85% of AI Startups Are Digging in the Wrong Layer

The world of AI is booming, with startups sprouting daily, each promising to be the next ChatGPT, the next NVIDIA, or the next transformational player in a multibillion-dollar industry. But the truth is harsh: 85% of AI startups are digging in the wrong layer. They are misaligned, chasing trends instead of creating value, and inevitably falling into the same traps that every gold rush throughout history has laid.

Think of the California Gold Rush. Thousands rushed to pan for gold, but very few struck it rich. The real winners weren’t the miners but the businesses that sold them tools?—?picks, shovels, and even durable jeans. Levi Strauss built an empire not by finding gold but by enabling the search for it. AI is following this same pattern, with one crucial distinction: the gold isn’t evenly spread, and most miners are digging in barren ground.

The AI Stack: Splitting into Four?Layers

To understand why so many startups are failing to find value, you must first understand how the AI stack is evolving. This stack has four distinct layers, each with its own economics, opportunities, and traps.

1. The Model Layer: The Commodity Trap

This is where the dreamers go to die. Building AI models feels sexy?—?it’s the stuff of headlines, Twitter trends, and VC pitches. Everyone wants to create the next ChatGPT or DALL-E. But the model layer is rapidly becoming a commodity.

• Commoditized Essentials: Like electricity in the early 20th century, AI models are becoming ubiquitous and cheap. They’re essential, yes, but the value is diminishing rapidly.
• Specialized Niches: Only in fields like biotech or genomics, where deep expertise is needed, do models retain significant value.
• Investment Sinkhole: The model layer demands massive investments, but very few are turning those investments into real profit. Most are chasing eyeballs, not dollars.

The allure of building foundational models blinds many founders to an uncomfortable truth: code doesn’t make money; solving problems does.

2. The Infrastructure Layer: A Giants’?Game

If the model layer is hard, the infrastructure layer is nearly impossible for small players. This is the domain of behemoths like NVIDIA, Amazon, and Microsoft. Infrastructure?—?think GPUs, cloud computing, and networking?—?follows the same consolidation pattern as the railroads in the 19th century.

• Economies of Scale: Infrastructure is a scale game. You can’t compete without billions in capital.
• Acquisition or Extinction: Small players get acquired or crushed. There’s no middle ground.
• A Closed Club: Competing in infrastructure requires entering a market that has already solidified around a few dominant players.

For startups, playing here is almost always a losing battle. The gold has been claimed, and the land fenced off.

3. Developer Tools: The Picks and Shovels?Play

This is where the “picks and shovels” analogy comes to life. Developer tools, like Stripe for payments or Snowflake for data, are essential for enabling innovation. But this layer isn’t a guaranteed gold mine either?—?it’s an arms race.

• Brutal Competition: The sheer number of players entering this space ensures that most will fail.
• Massive Winners: For every failure, there’s potential for a massive winner?—?a company that becomes indispensable to developers building in AI.
• Utility over Flash: The winners in this space will solve pain points that aren’t flashy but are critical.

If you’re building here, your odds are better than in the model or infrastructure layers, but don’t underestimate the difficulty. Developer tools are the most competitive layer in the stack.

4. The Application Layer: The Real Gold?Mine

Here lies the most fertile ground for building generational companies. The application layer focuses on solving real-world problems in specific industries. This is where 1,000+ winners will emerge?—?not by wrapping APIs, but by understanding boring, unsexy problems that plague massive verticals.

• Domain Expertise Wins: Technical excellence won’t save you here. Understanding an industry’s nuances and pain points is the real differentiator.
• Solving Real Problems: Success comes from building tools that improve efficiency, save money, or create value in ways that customers truly care about.
• The Opportunity in Boredom: It’s not glamorous, but this layer is where the real money is made.

The history of innovation teaches us a simple truth: the winners aren’t always those who create new technologies but those who apply them effectively to solve enduring problems.

The Value of Data: The Landowners in a Growing?City

In this evolving AI ecosystem, one asset stands out as more valuable than any other: data. Companies that own unique data are like landowners in a booming city. Their assets appreciate over time, without additional effort.

• Appreciating Assets: While algorithms and compute power depreciate or commoditize, unique datasets become more valuable as industries digitize.
• The Real Leverage: Owning exclusive data is the ultimate moat. It’s what separates a fleeting AI startup from a sustainable business.

If you don’t own unique data, you’re renting your competitive advantage?—?and landlords always win.

Lessons from History: AI and the Electricity Revolution

When electricity was new, thousands of entrepreneurs tried to generate and sell it. Most failed. The real winners were those who used electricity to solve specific problems?—?lighting factories, powering homes, enabling mass production.

The parallels with AI are striking. The biggest opportunities today aren’t in building new models or infrastructure; they’re in solving industry-specific problems.

Technology Changes, Economics Don’t

While technology evolves, the fundamentals of business remain constant.

• Commoditization Happens: Every technological breakthrough becomes a commodity over time. AI models are no exception.
• Scale Matters: Giants dominate in markets where scale creates insurmountable barriers.
• Solving Problems Wins: The companies that thrive are those that solve tangible, meaningful problems for real customers.

The hype around AI often blinds founders to these truths. It’s not about having the best technology?—?it’s about using technology to create value.

The Path Forward: Building Generational AI Companies

If you’re an AI founder, here’s your wake-up call:

1. Stop chasing trends. Sexy doesn’t scale. Solve boring problems in massive verticals.
2. Own your data. If you’re not building unique datasets, you’re building on quicksand.
3. Specialize ruthlessly. Domain expertise will outlast technological trends.
4. Think like Levi Strauss. Don’t mine for gold?—?sell the tools, own the data, or solve the problems miners face.

The future of AI belongs to those who see beyond the hype and focus on timeless business fundamentals. Technology may change everything, but economics never do.