The cost of agentic AI workflows is forecast to increase more than fivefold by the end of 2028 as users adopt more complex applications of the technology.

As Nvidia and other tech giants push inference and agentic AI as the next stage of the AI wave, Gartner warns that the cost of implementing these systems will rise even as foundation models become cheaper.

The analyst firm has cast its eye over the nascent world of AI agents – systems designed to act independently in pursuit of a goal – and sees multiple challenges ahead.

Leaving aside the substantial security concerns, these software agents are considerably more complex than chatbots.

Gartner believes falling model prices are tempting users to build more complex workflows, whose greater token consumption can outweigh those savings and drive up overall inference costs.

In other words, tokens are becoming more cost-efficient, but those savings are not keeping pace with the rising cost of more advanced AI capabilities. The rate of innovation is outpacing the cost curve, Gartner claims.

“The harsh economics of the inference paradox are exemplified by the differences between a simple chatbot and an AI agent,” says Gartner senior director analyst Will Sommer.

“Where a simple chatbot must read and interpret a query and quickly respond with a probabilistic reasonable answer, an AI agent must constantly reason, negotiate, and question itself,” he explains.

Those processes add up: routing a task to an agentic reasoning model increases inference costs at least fivefold, and potentially by much more as the task becomes more complex.

Securing a return on investment from such advanced AI tools therefore demands either much greater returns than basic models provide or better optimisation of inference, routing, and orchestration, Gartner warns. This could mean assigning each task to the most cost-efficient model capable of handling it.

The move by some AI providers from flat-rate subscriptions to usage-based billing hasn’t helped. Token-heavy workflows can produce runaway costs under the new model.

Perhaps it is no wonder Gartner predicted earlier this year that 40 percent of organisations would demote or decommission AI agents because of problems with the heavily hyped technology.

The analyst biz also cheerily forecast last year that at least half of all generative AI projects would blow their budgets because of poor architectural choices and a lack of expertise, while most attempts to build custom models would be abandoned.

According to a new report from Taiwanese publication United Daily News (UDC) Microsoft is raising the cost of Windows license fees for hardware makers by a much higher amount than normal. According to the report, certain OEMs are seeing licensing price hikes of between 7% and 10%.

Usually, Microsoft raises Windows license fees for OEMs by a low digit percentage, making this 7% to 10% increase much more significant than normal.

“A PC brand executive stated that Microsoft increases licensing fees for Windows OEMs (device manufacturers) every year. In the past, licensing fees mostly increased by single digits each year, but starting in July this year, the average increase has reached 7% to 10%, which is a significant increase compared to previous years,” UDC says.

Of course, the cost of the Windows license fee is usually (at least partially) passed onto the consumer buying a computer, and UDC claims that their source says laptop pricing is already being impacted by these recent license fee increases. Microsoft charges different OEMs their own fees for Windows, so the exact increase in cost will differ for each hardware maker.

Regardless, with the ongoing memory component shortage, PCs have already seen astronomical price increases over the last year, with experts claiming that we haven’t yet seen the worst of it. With Microsoft raising prices higher than usual on Windows licenses, it feels like the Redmond giant is adding extra fuel to the fire.

To be clear, the Windows license fee price hikes only apply to OEMs building hardware that ships with Windows. The cost of buying a license for Windows 11 as a consumer remains unchanged, being $99 for Windows 11 Home and $199 for Windows 11 Pro.

The ongoing component crisis has made computing unaffordable for many people. PCs that used to target the $600-$800 price bracket now cost close to $1,000. And flagship Windows laptops that used to cost $1,200-1,500 are now closer to $2,000. Even Apple has been hit by the shortage, being forced to raise prices by hundreds of dollars across its Mac line-up. Microsoft’s own Surface line was hit especially hard, with the flagship Surface Pro and Surface Laptop being $400-$500 more expensive than the last-gen models.

The UK’s Competition and Markets Authority has opened a fresh investigation into Microsoft, this time over how the company is marketing its Microsoft 365 Personal and Family subscription plans.

A cursory glance at the probe highlights that the CMA is trying to prevent Microsoft from sneaking in Copilot and AI features as justification for recent subscription price hikes. A closer look at the investigation displays a pattern of friction Microsoft runs into with regulators regarding its subscription practices, automatic renewals, and software bundling almost every time it adjusts its services.

The CMA made the investigation public on July 29, 2026, though the case itself was opened two days earlier. According to the CMA’s case announcement, the price gap between plans is concrete. The Classic Microsoft 365 Personal plan costs £59.99 a year, compared to £84.99 a year for the Copilot-equipped version, a difference of £25 annually. The Family plan costs £79.99 a year for Classic versus £104.99 a year with Copilot bundled in.

The CMA’s announcement hinges on Microsoft’s change in January 2025, when the company automatically added new features such as Copilot to existing Microsoft 365 plans with no increase in fees for the remainder of the subscription period. Unfortunately, once the plans were up for renewal, customers who were not paying attention were re-enrolled into higher-priced plans. Customers who were paying attention could avoid the bait-and-switch by selecting a ‘Classic’ plan or cancelling during the re-enrolment period.

This price change was covered when it first happened in February 2025, and even then, Microsoft would not directly attribute the increase to Copilot in their own communications. Personal plans jumped from $6.99 to $9.99 a month in the US, and Family plans went from $9.99 to $12.99, with the Classic option only showing for existing subscribers who knew to look for it during cancellation.

CMA regulators are now examining whether Microsoft provided adequate messaging for customers to understand what changes occurred and how much more they were going to pay despite spending months using features that were initially presented as free.

As Senior Director for Consumer Protection, Hayley Fletcher puts it, “When a business changes its subscription plans, customers need clear and timely information about their options.”

Gartner has slammed AI vendors’ lack of transparency, saying developers are facing sharply increased costs from coding agents.

Since the main AI coding agent vendors have shifted from seat-based licensing to consumption-based pricing, developer teams now face highly variable cost structures.

Developer teams face an emerging problem of escalating costs, said Nitish Tyagi, senior principal analyst at Gartner. AI coding bills were leaping from $20 or $100 to $2,000 to $5,000 per developer per month, while in extreme cases, the bill might hit $20,000 in token charges.

However, software engineering departments get little insight into how token consumption is calculated and billed, making it difficult for them to forecast and control costs accurately. AI coding vendors have not yet delivered built-in features to allow developers to optimise costs in AI coding agents, resulting in cost escalation, Gartner said.

“None of the vendors have incredible features when it comes to cost optimisation,” said Tyagi. Instead, he said vendors were focusing on the concept of “tokenmaxxing” to “boost the high” of token consumption, suggesting that if

developers increase the number of tokens, they will increase productivity gains.

“There is no direct relation between the increase in token consumption and an increase in productivity gains,” he said.

Gartner recommends developer teams optimise token consumption and adopt strategies such as context engineering practices, where software engineers improve the input context provided to AI systems. Another recommended strategy is model routing, where engineering and platform teams direct simpler, high-frequency tasks to smaller models, using frontier models only for complex, high-value work

“All of these things will improve the output quality, and, therefore, will increase the productivity gains as well, so while there is no direct relation between tokenmaxxing and productivity gains, there is a relation between optimisation of token consumption with the output quality,” Tyagi said.

As a result of the lack of cost optimisation tools among vendors and consumption-reduction strategies among users, AI development is in a situation where a developer’s coding agents may cost more than they earn, at least in some parts of the world. Gartner predicts that by 2028, AI coding costs will overtake the average developer’s salary due to rising LLM token consumption and the shift to consumption-based licensing models.

“We’re not saying AI token cost will be higher than every developer’s salary on the planet, because US salaries tend to be higher than in India, for example. But current token costs are already more than most of the salaries in India,” he said.

However, the cost of coding agents does not vary according to where they are consumed, meaning AI coding costs for a developer in India may be equivalent to the salary of an engineer with four to six years’ experience, Tyagi said.

“We’re not saying AI token costs will be higher than every developer’s salary on the planet, because US salaries tend to be higher than in India, for example,” Tyagi said, adding that token costs do not vary by location, and in India they may already be equivalent to the salary of an engineer with four to six years’ experience.

Zendesk has rethought service pricing in an AI-first era beyond the evolution from seat-based to token-based pricing – the company now commits to outcome-based pricing marking one of the most important commercial shifts in the enterprise AI market.

Announced during the company’s annual Relate conference, the new model charges customers only when its AI systems have successfully revolved support interactions.

The company said every resolution would be independently verified with its own dedicated AI evaluation model, meaning that low-value exchanges would be excluded from billing.

The pioneering move reflects growing pressure across the AI industry to prove tangible business value rather than simply showcasing model capabilities for high and often wasted costs. With organisations now becoming more cautious about AI spend, vendors are increasingly being asked to demonstrate measurable ROI instead of charging for access alone, and Zendesk’s shift in pricing model holds the company accountable to pass on ROI to customers.

However, what this means in terms of revenue is less clear. For years, SaaS companies have largely relied on predictable subscription pricing tied to the number of users or licenses.

Zendesk’s approach effectively treats AI agents as digital workers whose compensation depends on measurable outcome, rather than availability. For Zendesk’s customers, it means they’ll only pay when meaningful work is completed.

As Shashi Upadhyay, President of Product, Engineering, and AI, put it in an interview at the event: “Stop thinking of agents as software… start thinking of them as a unit of labour.”

This marks the first time that a company like Zendesk could risk incurring costs without being able to recuperate them from clients, and it marks a major turning point for revenue models in general. The implications could extend far beyond customer service, as companies are expected to experiment with pricing tied to sales conversions, resolved tickets, completed workflows, productivity improvements and other outcomes.

This new strategy also reflects the growing competition within enterprise AI – looking ahead, pricing innovation may become just as important as model performance in acquiring and retaining customers.

So while Zendesk may be among the first to implement such pricing at scale, it offers a glimpse into how AI software could likely be commercialised in the future.

Solvimon says companies are adopting more complex pricing models as AI changes how software is billed. According to the billing platform, usage-based charges have risen sevenfold since 2025.

Its data points to a shift away from flat monthly fees and seat-based pricing towards combinations of subscriptions and consumption-based charges. The fastest-growing software and AI businesses now use nearly five pricing structures on average, up from three over the past year.

That shift is putting pressure on finance teams as AI products create value in ways that do not fit neatly into older billing systems. Businesses are trying to bill for credits, tokens, usage and outcome-based measures alongside standard subscriptions, increasing the risk of invoicing errors and missed revenue.

A Bain & Company analysis of more than 30 established SaaS vendors found that 65% have already introduced hybrid pricing, adding AI usage or outcome metrics to traditional seat-based models. Solvimon argues that the spread of these structures is making billing one of the more fragile parts of the revenue cycle for companies with expanding product lines and more tailored commercial contracts.

Research from MGI found that SaaS companies can lose between 1% and 5% of annual recurring revenue each year through billing leakage caused by errors that repeat in each billing cycle until detected. For chief financial officers, that is bringing greater scrutiny to systems once treated as back-office tools rather than strategic infrastructure.

Oracle customers have been warned to watch for changes in support and pricing as Larry Ellison’s company makes huge datacenter spending commitments to support its AI ambitions.

Advisors in Oracle negotiations said support levels are declining, while costs might increase as discounts disappear. Others have warned of a tougher approach to contract negotiations.

Oracle is aggressively expanding AI datacenter capacity to serve demand tied to customers and partners including OpenAI, xAI, Meta, Nvidia, and AMD. The deal for OpenAI alone is set to be worth about $300 billion over five years.

Widespread reports of layoffs affecting thousands of staff came as investor concerns grew over how Oracle would finance its AI datacenter build-out, while borrowing heavily to support the expansion.

In March, Oracle increased the estimated cost of its fiscal 2026 restructuring plan by $500 million, taking the total to as much as $2.1 billion, as observers anticipated broader job cuts.

Oracle previously raised $18 billion in debt and has since outlined plans to raise far more, as it funds datacenter investments aimed at meeting surging demand from AI model builders and enterprise customers. Ratings agency Moody’s warned of further borrowing and negative free cash flow as Oracle ramps up its AI infrastructure investment.

Nick Walter, CTO of professional consulting at commercial advisory firm House of Brick Technologies, said that there were plenty of anecdotal reports that Oracle support outcomes are getting worse, meaning slower service, less skilled help, and “more Oracle Support responses that were obviously written by AI.”

House of Brick had also seen a recent surge in customers asking for help in reducing Oracle Cloud Infrastructure (OCI) usage or help in negotiation with Oracle over long-term OCI deals, he said. “Many customers who adopted OCI in the last three to five years are suddenly finding out that their renewals are much more expensive because Oracle has eliminated or reduced enterprise discounts, and they are being forced close to public listed pay-as-you-go prices.”

Customers were also seeking help in response to pressure on Java, with audits and sales pushes and “scary-sounding threats.”

“This is not a new phenomenon but pressure continues and Java is the main front in Oracle’s war against their customer’s pocketbook,” he said.

Craig Guarente, founder and CEO of Palisade Compliance, said that Oracle was no longer simply putting pressure on customers to license Java; it was conducting official audits. “We’ve definitely seen in the last few months a return to classic Oracle auditing practices,” he said.

In 2023, Oracle changed its Java SE subscription model, shifting from a per-user or per-processor basis to per-employee. Critics called the move “predatory” as organisations that were using little Java but had a large number of employees could be hit hard by the cost increase. Later that year, research from Gartner showed that costs could be between two and five times greater under the new licensing model for using the same software.

Guarente also warned of price hikes at OCI contract renewal. “If you’re not adequately protected, then we’re getting 9 percent or 10 percent for long term lock in. It all goes back to that original purchase. You had some flexibility in there if you were thoughtful in how you negotiated those contracts. But if you signed the local standard agreement, there’s no price protection.”

Eric Guyer, founder of Oracle advisory and consulting firm Advizant, pointed out in a social media post that the margins in software support were much higher than in datacenters.

“This means the value to Oracle of your software support payments is increasing, desperately so, whereas the value to you is diminishing, and at an increasing rate. I mean this both technically via waning innovation and the more than 8 percent annual inflationary adjustments,” he said.

Customers are also concerned about the long-term impact of the job losses on their relationship with Oracle, Guyer said.

“The trend in license negotiations is only indirectly related to Oracle’s shift for now. The layoffs mean companies have lost one or more of their sales contacts. But for apps or the more strategic long-standing relationships, this means customers are wondering what this loss of institutional knowledge will mean for them. I’m not hearing customers wonder if Oracle is de-prioritising its legacy software business just yet, although I see this as an obvious outcome,” he said.

Does your AI agent count as an employee?

At a recent conference, Microsoft executive Rajesh Jha floated a provocative idea. In a future where companies deploy fleets of AI agents, those agents may need their own identities — logins, inboxes, and even seats inside software systems. If so, AI wouldn’t shrink software revenue. It could expand it.

“All of those embodied agents are seat opportunities,” Jha said, envisioning organisations with more agents than humans — each effectively a user that must pay for a software license, or “seat” in industry lingo.

That’s a radical twist in the SaaS pricing debate rattling companies like Microsoft, Salesforce, and Workday. Investors worry AI could hollow out seat-based pricing, the backbone of enterprise software. If one human can manage dozens of agents, why pay for dozens of licenses?

Jha’s answer: because those agents are the new users. A company with 20 employees might buy 20 Microsoft 365 licenses today. If each employee gets five AI agents, and the workforce shrinks to 10 people, that could still mean 50 paid seats.

Not everyone buys it.

Nenad Milicevic, a partner at AlixPartners, sees the opposite. AI agents will reduce the number of humans interacting with software, slashing licenses. Instead of 20 employees, you might have one person overseeing a handful of agents. That shift would pressure vendors and empower customers to push back on pricing that no longer makes sense.

Milicevic argues the winners will be open platforms. Companies could charge extra for machine-based access, but risk losing customers to software rivals that let agents operate freely.

Which brings us back to the core tension: if AI agents are just extensions of you, charging extra feels like double billing. If they’re autonomous workers, paying for them may be inevitable.

The answer could define the next decade of software economics.

Microsoft has told its channel partners to get ready for a 20 percent price cut for Windows 365 cloud PCs, effective May 1st.

The software giant told partners it will cut prices “to make Cloud PCs more cost-effective for small and medium businesses.”

Microsoft has framed the price cuts as an “update” to the service that will change its performance in two ways: