The artificial intelligence boom has created one of the largest technology investment waves in modern history. Companies are spending hundreds of billions of dollars on chips, data centers, models, robotics, and autonomous systems.

But according to several executives shaping the AI economy directly, the industry’s biggest problems are no longer theoretical. They are operational, structural, and increasingly difficult to ignore.

At the Milken Global Conference in Beverly Hills, five major figures across the AI supply chain discussed where the industry is beginning to break under its own momentum. Their concerns ranged from data shortages and infrastructure bottlenecks to energy constraints and the possibility that current AI architectures themselves may be fundamentally flawed. 

The AI Industry Is Running Into Real-World Limits

One of the clearest themes from the discussion was that AI growth is increasingly colliding with physical-world constraints.

For years, much of the AI race focused on software breakthroughs and scaling compute. But as systems move into robotics, autonomous vehicles, industrial automation, and defense applications, executives say the bottlenecks are shifting.

Applied Intuition CEO Qasar Younis explained that the hardest problem is no longer just processing power. It is collecting enough real-world data to train systems effectively. 

Unlike language models trained on internet-scale text datasets, physical AI systems require actual environmental interaction:

• vehicles driving on roads
• drones navigating terrain
• robots operating in unpredictable environments
• industrial systems responding to real-world conditions

“You have to find it from the real world,” Younis reportedly said, arguing that synthetic simulations still cannot fully replace physical testing. 

That creates a scaling challenge the software-centric AI world has not fully solved yet.

Infrastructure Spending Is Becoming Unsustainably Large

Another concern raised during the conversation was the enormous cost of maintaining the current AI growth trajectory.

The industry is now spending at historic levels on:

• GPUs
• networking hardware
• data centers
• energy infrastructure
• cooling systems
• semiconductor fabrication

Major tech companies are collectively expected to spend hundreds of billions annually on AI infrastructure over the coming years. That expansion is already putting pressure on power grids, semiconductor supply chains, and global energy markets. 

The economics are becoming especially difficult for startups trying to compete against hyperscalers like Microsoft, Amazon, Google, and Meta.

AI development increasingly favors companies with massive capital reserves and direct access to infrastructure.

Some Leaders Are Questioning the Entire AI Architecture

Perhaps the most notable part of the discussion was that some participants questioned whether current AI systems are even built on the right long-term foundation.

The concern is not just that models are expensive. It is that scaling today’s architectures may eventually produce diminishing returns.

That debate has intensified across the AI industry over the past year.

Some researchers believe simply increasing model size and compute power will eventually plateau, forcing the industry toward entirely new architectures or learning methods. Others argue current transformer-based systems still have substantial room to improve.

The Milken discussion reflected growing uncertainty around whether the current AI race is sustainable in its present form. 

Physical AI Is Creating New Data Problems

The conversation also highlighted how AI is moving beyond digital tasks into physical environments.

Companies working on autonomous systems increasingly face problems that cannot be solved purely through internet-scale training data.

Defense startup Scout AI, for example, recently raised $100 million to train AI systems for autonomous military coordination and battlefield operations. The company operates training programs using real vehicles and physical terrain because simulation alone remains insufficient. 

That reflects a broader trend:

The next generation of AI systems may require massive real-world data collection pipelines rather than simply larger internet datasets.

This creates a much more expensive and operationally difficult scaling problem.

The Energy Problem Is Becoming Impossible to Ignore

One issue hanging over the entire discussion is electricity.

Modern AI systems consume extraordinary amounts of power, especially large inference and training clusters.

As companies continue building massive data centers, concerns are growing around:

• electrical grid capacity
• cooling requirements
• environmental impact
• access to sustainable energy

The AI economy is no longer just a software story. It is increasingly an industrial infrastructure story.

That shift may ultimately favor companies capable of controlling both compute and energy supply simultaneously.

AI’s Winners May Depend on Infrastructure, Not Models

Another major takeaway from the discussion is that the AI market may not ultimately be decided by who builds the smartest model.

Instead, the winners may be the companies that solve:

• energy supply
• chip manufacturing
• physical-world data collection
• deployment infrastructure
• real-world operational reliability

That changes the competitive landscape significantly.

It also explains why semiconductor firms, cloud providers, autonomous systems companies, and industrial infrastructure players are becoming just as important as frontier AI labs themselves.

The AI Boom Is Entering a More Difficult Phase

The conversation at Milken reflected a broader shift happening across the industry.

The first phase of the generative AI boom was dominated by excitement:

• chatbot launches
• viral image generation
• massive valuations
• explosive growth
• The next phase appears more complicated.

Now the industry must confront whether its infrastructure, economics, energy systems, and technical foundations can actually support the scale of ambition currently driving the market.

The AI economy is still expanding rapidly.

But increasingly, even the people building it are warning that some of the foundations underneath it are starting to strain.