The Four Speeds of AI Adoption

AI Has Conquered Code. The Next Challenge Is Reality.

Every technological revolution creates a dangerous illusion. The illusion is that whatever happened first will happen everywhere else.

Today, many in tech including investors look at the explosive growth of AI coding tools and assume the rest of the economy will adopt AI at a similar pace.

I believe that assumption is wrong.

AI is not one market. It is at least four different adoption markets, each moving at its own speed.

And understanding those speeds may be one of the most important investment questions of the next decade.

Speed #1: Consumer AI — Measured in Weeks, relatively small market today

Consumer AI products spread at internet speed. A user downloads an app, uploads a photo, asks a question, or generates an image. No procurement process. No integration. No training. No approval committee. The adoption friction is nearly zero.

ChatGPT reached hundreds of millions of users faster than almost any technology product in history.

This is what software people recognize as a classic viral adoption curve. When the product is useful, adoption can happen almost overnight. However, this adoption curve won’t repeat elsewhere beyond ChatGPT or AI Coder. And more importantly, the whole software industry is only about 2% of the overall GDP today.

Speed #2: Prosumer AI (weeks to months)

These are individual productivity tools purchased directly by highly skilled workers. Including Software engineers, Lawyers, Designers., Scientists.

The user, buyer, and beneficiary are the same person. That collapses the adoption cycle.

example is Claude Code in prosumer market ( mostly). It crossed $2.5B ARR in 6 months or so in early 2026, while the ubiquitous Zoom took 7 years to clear $1B ARR. Once again, this adoption curve won’t repeat for so called AI native applications or agentic AI for enterprises, let along the enterprise migration from SaaS to the AI world.

Speed #3: Enterprise AI — (quarters to years), median market

Here, examples are Copilots, AI agents or Workflow automation. The user, buyer, manager, compliance team, and security team are all different people and they all contribute to the adoption decision. Here, adoption slows dramatically.

Speed #4: Physical AI — Measured in Years, huge and biggest market

Examples are Robotics, Manufacturing, Energy systems, Construction, Logistics, Agriculture. The fourth speed is where the real economy begins. This is where AI stops moving bits and starts moving atoms.

And everything changes. In software, deployment is easy. In the physical world, deployment is the product.

A manufacturing robot does not simply need to work. It must work safely. It must work reliably, many times above 99%. It must integrate into existing workflows. It must pass quality assurance. It must survive edge cases. It must earn trust.

Throughout my years investing in advanced manufacturing and robotics, I have repeatedly seen highly capable technologies require years — not months — to move from pilot projects into production environments.

Not because the technology failed, Because reality is complicated. Operations teams are cautious. Factories cannot afford downtime. Safety standards matter. Existing processes are deeply entrenched. The bottleneck is no longer intelligence. The bottleneck is trust.

A coding assistant can be adopted by a single engineer on Friday afternoon.

A factory robot may require approval from operations, safety, manufacturing engineering, quality control, procurement, and executive management before it ever touches a production line. No robots can escape this adoption curve, None.

One deployment happens in hours. The other happens in years.

Why Investors Keep Getting This Wrong

Software investors naturally view the world through software adoption curves. For thirty years, faster adoption has generally meant larger outcomes. But Physical AI introduces a different dynamic.

The largest markets may not be the fastest markets. The sectors that move slowly often represent the greatest share of GDP. Manufacturing alone is larger than most software markets combined.

Energy is larger. Healthcare is larger. Construction is larger. The economic prize is enormous. The adoption cycle is simply longer. This creates a mismatch between headlines and reality.

Consumer AI creates headlines in weeks. Enterprise AI creates revenue in months. Physical AI creates economic transformation over years.

The Next Decade

The first chapter of AI was about generating information. The second chapter is about generating economic output.

The first chapter rewarded software. The second chapter will reward those who can bridge intelligence with execution.

Many of today’s AI winners will be measured by how quickly they acquire users. Many of tomorrow’s winners will be measured by how deeply they integrate into the real world.

AI has conquered code. The next challenge is reality. And reality moves at a different speed.

Physical AI will not be judged by intelligence alone. It will be judged by economics. Sector matters and is not born equal. A robot assembling GPU systems on a production line can create millions of dollars in value by removing critical bottlenecks in a supply-constrained industry. A robot folding clothes may face a completely different ROI equation. As Physical AI moves from demonstrations to deployments, adoption will increasingly be driven by measurable economic impact rather than technical novelty. My view is that the next phase of the market will be less about what robots can do, and more about where they create undeniable value. That is when the distinction between real businesses and hype will become clear.