The AI Investment Paradox — A 1962 Book Explains Why Billions Don't (Yet) Deliver

The Paradox

$37 billion invested in generative AI in 2025 alone. A 3.2x increase from the year before [1]. And yet, 95% of businesses have yet to see measurable ROI from their AI investments [2]. 42% of companies abandoned their generative AI initiatives entirely [3].

How is this possible? How can an industry attract this much capital while delivering this little return? I’ve been tracking this disconnect for a while — the “disappointment gap” between AI hype and actual outcomes is real, and it keeps widening.

The answer isn’t in a 2026 analyst report. It’s in a book from 1962.

A Sociologist Saw This Coming

In 1962, Everett M. Rogers published Diffusion of Innovations, a book that would become one of the most cited works in social science. Rogers studied how new ideas spread through populations. Not how they’re invented. How they’re adopted.

His core insight: adoption isn’t a technology problem. It’s a social process. And social processes take time, regardless of how much money you throw at them.

Rogers identified five groups that adopt any innovation in sequence:

The cumulative adoption follows an S-curve, slow start, rapid acceleration, then tapering off. The critical insight: you can’t skip stages. The early majority won’t move until they see early adopters succeed.

As Andy Jassy famously put it: “There is no compression algorithm for experience.” [5] The quote was originally about AWS’s head start in cloud computing, but it applies perfectly to innovation diffusion. You cannot compress the social process of adoption. Early adopters need time to build real implementations, fail, learn, and share — and the early majority needs time to observe those results before committing. No amount of investment accelerates this human learning curve. You can make it smoother, but you can’t skip it. I explored this idea earlier — and it keeps proving true.

Where AI Sits on the Curve

Through Rogers’ lens, the picture becomes clear:

• Innovators (2.5%). AI researchers, ML engineers building foundation models. They’ve been in for years. Done.
• Early Adopters (13.5%) — People building real-world implementations, publishing what works and what doesn’t. We’re just starting to show concrete results. I recently used a coding agent for eight hours straight without writing a single line of code — preparing customer meetings, researching papers, assembling expense reports. That’s what early adoption looks like: messy, personal, and surprisingly productive.
• Early Majority (34%). Watching. Evaluating. Waiting for proven patterns before committing budget and organizational change.
• Late Majority + Laggards (50%). Not even considering it yet.

This means the vast majority of potential users, the ones who would generate ROI for early infrastructure investors, haven’t even started implementation. The money is flowing into supply (models, compute, infrastructure) while demand-side adoption is still in its infancy.

Geoffrey Moore built on Rogers’ work in Crossing the Chasm (1991), identifying a dangerous gap between early adopters and the early majority. Many technology products die in this chasm. AI is standing right at its edge.

This also explains a pattern I see with customers: most are still building single point AI solutions — replacing one component while keeping everything else the same. That’s natural for early adopters. But the real ROI — the kind investors are waiting for — comes from system-level solutions that rethink entire workflows. And those require the early majority to get involved.

Why Diffusion Is Slow — It’s Not the Technology

Rogers identified five attributes that determine how fast an innovation spreads. Let’s map them to AI in 2026:

Relative Advantage — Clear for some use cases, unclear for many. “What exactly should I use AI for?” is still the most common question I hear from customers.

Compatibility — Low. AI requires new skills, new workflows, new data practices. It doesn’t plug into existing processes easily. This is a major friction point.

Complexity — High. Model selection, prompt engineering, RAG architectures, evaluation frameworks, guardrails, cost management. The early majority won’t tolerate this level of complexity.

Trialability — Improving. ChatGPT and Amazon Bedrock playgrounds help, but enterprise-grade trials with real data are still hard to set up.

Observability — Low. Most successful implementations are internal. Results are hard to quantify and even harder to share publicly. The early majority can’t see others succeeding.

That last one, observability, is the biggest bottleneck. Rogers was clear: people adopt innovations when they can observe others benefiting from them. Right now, most AI success stories are locked behind corporate walls.

How to Accelerate Your Own Diffusion

Rogers’ framework doesn’t just explain why adoption is slow, it points to specific levers you can pull. But the levers are different depending on where you sit.

If You’re an Individual: Become the Early Adopter in Your Organization

You don’t need permission to start. Rogers found that early adopters succeed because they make judicious adoption decisions, they don’t chase every shiny tool, they pick one real problem and solve it.

• Pick one workflow that frustrates you and apply AI to it. Not a side project, something you do every week. The ROI becomes obvious when it’s personal.
• Document what you learn. Write it down, share it in a team meeting, post it on LinkedIn. You’re building observability, the attribute Rogers identified as the biggest adoption accelerator. Every time you share a real result, you move someone in the early majority closer to trying it themselves.
• Accept the learning curve. You will waste time. You will hit dead ends. That’s the experience that can’t be compressed. But it compounds, each failed experiment teaches you what works for your context.

If You’re a Company: Design for the Majority, Not the Enthusiasts

Most corporate AI strategies are designed by innovators for innovators. That’s why 42% of initiatives get abandoned [3]. The early majority has fundamentally different needs:

• Start with the problem, not the technology. The early majority doesn’t care about model architectures. They care about “will this make my Tuesday morning less painful?” Frame AI adoption around business outcomes, not technical capabilities. I once spotted a foot-operated door handle in Barcelona — someone had designed for real behavior instead of imaginary users. The same principle applies: build AI for the workflow people actually have, not the one you wish they had.
• Reduce complexity ruthlessly. Don’t give teams a platform and say “go explore.” Give them a specific use case, a pre-built solution, and a two-week pilot. Rogers’ research is clear: complexity kills adoption. Every decision you remove accelerates it. I explored this tension between agency and control in From Chaos to Control — the early majority needs guardrails and predictability, not infinite flexibility.
• Find your internal opinion leaders. McKinsey’s research confirms what Rogers predicted: organizations that scale AI successfully find their most enthusiastic internal adopters and put them at the center of the cultural story [6]. These aren’t necessarily your most senior people, they’re the ones others trust and watch.
• Measure leading indicators. Don’t wait for revenue impact. Track adoption rate, use case pipeline, skill development, and time-to-first-value. Deloitte’s 2-4 year ROI timeline [4] is normal for a model shift, but you need early signals to stay the course.

If You’re a Thought Leader or Technology Partner: Bridge the Chasm

This is where companies like AWS, and people in roles like mine, carry a specific responsibility. Rogers called this the change agent, someone who bridges the gap between the innovation and the people who need it.

• Make trialability effortless. Sandbox environments, workshops, immersion days with real data — not slides. I’ve seen more adoption decisions made during a two-hour hands-on session than in months of presentations. The Generative AI ATLAS is one example — a holistic, open-source navigation aid from business strategy down to implementation code.
• Connect adopters to each other. Peer networks accelerate diffusion faster than any vendor engagement. When a CTO hears “we tried this and it worked” from another CTO in their industry, that’s worth more than any reference architecture. Communities of practice, user groups, and practitioner events are the diffusion networks Rogers described.
• Publish prescriptive guidance, not just possibilities. The early majority doesn’t want “here are 50 services you could use.” They want “here’s how to solve your specific problem, step by step.” Opinionated defaults, reference architectures, and managed services that abstract complexity, these are the tools that move people across the chasm.

The Time Gap Is Normal

Here’s the uncomfortable truth for investors: Deloitte’s research shows AI ROI payback takes 2-4 years [4], while typical technology investments are expected to pay back in 7-12 months. This isn’t a failure, it’s the natural pace of diffusion for a model shift.

At Amazon, one of our Leadership Principles is Ownership: “Leaders think long term and don’t sacrifice long-term value for short-term results.” [7] This principle was written long before the current AI wave, but it captures exactly what Rogers’ research tells us about innovation adoption. The companies that will benefit most from AI are the ones that invest now with a long-term lens, building capabilities, training teams, running experiments, rather than demanding quarterly ROI from a technology that’s still crossing the chasm.

Rogers’ S-curve tells us the acceleration is coming. But it requires the early majority to start moving, and the early majority requires proof. That proof is being built right now, by early adopters publishing their learnings, by managed services reducing complexity, by communities making success visible.

The billions being invested aren’t wasted. They’re building the infrastructure for a wave that hasn’t crested yet. The question isn’t whether adoption will happen. Rogers’ 60 years of research says it will. The question is what you’re doing to accelerate it.

Are you making your AI successes visible? Are you reducing complexity for the next wave of adopters? Or are you waiting for someone else to cross the chasm first?

If you’re navigating your own AI adoption journey, whether you’re just getting started or looking to scale what’s working. I’d love to hear from you. I’ve had the privilege of working with customers across industries on exactly these challenges, and the building blocks AWS provides today make it possible to turn ideas into reality faster than ever. Let’s talk about where you are on the curve and how to accelerate from there. Reach out. I’m happy to support the process.

Sources

[1] Menlo Ventures — Enterprise AI Investment 2025

[2] MIT Media Lab Project NANDA — AI ROI Study

[3] Janea Systems — Enterprise AI Maturity Gap 2026

[4] Deloitte — AI ROI: The Paradox of Rising Investment and Elusive Returns

[5] Andy Jassy — “No compression algorithm for experience”, CNBC / AWS re:Invent

[6] McKinsey / Forbes — Culture Will Make Or Break Your AI Strategy

[7] Amazon Leadership Principles — Ownership

About the Author

Stefan Christoph is a Principal Solutions Architect at AWS, focused on agentic AI, media & entertainment, and helping builders move from demo to production. He writes about AI architecture, developer productivity, and the future of software.

This is a personal blog. Opinions expressed here are my own and do not represent the views or positions of my employer.

Learn more →

• AI
• Innovation
• Strategy
• Adoption