In this photo illustration, the American multinational investment bank and financial services company J.P. Morgan (NYSE: JPM) logo seen displayed on a smartphone with an Artificial intelligence (AI) chip and symbol in the background. (Photo Illustration by Budrul Chukrut/SOPA Images/LightRocket via Getty Images)

SOPA Images/LightRocket via Getty Images

Some trends about the future of of AI in enterprise are starting to gel, based on our collection of data from hundreds of cases studies. Recent news events have only served to magnify these trends.

Just a few of the trends to watch: The largest enterprises are more interested in protecting their proprietary data than handing it over to the largest frontier AI models. They are favoring open AI models with proprietary software customization, rather than off-the-shelf solutions. And their pilot projects tend to be very focused on specific problems.

Finally, despite the hype, real-world adoption of AI in global enterprises remains difficult at best. This adoption is by no means nonexistent, as reported in some poorly designed studies that seem to attract journalists, but the movement is nuanced and it’s pivoting toward generating critical business outcomes.

From Bubble to ROI

The question that everybody is waiting to answer is: Will it all pay off? With trillions of dollars in infrastructure being deployed, that means the AI boom must return trillions in ROI to pay off.

The reality is that much of this spend is speculative and potentially out of scale. Futuriom is estimating that approximately $3 trillion in capex will be allocated over the next three years. To put things in perspective, $1 trillion a year is 40% of the total estimated profits of the S&P ($2.3 trillion for 2026).

MORE FOR YOU

This investment will only be worthwhile if AI services and workloads are widely adopted by major enterprises as well as consumers, a small number of which are using it today.

“This is about the deployment of AI infrastructure becoming real,” said Kartik Srinivasan, the CEO of an AI infrastructure provider Napatech. “ Once enterprise starts deploying AI infrastructure, that's when it has hit reality.”

So Where Are the Enterprise Gains?

The consensus from many sharp investment minds and this particular analyst is that in order to deliver the goods, AI adoption has to move from circular financing deals with NVIDIA to widespread adoption by consumers and enterprise users. In a recent IBM study, only 11% of respondents say they’re completely prepared for the scale of AI agent deployment.

So far, enterprise boards and CIOs have grappled with struggles in data governance, security, and token budgets, with the last grabbing the headlines recently when Uber announced its full token budget had been spent in four months.

“Enterprises are trying to reduce their Anthropic bills,” Haseeb Budhani, Cofounder and CEO Rafay Systems, an AI infrastructure software provider, recently told me. "The enterprises are looking at their own infrastructure. The only option is open-source models."

The conversation is no longer about whether companies should experiment with artificial intelligence but about which organizations can turn AI into business value. That is the central message of Futuriom’s latest report, The Futuriom 25: The Top AI-Forward Enterprises, which profiles 25 AI-forward enterprises and draws lessons from more than 200 case studies across industries (preview free, subscription required for entire report).

From the analysis in our report, the companies winning at AI are not treating it as a bolt-on tool or a pilot project. They are using it to redesign workflows, improve decision-making, and scale proprietary advantage.

The 25 firms we selected were based on results, not geography or vertical. Taken together, the list suggests that the most successful AI efforts share a common trait: they are linked to concrete business outcomes. Leaders of the projects are moving from efficiency metrics to outcome metrics. These metrics are used to show how AI changes revenue, risk, customer satisfaction, operational resilience, and product velocity.

This shows that AI has already entered its second act. The first wave was defined by chatbots, copilots, and experimentation. The new wave is defined by agentic systems, vertical use cases, and infrastructure embedded into core operations. In our view, companies are past the point of asking how to start using AI. The harder question is how to scale it without losing control of data, governance, or human oversight.

Proprietary Data Rules the AI World

One of the report’s clearest conclusions is that proprietary data is becoming a strategic asset. The strongest enterprise AI programs are often built around internal data that gives models more relevance than general-purpose tools can provide. By training models on proprietary information, companies can streamline internal workflows, support employees more intelligently, and make decisions that reflect their own operating context. This is not just about smarter software; it is about using AI to amplify the knowledge already sitting inside the business.

In data-sensitive industries such as financial services and healthcare an over-reliance on automation, requires a delicate approach. AI can accelerate diagnosis, surface anomalies, and handle repetitive service tasks, but it can also produce false outputs, weaken oversight, and create security vulnerabilities if deployed carelessly. The companies that scale successfully are the ones that build governance, human review, and security into the deployment model from the start.

The industries moving the fastest in AI are the retail, financial services, insurance, healthcare, and manufacturing verticals. These sectors are not just experimenting with AI in isolated pockets. They are using it to transform core business functions. Retailers are adopting predictive personalization and agentic commerce to shape everything from discovery to checkout. Banks and insurers are deploying AI for fraud detection, service automation, and personalized recommendations. Manufacturers are combining AI, robotics, and digital twins to improve quality, optimize production, and reduce downtime. Healthcare firms are using AI to accelerate drug discovery, support clinical workflows, and improve patient matching.

The report returns to a pattern: Successful enterprise AI is rarely a single model. It is an integrated system. The strongest examples embed AI directly into existing platforms and workflows. Alphabet’s fleet reporting assistant, for example, converts large sets of operational data into charts and summaries inside a portal managers already use. American Express applies AI across fraud detection, expense management, and commercial analysis. BMW uses digital twins, computer vision, and AI-powered robotics to support factory operations and quality control. Each example points to the same lesson: Adoption rises when AI is invisible in the best possible way—available where work happens, not off to the side as an extra step.

Infrastructure is also shaping outcomes. While many enterprises are building proprietary systems, they still rely heavily on hyperscalers such as Google, Amazon, Microsoft, and NVIDIA to provide the foundational layers.

Sovereign Infra Control on the Rise

Futuriom’s research also detects a broad movement toward sovereign AI, where companies keep models and sensitive data in private clouds or regional infrastructure rather than relying on fully public environments. That shift reflects growing concerns about privacy, compliance, and leakage. For many enterprises, strategic independence is becoming as important as model performance.

This is especially relevant in regulated industries. JPMorgan Chase uses AI to improve fraud detection and employee productivity at scale, while maintaining tight control over access. Pfizer is using AI to speed drug development and improve the quality of documentation. Johnson & Johnson, Roche, and Massive Bio are applying AI in ways that could materially change how medical products are developed, tested, and matched to patients.

Our data also reflects the reality of most AI transformations: The biggest constraint is often not model quality but data readiness. Many organizations still struggle with fragmented, incomplete, or unreliable data. That undermines confidence in outputs and limits the ability of AI systems to operate at scale. The promise of generative AI can only go so far if the underlying business data is messy. The enterprises seeing real gains are the ones that have invested in data pipelines, governance, and reusable architecture rather than one-off demos.

The more successful organizations appear to be those that use AI to augment people, not simply reduce headcount. American Express highlights this with its human-centric support tools. Nedbank’s virtual assistant handles routine calls so live agents can focus on complex issues. Many of the strongest case studies show the same principle: AI works best when it frees people to do more valuable work.

Enterprise AI is evolving every day, and we are seeing a shift toward how companies can leverage their proprietary data edge with unique projects. The winners are not necessarily the companies with the most advanced models. They are the ones with clear use cases, strong data, the disciplined governance.