TLDR
- Generative AI adoption by enterprises is rising fast, with most business leaders already reporting positive ROI, according to a new Wharton study.
- Culture and talent, not tech, are now the biggest adoption barriers.
- Companies plan to step up spending on generative AI, and expect to hire more junior staff as adoption expands.
A new Wharton study finds that most U.S. companies are shifting from dabbling in generative AI to deploying it broadly in enterprise AI solutions. However, they worry that such dependence could erode core human skills. In a surprise, they also expect AI to lead to the hiring of more junior staff.
The 2025 Wharton–GBK gen AI adoption report, based on a survey of more than 800 senior decision-makers at U.S. firms with at least 1,000 employees, shows that 82% of business leaders use generative AI at least weekly and 46% use it daily, up from 29% in 2024.
At the same time, about 72% of enterprises say they now track formal return on investment (ROI) metrics for generative AI, and roughly three in four expect positive payback from their initiatives.
The findings − published by Wharton Human-AI Research and GBK Collective in an annual report titled “Accountable Acceleration: Gen AI Fast-Tracks into the Enterprise” − offer a counterweight to a widely cited recent MIT study that found 95% of generative AI pilots have a zero return.
The Wharton study found a rapid increase in enterprise AI adoption in 2025. “That would be the biggest change” over findings in prior years, said Stefano Puntoni, co-director of Wharton Human-AI Research and a marketing professor who co-authored the study, in an interview with The AI Innovator.
“People are excited about generative AI. They think this is not just hype,” he said. While there are concerns about issues such as labor and data privacy, “overall, there is a sense that this technology is powerful and it is going to affect everything.”
The study revealed that the most popular gen AI tool at work is OpenAI’s ChatGPT, with a 67% share, up 5 percentage points from a year ago. Second is Microsoft Copilot, at 58%, up 6 percentage points. Google’s Gemini is third, with 49% but the fastest gaining with a 9 percentage point uptick. Meta AI is next at 37% and internal custom chatbots are at 29%.
The report tracks how generative AI has moved from experimental pilots to “everyday AI” woven into office workflows. Half of the top use cases directly boost employee productivity, such as coding, data analysis, writing and editing, summarization and slide or text preparation.
Enterprise AI ROI moves center stage
The Wharton–GBK team waited until the third year of the study to ask about ROI, on the assumption that earlier years were too early for meaningful numbers. The answers in 2025 surprised them: 74% already see positive ROI and four in five expect positive ROI in the next two to three years. Small- and mid-sized firms got to positive ROI faster, but larger firms are closing the gap.
“I had expected a rosy picture,” Puntoni said, “but I wasn’t expecting such strong results, both in terms of the measurement and the results.”
Recent headlines have focused on trillion-dollar data-center build-outs by hyperscalers such as Amazon, Microsoft and Google. Markets have sold off as investors worried that such unprecedented data center spending means an AI bubble is forming. But Puntoni said enterprise demand is there.
“I don’t think that those investments make any sense unless you think that gen AI is going to have a transformational effect, not only on efficiency, but also on effectiveness,” he said, meaning that gen AI is not going to be used only for cost-cutting, but also for value creation and innovation.
“If you don’t believe the latter, then I think it makes no sense to spend a trillion” dollars on data centers, Puntoni said. “I don’t think the kind of productivity growth you’re going to get is going to justify that kind of money.”
“To me, it’s about value creation,” he continued. “How do we use this technology to create a new customer experience, a better customer experience, new product categories, new industries?”
Wharton data shows that companies are prepared to spend on gen AI for enterprise AI solutions over the next few years: 88% expects to increase spending in the next 12 months, up 16 percentage points from a year ago, while 62% expect to raise budgets by at least 10% over the next two to five years. One out of 10 companies plans to divert funds from legacy IT, HR and workforce programs.
A chunk of that gen AI budget is going to internal R&D, consulting and human capital – not just model access or software licenses.
“It’s not like the Microsofts and the Googles are (the only ones) making the money,” Puntoni said. “Actually, the majority of the budget is going to other things. Some of it is internal R&D, and … lots of money for consultancies. Some of it is for upskilling and onboarding − basically investing in human capital. To me, gen AI is very much an HR story, as it is also a technology story.”
Human capital becomes the bottleneck
The constraint to gen AI adoption is no longer tools, but people, according to the study. Culture, as well as recruiting people with advanced AI skills and training existing employees effectively, are among the top obstacles to capturing AI value.
“Everybody’s worrying about jobs, but in fact, senior executives tell us they cannot find the people they need,” Puntoni said. “Change management is hard. So when you have a new technology coming in with a very powerful set of capabilities, it’s going to take time to figure out, first, what to do with it, and second, how to develop the talent that you need to be a good complement to this technology.”
Executives are also increasingly aware of a subtler risk: skills atrophy. According to the report, 89% of leaders say gen AI augments employees’ skills, but 43% worry it could lead to declines in skill proficiency over time.
Puntoni used his own writing process as an example. “For me, thinking is very much bound to language,” he said. “So if I outsource all my writing to AI, I’m also outsourcing all my thinking, and that’s not a good idea. So what I do is that I use gen AI, but only for editing. … If I start with gen AI from scratch, the blank page, then I do not engage in thinking as much and that’s not good for me.”
He urged leaders to treat AI adoption as a long-term human-capital challenge for which they must plan strategically. They also must prepare for employee career paths changing as capabilities evolve. If junior employees are not adequately trained in AI skills because gen AI is running operations, who will replace older workers who are retiring or leaving?
Puntoni’s advice: “Carve out time for your most talented employees to use gen AI to do something amazing rather than something cheap.”
Are junior jobs really at risk?
One finding in the Wharton data flips the narrative on the impact of gen AI on junior staff. The predominant belief today is that companies will use gen AI for basic tasks and thus hire fewer young, entry-level workers. The Wharton study confirms this view, with a caveat.
While 17% say gen AI will lead to fewer junior hires, 49% said it would lead to more junior roles as long as these are related to AI. “There are more people thinking it would lead to more jobs than people who believe it would lead to fewer jobs. And those who believe it’s going to lead to more jobs tend to believe it’s going to be predominantly for junior positions too,” Puntoni said.
“Basically, what we are finding is that junior positions are expected to change,” he added. “The net result might actually not be a bad one for junior people. … My advice for junior talent is to make sure they are ready to make a difference in a workplace where gen AI is playing an important role.”
For Puntoni, the long-term picture of an AI-infused enterprise is less about chatbots and more about a quiet shift in what human expertise means.
“We’re going to see that increasingly, in many functions within a company, the job of the human expert is not going to be doing things, but deciding what needs to be done,” he said. “A lot of the doing is going to be done by AI agents, but what these agents should do is going to remain a human question for a bunch of reasons. One is accountability. Second is values matter. Third, … a very competent person is better than AI at that job.”
Puntoni called this the “art critic model”: AI generates the output; a highly skilled human judges whether it is good enough, how it should be improved, and what should be attempted next.
“Software development has been a bit of the canary in the coal mine there,” he said. “If you are a software engineer today, your role is becoming almost like a vice president-level role in a software company where your job is not to write the code, but to decide what code needs to be written.”
Some developers may relish the move toward more strategic, high-impact work. Others “will really struggle,” he warned, if they “love writing code and … now find a situation where their competence is no longer that valuable.”
As gen AI becomes more deeply embedded into operations, 2026 could be the start of an inflection point for enterprise AI.
“2026 could be the turn from accountable acceleration to performance at scale − where today’s ROI metrics, playbooks, and guardrails let enterprises rewire core workflows, deploy agentic systems, and reallocate budgets toward proven returns,” according to the study.
