The six AI trends that will define 2026

After a stretch of super-charged investment and unrestrained enthusiasm, the AI market will enter 2026 with a very different temperament. The mood across boardrooms, engineering teams, and investor circles is markedly more demanding, more grounded, and far less tolerant of the same old lofty narratives. Many leaders expect the year to be a broad market correction; a moment when the AI power brokers are expected to show tangible outcomes and justify the scale of their promises. The atmosphere carries a sense of reckoning, as if the industry must now account for the gap between expectation and execution.

Forrester captures this pivot with a vivid metaphor, describing 2026 as the likely moment AI trades its tiara for a hard hat. Their analysis points back to a turbulent 2025 shaped by exuberant ambitions and frantic experimentation. Enterprises, bruised by that experience, are trimming back their theatrical showcases and asking harder questions about the practical jobs AI can perform. You could say that the glow of the honeymoon period has faded; the budgets that once flowed toward innovation theatre – and toward what many executives now describe as billions squandered on superficial ChatGPT wrappers and vapourware – face sharper scrutiny. As with all things, you can only get so far on hype alone; now the market expects something more tangible. Say, a solution that works in real environments, not just elegant presentations or aspirational demos. 

Gartner’s market-cycle outlook affirms this outlook, placing 2026 firmly in the trough of disillusionment. The term often raises alarm, but here it marks something far more constructive: a movement toward maturity. Inflated expectations are settling, low-value applications are losing oxygen, and attention is shifting from technology-led exploration to business-led implementation. The conversations happening now revolve around measurable financial impact, operational reliability, and accountable deployment. The guiding question for the year turns toward profit-and-loss relevance and away from theoretical potential. In this article, we’ll explore the trends that show where the real money, real innovation, and real transformation will happen as the market sobers up.

“Organisational structures today have humans reporting to humans, but that’s changing to a place where you will have AI colleagues reporting to humans and AI collaborating with humans.”

Adit Jain, Leena AI’s Chief Executive and co-founder

The rapid ascent of agentic AI

Autonomous AI agents are set to enter core business workflows, speeding operations and reshaping how work gets done across the enterprise.

The most significant technological shift of 2026 will likely come from a surge in enterprise adoption of agentic AI. Momentum has been building for months, and Gartner now expects 40% of enterprise applications to be intertwined with task-specific agents by year’s end; a steep climb from the single-digit adoption we’re seeing today. The appeal becomes obvious when you watch these systems operate. Instead of waiting for a prompt at each step, agentic AI can interpret objectives, adjust to changing conditions, and carry out multi-stage tasks with minimal oversight, paving the way for the large-scale automation of tasks across entire departments.

Agentic ecosystems

Gartner forecasts that AI agents could soon become a massive market – perhaps the largest among the AI segments. In a best-case scenario, agentic AI could generate close to 30% of all enterprise application software revenue by 2035, totalling more than US$450bn. “AI agents will evolve rapidly, progressing from task and application-specific agents to agentic ecosystems,” says Anushree Verma, Senior Director Analyst at Gartner. “This shift will transform enterprise applications from tools supporting individual productivity into platforms enabling seamless autonomous collaboration and dynamic workflow orchestration.”

Early adopters in areas like software development, legal research, marketing operations, and customer support are already reporting meaningful gains. Boston Consulting Group’s analysis suggests that AI agents can accelerate business processes by 30% to 50%. Customer-facing functions in particular are expected to experience a dramatic change. Gartner predicts that, by 2029, agentic systems will autonomously resolve around 80% of standard customer issues, cutting operational costs by roughly 30%. 

AI for everyone

The conversation around AI in the enterprise has started to shift as a result. While early adopters experimented with replacing workers outright, the smarter approach seems to be augmentation, dramatically increasing what each person can accomplish. Many strategists have started using the term “superstaffing” to describe a future workplace where each employee is supported by an AI chief of staff that manages logistics, priorities, and background coordination.

Leena AI offers an early glimpse of what this future might look like. The company recently introduced what it calls AI colleagues: agents that communicate using natural conversation and work across IT, HR, finance, marketing, sales, and procurement. Each agent is given a full identity; name, phone number, email, Slack and Teams profiles, you name it. The company even gives them favourite sports teams and other personal details to make interactions feel more natural. “Organisational structures today have humans reporting to humans, but that’s changing to a place where you will have AI colleagues reporting to humans and AI collaborating with humans,” says Adit Jain, Leena AI’s Chief Executive and co-founder.

The US$600bn infrastructure boom

A surge in AI demand triggers a record infrastructure investment, intensifying competition between hyperscalers and fast-growing neoclouds.

While investments in AI may become more cautious and results-oriented next year, we are still likely to see a distinct push toward generative AI and autonomous agentic systems; one that will be reflected in capital investment. Spending will likely be targeted towards the physical infrastructure that underpin these systems more so than the tools themselves. Market analyst TrendForce projects that the top eight cloud service providers – Google, AWS, Meta, Microsoft, Oracle, and others in the group collectively referred to as ‘hyperscalers’ – will pour over US$600bn into AI infrastructure in 2026 alone. While this sum may seem colossal, it represents only a 40% jump over the 2025 sum. The broader market is following a similar trajectory to the big eight: indeed, Gartner expects global AI spending to reach close to US$1.5tr this year and to exceed US$2tr in the next. Alongside hyperscaler investments, accelerating enterprise adoption and the spread of AI into consumer hardware such as smartphones and PCs are expanding demand at a pace the industry has rarely seen.

Hyperscalers are treating this spending wave as a defensive play every bit as much as a capacity upgrade. Each is working to build and deploy end-to-end ecosystems that keep customers close before a new class of competitors has the chance to garner a meaningful foothold. Despite controlling the bulk of global cloud spending, hyperscalers have struggled to keep up with AI compute demand since 2024. This has opened the door for a new class of cloud providers built specifically for AI workloads. Often referred to as neoclouds, they concentrate on delivering GPU-backed servers and virtual machines, generally at prices that significantly undercut established players. A 2025 report by Uptime Intelligence found that an Nvidia DGX H100 instance costs about US$98 per hour on-demand from a hyperscaler, while the equivalent from a neocloud costs just US$34 per hour – essentially a two-thirds discount.

Hyperscalers vs. neoclouds

That massive disparity in price we just mentioned? It stems from different business models rather than cheaper hardware. Hyperscalers maintain sprawling platforms with millions of SKUs, covering every imaginable enterprise workload ranging from legacy mainframe emulation to cutting-edge AI. They amortise massive infrastructure investments across decades-old systems alongside brand-new GPU clusters. All that breadth drives up overhead and pushes gross margin targets higher. Neoclouds avoid that complexity by narrowing their focus to a small set of high-value GPU configurations and streamlined services. The lighter footprint reduces their management and R&D burden, allowing the savings to flow back to customers in the form of lower prices.

Even with this momentum, GPU cloud providers generally avoid describing themselves as hyperscaler rivals. Their preferred language positions them as components in an enterprise multi-cloud stack, supplying specialised compute while leaving general-purpose workloads to the established platforms. Interestingly, hyperscalers themselves have started leaning on neoclouds to meet the scale of demand. Microsoft, for example, has committed to spending US$10bn with CoreWeave by the end of 2029. That kind of partnership suggests the lines between categories may blur, and many of these specialised providers will likely get acquired by hyperscalers once the infrastructure crunch eases and the strategic value becomes clear.

“The cost of software is going up, and both the cost of features and functionality is going up as well, thanks to generative AI.”

John-David Lovelock, analyst at Gartner

Generative AI becomes ubiquitous (and more expensive)

Generative AI is becoming a standard feature in enterprise software, bringing new capabilities – and a growing operational cost.

Generative AI will have a nearly ubiquitous presence in 2026. Software vendors have spent the past year integrating AI capabilities into their platforms, and those features are now shipping as stand ard components rather than optional add-ons. You’ll find them embedded in the CRM you’ve used for years; the collaboration tools your team relies on; even the enterprise applications that run core business processes. Whether you planned to adopt AI or preferred to wait, the decision is increasingly being made for you. The catch is that ubiquity comes with a price tag. “Generative AI features are now ubiquitous across software already owned and operated by enterprises, and these features cost more money, aligning with this flush,” says Gartner analyst John-David Lovelock. “The cost of software is going up, and both the cost of features and functionality is going up as well, thanks to generative AI.”

The hidden costs of generative AI

Flexera’s 2026 IT Priorities Report puts hard data to Lovelock’s assertion: 80% of organisations report that their spending on AI applications has increased, with the voracious appetite of AI for compute power driving most of those rising costs. After committing billions to the global AI infrastructure, hyperscalers are now folding those expenditures into their pricing models as a way to recoup their investments. Every single query your team runs through an AI feature costs them money in compute resources, and they’re passing the cost – not to mention a healthy margin – straight through to your monthly invoice. Of course Adobe won’t just eat the cost when you use their AI image generator. Nor does Microsoft absorb the expense when Copilot writes your emails. You do.

The economics of enterprise software are shifting as a result. IT budgeting used to revolve around predictable capital expenses – you bought licenses, renewed subscriptions, and planned accordingly. In 2026, more of that spending will behave like a utility bill, fluctuating based on your usage. If your sales team starts leaning heavily on AI-powered email drafting or your support team automates responses with LLMs, your monthly software costs can spike in ways that traditional license agreements never did. IT leaders now need to track AI model usage with the same scrutiny they apply to cloud egress or network utilisation, and many are revisiting their deployment strategies to avoid runaway operational spending as generative AI continues its climb toward default status across the enterprise stack.

Shadow AI agents create a new cybersecurity crisis

From hyper-realistic phishing powered by multimodal generative tools to shadow AI agents, 2026 will expose companies to greater risks than ever before.

In 2026, AI will take on a more prominent role in the cybersecurity landscape, becoming both the weapon of choice for sophisticated attackers and a growing point of vulnerability for organisations. According to Google Cloud’s 2026 Cybersecurity Forecast, the most immediate risk comes from multimodal generative tools capable of manipulating voice, text, and video at the same time. Google warns that these tools will fuel a new wave of business email compromise and hyper-realistic phishing. By giving threat actors the ability to mimic an executive’s speech patterns and facial expressions on a live call with little friction, generative AI will make deception more difficult to recognise and nearly impossible to counter in the moment.

Another potent threat highlighted in the report comes from shadow agents; that is, unauthorised AI agents deployed quietly by employees who want to automate repetitive work or speed up some of their tasks. Their work happens outside of the familiar (i.e. approved) systems, creating invisible data flows that may expose sensitive information or bypass compliance controls. The idea builds on the long-standing issue of shadow IT, where staff adopt tools without approval from the IT department. Shadow AI agents escalate the risk even further because they can operate autonomously and make decisions without human oversight. “These agents – like any AI model – can experience inconsistent performance, hallucinations, bias, drift or misalignment with evolving compliance standards,” explains Hans Petter Dalen, IBM’s AI for Business Leader for EMEA. “Left unchecked, they can introduce reputational risk, decision-making errors, and even legal violations.”

Businesses under attack

This shift unfolds against a backdrop of intensifying cyberattacks. About 27% of UK businesses say they experienced an incident in the past year, according to the Royal Institution of Chartered Surveyors (RICS). What’s more, among more than 8,000 business leaders surveyed, 73% expect a major disruption within the next one to two years. The recent high-profile breach at Jaguar Land Rover (JLR) shows just how severe the consequences of a modern cyberattack can be.

The attack, which unfolded at the end of August, forced the company to take its computer networks offline, halting highly automated production lines for nearly two months. Analysts estimate the financial cost at £1.9bn (US$2.5bn), making it the most damaging cybersecurity incident in the UK to date. The disruption rippled outward through JLR’s extensive supply chain, affecting thousands of companies and pushing some smaller suppliers to the edge. Industry groups estimate that 5,000 businesses – largely Tier 1 and particularly Tier 2 suppliers – have been caught in the fallout, and a full recovery is not expected until January 2026. Meanwhile, the British government has been forced to intervene, providing financial support to firms whose short-term survivial depends on JLR’s business.

While AI amplifies the threat surface, it is also reshaping defensive capabilities. Security teams increasingly use AI to cut their mean time to detect, respond, and recover from incidents. Real-time analytics can sift through enormous volumes of telemetry, revealing anomalies that would otherwise go unnoticed. AI systems can map suspicious login behaviour, reverse-engineer malware samples, flag unusual network flows, and even forecast likely vulnerabilities by learning from historical attack patterns. The tools evolve quickly, and many organisations see them as essential for keeping pace with attackers who are equally eager to exploit the new terrain.

Physical AI arrives on the scene

Physical AI promises to revolutionise what robots can do, widening automation’s reach and reshaping global labour dynamics.

Industrial robots have been around for decades, doing the same jobs over and over with remarkable precision and speed. They weld car frames, assemble electronics, and move pallets; all tasks where absolute consistency is paramount. Automotive and electronics manufacturers have built entire production philosophies around these machines. But they’ve always had a fundamental limitation: inflexibility. You programme them to do one thing, and that’s all they can do. Introduce any kind of variation, and they come to a sudden, grinding halt. Manufacturing lines were thus designed around their limitations, with humans handling everything that required even basic adaptability.

The emergence of physical AI is starting to change that equation. Instead of being explicitly programmed for a single repetitive task, robots can now learn from simulated or real-world experiences using AI and machine learning. A robot trained this way can adapt to mid-volume production runs or even non-repetitive tasks that would have required human judgment in the past. Because much of the training happens in virtual environments, deployment timelines compress dramatically, and the range of tasks that become economically viable to automate expands. 

Robot takeover

What makes physical AI different from earlier automation attempts is perception. Modern robots come equipped with high-resolution cameras, tactile sensors, and other tools that let them “see” and interpret their surroundings in real time. Powerful AI foundation models tie it all together, integrating vision, language, and action. Give one of these systems a natural language prompt, and it can understand context, make decisions autonomously, and plan sequences of actions. That’s a meaningful departure from the rigidly programmed systems that came before.

It won’t be long before we see some of these machines in action. Tesla, Figure AI, Apptronik, and Agility Robotics have all announced plans to start shipping AI-powered robots by late 2025, with Goldman Sachs projecting global shipment of 50,000 to 100,000 units in 2026. By 2035, annual shipments could reach into the millions of units, with the total addressable market estimated at around five billion units – one for every working-age person on Earth, more or less. Unit economics are expected to settle between US$15,000 and US$20,000 per robot, driven by falling component costs and more scalable manufacturing.

The job apocalypse

The business argument for physical AI is becoming increasingly compelling as new figures come to light. According to the International Federation of Robotics (IFR), adopting these machines could increase productivity by as much as 20-30% by 2030. However, those efficiency gains come with predictable consequences. The IRF estimates that repetitive manual jobs could shrink by 10% to 15% globally within a decade, with low-skill workers in manufacturing, retail, and logistics facing the most exposure. Oxford Economics goes even further in its predictions, suggesting that 20 million jobs could vanish by 2035 because of AI and robotics.

Businesses, meanwhile, stand to benefit substantially. Lower labour costs, reduced downtime, and higher margins translate directly to improved profitability. Some of those gains may flow into hiring in other segments, though probably not at a one-to-one ratio. Robotics also offers a solution to demographic pressures that many developed economies are grappling with, including declining fertility rates and ageing workforces. By automating tasks that are undesirable, dangerous, or low-wage, companies can reduce workplace injuries, improve efficiency, and relieve labour shortages. Of course, whether that trade-off feels equitable will depend largely on which side of the labour market you’re on.

The great ROI reckoning

AI budgets tighten as leaders demand clearer returns, prompting a shift toward growth-focused models and more disciplined investment.

Unfortunately for the big AI players, every hype cycle eventually reaches its breaking point. In 2026, AI appears to be approaching that stage, and enterprise buyers are starting to ask uncomfortable questions about what they’ve actually gotten out of their hefty investments. ROI concerns are finally overpowering the lofty promises of vendors, and companies are pulling back on spending until someone can show them numbers that make sense. Forrester predicts that enterprises will delay 25% of their planned AI spend into 2027; essentially hitting pause on projects that can’t demonstrate clear business value.

The reason for the delay is pretty straightforward: most AI investments simply haven’t been very profitable. Forrester’s research found that only 15% of AI decision-makers reported an EBITDA lift for their organisation in the past 12 months; a dismal success rate given the staggering amounts of capital that have flowed into AI initiatives. Similar patterns appear across sectors. Deloitte’s inaugural Finance Trends report (based on input from more than 1,300 finance leaders) found that only 21% believe their AI investments have produced clear, measurable value so far. Among finance leaders in the early stages of AI adoption, 30% are struggling to justify ROI at all.

Those numbers are forcing a strategic recalibration at the executive level. International Data Corporation’s FutureScape 2026 research suggests the conversation is shifting away from incremental efficiency gains and cost reduction – the safe, easy-to-pitch use cases that dominated early adoption. By 2026, IDC expects around 70% of G2000 chief executives to refocus AI ROI on growth instead, using the technology to reinvent business models and scale revenue without proportionally expanding headcount. Whether that shift produces better outcomes than the first wave of AI spending ultimately remains to be seen. Growth-focused initiatives tend to be harder to measure and take longer to validate than straightforward cost-cutting projects. But when cost-cutting fails to deliver, growth becomes the next logical pitch.