AI Is Crossing the Chasm

Summary: AI is surging from early-adopter novelty to everyday utility. Yet this transition is uneven across countries and use cases. In some cases, AI has already crossed Geoffrey Moore’s famed chasm between visionary early adopters and the pragmatic early majority, whereas in other cases, AI diffusion is much slower.

To “cross the chasm” is Geoffrey Moore’s term for transitioning a product (or a technology like AI) from being limited to a small set of enthusiasts to becoming mainstream. Did the camel succeed in making it across the chasm? Watch my music video about this article (YouTube, 4 min.) to find out. (GPT Image-1)

Global consumer AI usage has reached unprecedented scale by 2025, with an estimated 1.8 billion humans having used AI tools and around 600 million using them daily. This marks a dramatic leap from just a year ago, signaling that AI is no longer confined to tech enthusiasts but has entered daily routines worldwide.

In Boston Consulting Group’s mid-2025 study of 11 countries, 72% of workers reported using AI regularly at work. This is up sharply from the prior year, aligning with McKinsey’s global survey, which found that 78% of organizations used AI in at least one business function in 2024, up from 55% the previous year. Such a rapid jump, from roughly half of businesses to over three-quarters in just one year, indicates that many firms have moved from piloting AI to broadly deploying it.

Many other recent studies show that AI is in the process of crossing Geoffrey Moore's famous chasm of technology adoption, moving from early adopters to mainstream use. However, “AI” isn’t one market; it’s a swarm of adoption curves moving at different speeds. Some uses have landed on the far side of the chasm. Others are still dangling by their fingertips.

Geoffrey Moore’s “chasm” concept: there are substantial differences between the early adopters of a new technology or product and the early majority of customers. Innovations must meet many more requirements and have much higher usability to make the crossing and appeal to the mainstream. (GPT Image-1)

“Mainstream” doesn’t mean “uniform.” Some countries and sectors are well into the early majority or even late majority adoption (using AI as a matter of course), while others are just reaching the inflection point. These gaps, between and within countries, will have significant implications. Those who successfully embrace AI will reap economic and social benefits (from higher GDP to better healthcare and education), while latecomers lose competitiveness and miss out on these benefits.

Geoffrey Moore’s “Chasm” Theory

Building on Everett Rogers' innovation diffusion theory, Geoffrey Moore identified a critical gap in the technology adoption lifecycle that has proven fatal to countless promising innovations. His insights, first published in his book “Crossing the Chasm” in 1991, are relevant for thinking about the diffusion of AI today.

The technology adoption lifecycle consists of five distinct customer segments, each with unique characteristics and motivations:

1. Innovators, comprising roughly 2.5% of the market, are technology enthusiasts who embrace new products for their own sake, often tolerating bugs and incomplete features for the privilege of being first.

2. Early Adopters, representing about 13.5% of the market, are visionaries who see strategic opportunities in new technology and are willing to take calculated risks to gain competitive advantages.

3. The Early Majority, a pragmatic 34% of the market, seeks proven solutions that integrate well with existing systems.

4. The Late Majority, another 34%, adopts technology only when it becomes an established standard or necessity.

5. Laggards, the remaining 16%, resist technological change and adopt only when forced by circumstances or when the technology becomes invisible within other products.

The last 16% of the population will take a long time to convert into AI users. Not that we should give up on them, but they are not the goal for now. Currently, focus on the Late Majority for those AI use cases that have already crossed the chasm and the Early Majority for those AI use cases that are about to break through. (GPT Image-1)

Moore’s crucial insight was identifying the chasm between Early Adopters and the Early Majority. This gap represents a fundamental shift in customer expectations and buying behavior. While Early Adopters are visionaries seeking revolutionary change and competitive advantage, the Early Majority consists of pragmatists looking for evolutionary improvements and reliable solutions. This transition is treacherous because the marketing strategies, sales approaches, and even product features that appeal to Early Adopters often fail to resonate with the Early Majority. Many promising technologies die in this chasm, having exhausted their resources serving the enthusiast and visionary markets without successfully transitioning to mainstream adoption.

Feature requests from users in the Innovators segment will often doom a product for mainstream use, as more features increase complexity beyond the ability (or willingness) of most users to comprehend. (GPT Image-1)

Enthusiasts tolerate terrible design. They enjoy complexity. They will forgive inconsistent interfaces, catastrophic errors, and the complete absence of documentation. They will spend hours learning a system because they gain status from mastering the technology first. When companies rely on feedback from these users, they are optimizing for the wrong criteria. Early Adopters demand features, customization, and power. If you listen to them, you will develop a bloated product that is impossible for normal people to navigate.

The early adopters enjoy complexity, which gives them a sense of mastery. Such featuritis is baffling to mainstream users. (GPT Image-1)

Central to crossing the chasm is Moore's concept of the “whole product.” While Early Adopters are willing to piece together solutions from multiple vendors and fill gaps with their own resources, the Early Majority demands complete, ready-to-use solutions. The whole product encompasses not just the core technology but everything required for the customer to achieve their desired outcome: training, support, complementary products, integration services, and established procedures. For example, when personal computers crossed the chasm, the whole product included not just the hardware and operating system, but also business software, printers, networking capabilities, training programs, and technical support infrastructure.

Mainstream use requires a whole product that includes everything needed to solve real-world problems. You can’t expect Early Majority users (and especially not Late Majority) users to cobble together a solution from bits and pieces that may or may not work. AI often doesn’t offer this complete package yet. (GPT Image-1)

Reviewing the data about AI adaptation as of now shows that this technology doesn’t have a single fit within Moore’s adaptation model. In some cases, AI has already crossed the chasm and become mainstream. In other cases, it’s stuck (for now, that is!) in the early adopter stage.

AI contains multitudes.

That said, given that the first useful AI (GPT 4) wasn’t released until March 2023, it’s remarkable that any AI uses have crossed the chasm. AI is moving at unprecedented speed. It’s also improving at an incredibly rapid pace, meaning that most of those AI uses that have not yet become mainstream will probably cross the chasm in 2026 or 2027.

Already Crossed the Chasm

• Countries: China, USA, India, Gulf States. 61% of U.S. adults report using AI, and 81% of Chinese respondents make the same claim. Despite widespread AI adoption in all these countries, only China has a high public optimism score at 83% (people expecting AI to be more beneficial than harmful). The USA has a disgracefully low AI optimism score of only 42% which does not bode well for this country’s future prospects in the world economy. (At least the American optimism score recorded in 2025 is up by 7 percentage points from the abysmal 35% in 2022.) India has extremely high AI use among professionals (92%), but low use among the general public. 59% of Indian companies report active AI adoption, which also represents a crossing of the chasm.

Despite the prejudice of some outsiders, the Gulf States are already high tech. (I particularly recommend visiting The Museum of the Future when in Dubai, and not just for Shaun Killa’s stunning architecture.) They will likely accelerate faster than many other regions, due to strong AI optimism scores and plentiful energy, which is required for AI training and inference. (GPT Image-1)

• Company type and industry: Large firms, tech sector, finance and banking, e-commerce, manufacturing.

• Business functions: Marketing and sales automation (customer segmentation, personalization), software development (DevOps automation, code generation).

AI can often help with customer segmentation, resulting in an immediate increase in sales, which is one reason for the common AI implementations in sales departments. (GPT Image-1)

Once you’ve segmented the customer base, AI allows for easy adjustment of the tone of voice for marketing copy. Marketing departments are another business function with many well-funded AI projects. (GPT Image-1)

Software development is currently the number-one AI use case, with around $10 billion in annual revenue. (GPT Image-1)

• Consumer functions: Chatbots, voice assistants, research before purchases, healthcare information (but not actual care).

AI now has mainstream adoption for use in product research: it is much easier for consumers to ask AI for a comparison summary than it is to extract the same information from multiple websites, often with deviant and low-usability design. (GPT Image-1)

Widespread use isn’t limited to young techies: nearly 45% of U.S. Baby Boomers (ages 61–79, such as yours truly) have tried AI tools, and 11% use them daily. (In fact, AI has special advantages for old users.)

Seniors are frequent AI users. (GPT Image-1)

The breadth of use cases shows that AI has become a general-purpose tool for consumers, akin to a new “utility” like search engines were 20 years ago. Creative AI tools for image generation (from Midjourney to Grok’s Imagine feature) first became mainstream in niche communities (artists, designers) but are now used by regular folks for social media content or fun (though not as universally as text chatbots).

Laggards

• Countries: Europe (both EU and UK), Japan. Even Denmark, which is one of the most advanced countries in Europe regarding AI, only has 28% of firms using AI. Countries like Romania, Bulgaria, and Turkey are sitting around 5%. (Compare with 65% in the US and 80% in China.)

The predominantly negative media coverage of AI, such as the demeaning “stochastic parrot” myth, has led to low AI optimism scores among populations in many countries, which will suffer as a result, as they fall behind more optimistic regions, like most of Asia. (GPT Image-1)

• Company type and industry: Small firms, tech sector, finance and banking, healthcare, education.

• Business functions: Logistics, R&D, operations.

• Consumer functions: autonomous vehicles, AI agents that take action on the user’s behalf.

Making AI agents assume the desired level of agency on the user’s behalf (neither more nor less) may require significant tuning, which presents a usability barrier to adoption. (GPT Image-1)

In advanced countries, healthcare providers are on the verge of crossing the chasm. They aren’t true laggards anymore, but don’t yet have mainstream use. In the US, 66% of physicians report using some AI, but only 12% use it to assist in diagnosis, which would seem to be required to claim real AI healthcare adoption.

Only 12% of human physicians currently use AI to help in diagnosis, despite research showing that AI often outperforms them. (GPT Image-1)

Research shows that patients perceive AI doctors as exhibiting more empathy than human physicians, and AI’s diagnostic accuracy is also higher in many cases. However, healthcare customers are currently unable to access AI providers outside of these research settings. Likely progress will be slow in the medical domain. (GPT Image-1)

Education is another in-between segment, with most students and teachers utilizing AI for their personal work, but few educational institutions systematically employ AI to aid in teaching (other than pioneers like the private Alpha School).

AI is already being heavily utilized in education for support tasks, such as creating lesson plans, but is not yet commonly used for actual instruction. (GPT Image-1)

Why Some Areas Are Stuck at the Wrong Side of the Chasm

The laggards need a “Whole Product” for using AI, but this is rarely offered yet: AI solutions in these fields are rarely off-the-shelf. They are often bespoke projects requiring significant in-house data science expertise, extensive data preparation, and complex integration with legacy operational technology (OT) systems. This forces the customer to assemble the “whole product” themselves, a task only a visionary organization is willing to undertake, and which often fails (95% of internal corporate “build-it-ourselves” AI projects are reported to have failed). A majority of organizations report that their data is not “AI-ready,” a fundamental barrier to creating a whole product.

Many legacy enterprises have not yet transformed their masses of data into an AI-ready format. (GPT Image-1)

In countries where people are generally optimistic about technology and risk-tolerant (such as the United Arab Emirates or India, where more than 90% express approval of AI), individuals are more likely to try new AI services, and businesses are more likely to deploy them without fearing public backlash. By contrast, in societies with skepticism or technophobia among the public or even within corporate culture, adoption stalls or at least proceeds cautiously. Japan’s case hints at this: a tradition of valuing perfection and a cautious corporate environment meant that many were hesitant to use a “sometimes wrong” AI chatbot, delaying its mass adoption. Similarly, Germany historically has been tech-cautious (e.g., initially skeptical of ride-sharing, strict on privacy), which might explain its slow AI rollout.

High-income countries have the resources to invest in AI, but legacy systems and bureaucracy hinder their progress; emerging economies have fewer legacy systems and can leapfrog with new technology. India, Brazil, Middle East thus have high AI usage growth.

Connecting something new, such as AI, to legacy systems often means that you’re in for a world of pain. Thus, the prominent “Cry” button, which you’ll be using often. Unless you’re lucky enough to reside in a country with fewer legacy systems. AI can make leapfrogging real. (GPT Image-1)

Additional factors that explain why different AI uses are on different sides of the chasm today:

• Risk/Consequence Level: Low-risk uses (entertainment, writing drafts) adopted faster; high-risk uses (driving, surgery) slower.

AI rollout is delayed for use cases with high penalties for errors. This, despite the likely much higher benefits that could be realized worldwide in the case of faster adaptation: kill one patient to save thousands? That’s the very real AI version of the “runaway trolley” track-switching dilemma. (GPT Image-1)

• Complexity and Supporting Tech: Uses that only need an app or cloud (chatbots, image generation) soared; those needing complex hardware (robots, specialized equipment) lagged.

Integrating AI with physical operations that involve complex hardware can be challenging. (GPT Image-1)

• Economic Incentive: Customer service and marketing AI (cost-saving, revenue-generating) has been widely adopted; experimental or nice-to-have uses without proven ROI (some HR or creative AI tasks) have been slower.

AI improves every year, meaning it can accomplish more at a lower risk, driving the ROI curve in a favorable direction. Only the highest-ROI AI projects are common now, but over time, more and more AI projects will have high ROI. (GPT Image-1)

Customer support is a domain where ROI is already sky-high for AI, and so many companies are using AI for simpler customer problems, escalating only demanding cases to their remaining human agents, who can be expected to become fewer and fewer every year. (GPT Image-1)

• User Experience: Tools with high usability (conversational AI, simple API integrations) have mainstream acceptance, whereas low-usability AI tools that require significant training (advanced analytics platforms, some enterprise software) face slower uptake except in tech-centric firms. As AI implementations improve their usability, they should break into the mainstream, accelerating the next wave of adoption among the early majority who lack high-tech skills.

Low usability presents a formidable barrier for most users to climb, preventing bad design from going mainstream. (GPT Image-1)

• Availability of Data: Use cases that could leverage existing large datasets (like web text for language AI, or customer data for recommendation systems) matured faster. Areas that lacked data or had fragmented data (like certain medical AI requiring diverse patient data) take longer to reach effective performance, hence slower adoption.

• Ethical/Social Acceptance: If a use case aligns with social values or at least doesn’t overtly challenge them, it faces less resistance. AI art and AI writing, while resisted by certain legacy artists, have broad acceptance by end-users. But AI in law enforcement or surveillance raises societal hackles and thus sees pushback (e.g., facial recognition tech is mainstream in China, but stalled for now in the West).

AI-generated content is accepted by consumers who’ll watch a movie if it’s good, regardless of whether the special effects were created with AI or the traditional method. Soon, audiences will probably extend the same acceptance to fully AI-created actors and music performers: if a song is good, who cares how it was made? (GPT Image-1)

There is significant social pressure against the use of AI to dynamically adjust the prices customers are charged, despite evidence from Uber showing that surge pricing can effectively match supply with demand (when prices increase, more drivers hit the road to pick up the surge in riders). (GPT Image-1)

Transition Problems

U.S. businesses are still maturing in their adoption. Many employees use AI tools spontaneously (in BCG’s study, over half admitted they’d use AI tools even if not officially authorized, a rising “shadow AI” trend), but only 13% of companies have broadly integrated AI into their formal workflows. The U.S. has achieved habitual usage but not uniform integration: AI is common, but not yet fully trusted or systematically utilized in every organization. Top leadership is pushing for more: three-quarters of U.S. (and global) executives believe AI agents will be vital to future success.

It is extremely common for individual employees to use their personal AI accounts to perform work tasks, even in companies that ban or limit the use of AI. These employees don’t tell anybody, but they cash in on the raises or bonuses that result from their improved productivity. Because of such “shadow AI,” most companies have much greater AI use than their official data show. (GPT Image-1)

The current phase is thus transitionary: the U.S. has largely crossed into early majority adoption among both consumers and enterprises, but is now grappling with how to extract full value. Overcoming organizational inertia (through training, workflow redesign, and addressing employee fears) will determine how quickly U.S. adoption moves into late-majority, truly transformative use.

Overcoming inertia in large organizations requires strenuous pushing, but can be done. (GPT Image-1)

Larger organizations have more resources to adopt AI and are typically ahead of smaller firms. Early evidence shows big companies are twice as likely as small ones to implement best practices for scaling AI (like having roadmaps, training, and centralized AI teams). This aligns with OECD findings that large firms and knowledge-intensive sectors were adopting more rapidly than small and mid-sized companies.

Numerous studies document the lack of AI training in most organizations, yet training is a required prerequisite for a new technology to go mainstream. (GPT I-1)

Within the enterprise world, multinationals and tech giants may be approaching the late-majority stage (with multiple AI projects scaling). In contrast, many small businesses are still in the early-adopter phase with some experimentation and few implementations. For example, a local retailer might only now be considering using an AI chatbot on their website, whereas Amazon has AI integrated throughout logistics, pricing, and more. This gap means the “mainstreaming” of AI in business is uneven: it’s common to use AI somewhere in the org, but fully AI-driven companies are still the exception (though growing in number).

Beachhead strategy for the path across the chasm: While mainly stuck on the wrong side of the chasm, some visionary companies are establishing beachheads by following Moore’s recommended “D-Day” strategy: targeting a single, highly specific, high-pain niche and dominating it. This is a way to gain early success experience, serving as a foundation for subsequent AI expansion.

The beachhead strategy is the establishment of a secured position (akin to gaining a foothold on enemy territory) that enables subsequent advances. Find a niche where AI can help your company now, and secure it first. (GPT Image-1)

• BMW did not try to build an entire AI-powered factory. Instead, it targeted the specific problem of quality control on its assembly lines. By using AI-powered computer vision for real-time inspections, it reduced vehicle defects by up to 60%. This is a perfect beachhead: a single, high-value problem solved with a contained, measurable AI solution.

Quality control on BMW assembly lines: AI reduced vehicle defects by 60%. (GPT Image-1)

• Shell focused on the critical challenge of predictive maintenance. It has now deployed its AI platform to monitor over 10,000 assets, processing 20 billion sensor readings per week to generate 15 million daily predictions that prevent costly equipment failures.

AI can often predict equipment failures before they happen, leading to substantial cost savings in industrial operations, as happened at Shell. (GPT I-1)

2025 is the Dawn of AI’s Early Majority Era

AI has, in many contexts, already crossed the chasm from a futuristic novelty to an everyday necessity. The data shows a remarkable story: within just two years, a majority of businesses and a large fraction of the global population began using AI in some form. This is a testament to how quickly a transformative technology can spread in the modern digital age. Countries like the US, China, and India are now living in an AI-infused reality, where AI increasingly touches both work and personal life. Many other nations are not far behind, rapidly learning and adopting the innovations of the leaders.

However, crossing the chasm is not the end of the journey; it’s the beginning of the mainstream phase, which brings its own challenges. Now that AI is in the hands of mainstream users, the focus must shift to usability, reliability, and meeting the needs of pragmatists rather than just visionaries. For businesses, this means leveraging widespread AI usage to achieve genuine productivity and innovation gains by reengineering processes and upskilling employees. For consumers, it means integrating AI in ways that truly enhance daily life while respecting privacy and values.

Most AI services have settings that allow paying subscribers to opt out of having their data used for training purposes, but people need to trust that these settings will keep their information private, which is a harder sell. (GPT Image-1)

Enterprise AI adoption in 2025 is broad but shallow: most companies have crossed the threshold of trying or using AI, meaning the concept of AI in business is mainstream. Yet, the maturity of adoption (using AI to its transformative potential with new workflows) is still in progress. Some leading industries and firms have fully embraced AI (gaining competitive edge and productivity boosts), while many others are still learning how to integrate it effectively. The next few years will likely determine which companies convert today’s early-majority adoption into tangible performance gains versus those who stagnate. The ones that succeed will be those that train their people, redesign workflows, and build trust and governance around AI: in essence, they will carry the early majority over the finish line to true mainstream value realization.

A complete redesign of workflows for AI is necessary to unlock its full ROI potential and elevate a company’s AI maturity from the pilot stage to the fully-scaled stage. (GPT Image-1)

2025 marks a tipping point: AI is no longer just for early adopters. The conversation has shifted from “will AI be adopted?” to “how can we maximize its value now that it largely has been adopted?” The world’s major economies are navigating this new normal, each at its own pace, but the trajectory is clear. If the past year’s adoption rates are any indicator, by this time next year, even more of the laggards will have fallen in line, and AI’s presence will be even more ubiquitous, though often seamlessly blended into the background of everyday tools. In many ways, AI’s crossing of the chasm in 2025 echoes the spread of the Internet in the late 90s: a rapid, transformative mainstreaming that opens up vast new possibilities and challenges. The countries and organizations that recognize the factors behind successful adoption and actively address the gaps will be best positioned to thrive in AI’s early majority era and beyond.

Music video about this article (YouTube, 4 min.)

About the Author

Jakob Nielsen, Ph.D., is a usability pioneer with 42 years experience in UX and the Founder of UX Tigers. He founded the discount usability movement for fast and cheap iterative design, including heuristic evaluation and the 10 usability heuristics. He formulated the eponymous Jakob’s Law of the Internet User Experience. Named “the king of usability” by Internet Magazine, “the guru of Web page usability” by The New York Times, and “the next best thing to a true time machine” by USA Today.

Previously, Dr. Nielsen was a Sun Microsystems Distinguished Engineer and a Member of Research Staff at Bell Communications Research, the branch of Bell Labs owned by the Regional Bell Operating Companies. He is the author of 8 books, including the best-selling Designing Web Usability: The Practice of Simplicity (published in 22 languages), the foundational Usability Engineering (28,863 citations in Google Scholar), and the pioneering Hypertext and Hypermedia (published two years before the Web launched).

Dr. Nielsen holds 79 United States patents, mainly on making the Internet easier to use. He received the Lifetime Achievement Award for Human–Computer Interaction Practice from ACM SIGCHI and was named a “Titan of Human Factors” by the Human Factors and Ergonomics Society.

· Subscribe to Jakob’s newsletter to get the full text of new articles emailed to you as soon as they are published.

· Follow Jakob on LinkedIn.

· Read: article about Jakob Nielsen’s career in UX

· Watch: Jakob Nielsen’s first 41 years in UX (8 min. video)