have done by 2030

Look at the six trends together, and one word runs through all of them. Control. Sovereignty is control over where data lives. Decision governance is control over why a system acted. Streaming is control over when the truth arrives. Semantics is control over what the data means. AI-first, in Gartner’s usage, describes an enterprise that knows its own estate well enough to let something autonomous act inside it.

Which brings back the number. GPU spending will not decide who makes the 10%. The deciding work is the kind nobody live-streams: mapping decisions, killing batch jobs, rebuilding the semantic layer, arguing with legal about residency. Unglamorous, expensive, invisible in a keynote.

Everyone else gets to be a case study. In the other column.

Image credit: iStockphoto/beast01