What is Happening? Critical Supply Chain Bottlenecks Hit AI Industry
Five leading figures from across the AI supply chain recently gathered at the Milken Global Conference in Beverly Hills to discuss the mounting challenges facing the artificial intelligence economy. The panel included Christophe Fouquet, CEO of ASML; Francis deSouza, COO of Google Cloud; Qasar Younis, CEO of Applied Intuition; Dimitry Shevelenko, CBO of Perplexity; and Eve Bodnia, founder of Logical Intelligence.
The conversation revealed that the AI boom is hitting hard physical limits across multiple layers of the technology stack, from semiconductor manufacturing to energy consumption and data collection. These constraints are creating significant challenges for the industry's continued growth and development.
The Details: Multiple Bottlenecks Constraining AI Growth
Chip Manufacturing Crisis
Fouquet highlighted a critical supply constraint in chip manufacturing, stating his "strong belief" that "for the next two, three, maybe five years, the market will be supply limited." This means major hyperscalers including Google, Microsoft, Amazon, and Meta will not receive all the chips they are paying for.
The scale of this demand is evident in Google Cloud's financial performance. DeSouza revealed that Google Cloud's revenue crossed $20 billion last quarter with 63% growth, while its backlog nearly doubled in a single quarter from $250 billion to $460 billion, demonstrating that "the demand is real."
Data Collection Challenges in Physical AI
For companies working with physical AI systems, the bottleneck extends beyond silicon. Younis, whose company Applied Intuition builds autonomy systems for cars, trucks, drones, mining equipment and defense vehicles, explained that data constraints are equally critical. "You have to find it from the real world," he said, noting that synthetic simulation cannot fully replace real-world data collection. "There will be a long time before you can fully train models that run on the physical world synthetically."
Energy Infrastructure Limitations
Energy consumption presents another significant constraint. DeSouza confirmed that Google is seriously exploring data centers in space as a response to energy limitations, noting that space provides "access to more abundant energy." However, he acknowledged the technical challenges, explaining that space's vacuum environment eliminates convection cooling, leaving only radiation as a heat dissipation method.
Google's strategy involves co-engineering its full AI stack from custom TPU chips through to models and agents to improve efficiency. DeSouza claimed that "running Gemini on TPUs is much more energy efficient than any other configuration" because chip designers know what's coming in the model before it ships.
Einordnung: Why These Constraints Matter for E-commerce
These supply chain bottlenecks have significant implications for e-commerce businesses relying on AI-powered solutions. The chip shortage means higher costs and longer wait times for AI infrastructure, potentially delaying the rollout of advanced automation tools, personalization engines, and intelligent customer service systems.
The energy constraints highlighted by industry leaders suggest that AI deployment costs may continue rising, affecting the economic viability of certain AI applications in e-commerce. Companies planning AI implementations should factor these infrastructure limitations into their strategic planning.
Alternative AI Architectures Emerging
Bodnia presented a different approach through her company Logical Intelligence, which uses energy-based models (EBMs) instead of large language models. Her largest model runs with 200 million parameters compared to hundreds of billions in leading LLMs, claiming it runs thousands of times faster and can update its knowledge as data changes rather than requiring complete retraining.
"Language is a user interface between my brain and yours," Bodnia explained. "The reasoning itself is not attached to any language." For applications requiring understanding of physical rules rather than linguistic patterns, she argues EBMs are more suitable.
Geopolitical Sovereignty Concerns
Younis highlighted that physical AI creates new geopolitical challenges that purely digital AI never faced. "Almost consistently, every country is saying: we don't want this intelligence in a physical form in our borders, controlled by another country." He noted that fewer nations can currently field a robotaxi than possess nuclear weapons.
Fouquet emphasized how semiconductor access affects global AI capabilities. Without access to EUV lithography, Chinese chipmakers cannot manufacture the most advanced semiconductors, creating compounding disadvantages regardless of software improvements.
Praxis-Tipps: Strategic Recommendations for E-commerce
- Plan for Supply Constraints: Factor chip shortages and energy costs into AI implementation timelines and budgets
- Consider Energy Efficiency: Prioritize AI solutions that optimize for energy consumption rather than raw computational power
- Explore Alternative Architectures: Evaluate whether smaller, specialized models like EBMs might suit specific use cases better than large language models
- Implement Granular Controls: Follow Perplexity's approach of implementing detailed permission systems for AI agents operating within business systems
- Focus on Real-World Data: Invest in collecting real-world operational data rather than relying solely on synthetic datasets
Ausblick: Implications for the Future of AI in Commerce
The panel's insights suggest the AI industry is entering a more mature phase where physical constraints and geopolitical considerations will increasingly shape development. For e-commerce businesses, this means a shift from the current period of rapid AI adoption to a more strategic approach focused on efficiency and practical implementation.
The emergence of alternative AI architectures like energy-based models may provide new opportunities for businesses seeking more efficient solutions. Meanwhile, the growing focus on AI agents and automation suggests that the technology will continue evolving from tools that workers use to systems that work alongside human teams.
As Shevelenko described Perplexity's vision: "Every day you wake up and you have a hundred staff on your team. What are you going to do to make the most of it?" This transformation from AI as a tool to AI as digital workforce represents a fundamental shift in how businesses should approach these technologies, despite the current supply chain constraints.
The industry's focus on sovereignty and security suggests that regulatory frameworks will continue evolving, potentially creating new compliance requirements for businesses deploying AI systems, particularly those with physical world applications.