What does the next 12 months hold for AI?
Stephen Kelly, CEO, Cirata
As we enter a new year, AI feels very different from where it was even twelve months ago. The excitement hasn’t gone away, but it has been tempered by something far more important: realism.
After several years of huge investment, bold claims and rapid experimentation, 2026 is shaping up to be the year when AI is judged less on potential and more on performance.
From hype to hard work
Globally, hundreds of billions have been invested in AI, from foundation models to data centers and enterprise platforms. But many organizations are discovering that moving from pilot projects to real, repeatable value is harder than expected.
Industry research consistently shows that a majority of enterprise AI initiatives still fail to reach production, most often because of issues with data quality, integration and governance. As a result, investors and boards are asking tougher questions:
What’s the return? Where is AI actually saving money, reducing risk or improving outcomes?
In the year ahead, funding will continue to flow, but increasingly towards companies that can demonstrate real deployment, measurable impact and operational resilience. This is where the focus shifts from models alone to the foundations beneath them.
At Cirata, we see this first-hand. Organizations are less interested in experimenting with AI for its own sake, and more interested in whether their data is modern, secure, recoverable and fit for purpose.
Is there an AI bubble?
It’s tempting to compare today’s AI moment with the dot-com boom or even the financial crisis of 2008. There will undoubtedly be failures, that’s inevitable with any major technology shift, but the comparison only goes so far.
Unlike the late 1990s, AI already has real users, real customers and real applications embedded across industries, from fraud detection and logistics to software development and customer service. What we’re entering now is not a collapse, but a shake-out.
Some companies won’t survive. Others will emerge stronger, more focused and more sustainable. That’s a healthy phase, and one that rewards strong fundamentals over hype.
The real risks aren’t science fiction
Much of the public debate around AI focuses on distant or dramatic scenarios. But the risks organizations are grappling with today are far more practical.
AI can scale mistakes just as easily as it scales productivity. Poor-quality or biased data leads to poor outcomes. Weak controls increase exposure to fraud, impersonation and security breaches. Over-automation, without human oversight, can turn small errors into systemic failures.
As AI becomes embedded in more critical systems, resilience matters more than ever. Businesses are asking difficult but necessary questions:
- Can we trust our data?
- Can we recover quickly if something goes wrong?
- Can we understand how decisions are being made?
That’s why data governance, disaster recovery and data modernization are becoming inseparable from AI strategies. Without them, progress is fragile.
Jobs, skills and the year ahead
AI is already changing work, but not always in the way people expect.
Most evidence suggests that AI replaces tasks before it replaces jobs. Roles that are highly repetitive and rules-based are under the most pressure, while demand is growing for people who understand data, security, AI operations and governance.
The biggest risk for organizations over the next year isn’t mass unemployment, it’s skills mismatch. Companies that invest in modernising their data and upskilling their people will move faster and safer than those that don’t.
What 2026 will be remembered for
If the last few years were about possibility, the next twelve months will be about proof.
AI is moving out of its hype phase and into its hard-work phase, where value is created through solid data foundations, resilient systems and responsible use. The organisations that succeed won’t necessarily be the loudest or the fastest, but the ones that build AI on data they can trust.
That’s where the future of AI will be decided.
