Oct 24, 2026 · The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch · 5 min read

Energy Bottlenecks and the $10T AI Inference Market

Naveen Rao argues that energy, not compute, is the primary constraint for AI scaling. He predicts a $10 trillion inference market within a decade and advises investors to shift focus from software to hardware and physical intelligence.

Business Strategy Technology Investment Market Trends

The takeaways

Energy Is the Primary AI Constraint. Building one gigawatt of power costs $50 billion, exceeding semiconductor fab costs. Investors must prioritize energy infrastructure over pure compute expansion to avoid supply shocks.
Inference Market Reaches $10 Trillion. The inference market will expand to $10 trillion within ten years. This growth is driven by the shift from pre-training to real-time, high-volume intelligence delivery.
Hardware Outperforms Software for Returns. Pure software investors face diminishing returns as SaaS margins compress. Hardware and physical intelligence layers offer superior margin capture and long-term defensibility.
Verification Becomes the New Bottleneck. AI shifts the challenge from production to verification. Companies that own the verification layer control the automation pipeline and ensure output reliability.
Robotics Surpasses Agentic Coding Value. Physical robotics will become 100 times larger than agentic coding. The market for dynamic physical problem-solving dwarfs virtual task automation in total addressable value.
Open Models Dominate Token Traffic. Open-source models will handle 80% of token traffic within three years. Frontier models will capture 80% of margins by focusing exclusively on high-complexity, high-value tasks.

The Energy-Intelligence Nexus

The AI industry is transitioning from a compute-centric paradigm to an energy-constrained reality. Naveen Rao, founder of Unconventional AI, identifies energy supply as the critical bottleneck, noting that building one gigawatt of power infrastructure costs approximately $50 billion. This figure exceeds the cost of building a semiconductor fab, highlighting the physical and financial barriers to scaling AI. As data centers exhaust existing grid capacity, the market faces a hard ceiling on growth unless new generation methods are deployed. This shift necessitates a re-evaluation of infrastructure investment strategies, moving away from purely digital assets toward tangible energy solutions.

Market Dynamics and Margin Shifts

The inference market is projected to reach $10 trillion within a decade, driven by the ubiquity of AI agents. However, the value chain is restructuring. Rao argues that pure software and SaaS models are losing margin as intelligence becomes commoditized at the lower end. Conversely, hardware and physical intelligence layers are capturing the lion's share of value. This is because the conversion of energy to intelligence is the core economic engine, and those who optimize this conversion retain the highest margins. Investors who remain focused on traditional software metrics are mispricing the future of the industry.

Strategic Implications for Investors

The post-Transformer era requires a new investment thesis. The Transformer architecture is not the final form of AI; it is an inefficient intermediary. Future breakthroughs will come from co-designing models with hardware, optimizing for energy efficiency rather than raw speed. This creates opportunities in specialized chips and physical intelligence. Furthermore, the verification layer is emerging as a critical control point. As AI generates vast amounts of unverified data, the ability to validate outputs becomes a primary source of competitive advantage. Companies that master verification will own the automation layer, securing long-term defensibility in an increasingly automated economy.

Conclusion

The next decade of AI growth will be defined by physical constraints and hardware innovation. Energy is the new oil, and hardware is the new software. Investors must adapt their frameworks to value physical infrastructure, verification capabilities, and the massive, yet underappreciated, market for robotic intelligence.

Key insights

Energy infrastructure is the primary constraint on AI scaling, with a cost of $50 billion per gigawatt. This physical limit dictates the pace of model deployment and data center expansion.

Infrastructure →

Impact: Investors must prioritize energy-adjacent assets and infrastructure plays over pure compute to mitigate supply-side risks in the AI sector.
The inference market is projected to reach $10 trillion within ten years, driven by the shift from training to real-time intelligence delivery. This represents a massive expansion of the addressable market.

Market Size →

Impact: Companies focused on efficient inference delivery will capture significant market share, while those stuck in pre-training optimization may miss the growth curve.
Hardware and physical intelligence layers are capturing the majority of margins, while software and SaaS margins are compressing. The value of AI is shifting from code to physical execution.

Value Chain →

Impact: Pure software investors face diminishing returns, while hardware and robotics investors are positioned to capture the highest value in the AI stack.
Verification is becoming the new bottleneck in AI development. The ability to validate AI outputs is more valuable than the ability to generate them, creating a new layer of defensibility.

Operational Strategy →

Impact: Companies that own the verification layer will control the automation pipeline, securing long-term competitive advantages in AI-driven workflows.
Robotics will become 100 times larger than agentic coding, as physical problem-solving offers a vastly larger total addressable market. The market for dynamic physical intelligence is currently underappreciated.

Sector Outlook →

Impact: Investors should shift focus from virtual automation to physical robotics, where the long-term value creation potential is significantly higher.

Action items

Reallocate investment portfolios to prioritize energy infrastructure and hardware assets over pure software and SaaS companies. Focus on companies that optimize the conversion of energy to intelligence.

Impact: This shift aligns with the primary constraint of AI scaling and captures the highest margin opportunities in the current market cycle.
Develop internal verification frameworks for AI outputs to ensure reliability and trust. Invest in tools and processes that validate AI-generated content before deployment.

Impact: Owning the verification layer creates a defensible moat and ensures that AI automation delivers consistent, high-quality results in production environments.
Evaluate the potential of robotics and physical intelligence as a long-term investment thesis. Focus on companies building models and hardware for dynamic physical problem-solving.

Impact: Positioning early in the robotics sector allows investors to capture the massive value creation expected in the physical AI market over the next decade.
Monitor the adoption of open-source models for routine tasks and reserve frontier models for high-complexity problems. Optimize cost structures by leveraging the 80/20 split in token traffic.

Impact: This strategy reduces inference costs significantly while maintaining high performance for critical tasks, improving overall operational efficiency.
Shift investment metrics from market validation to technical progress for deep-tech startups. Focus on the rate of problem-solving rather than immediate revenue or user adoption.

Impact: This approach aligns with the nature of deep-tech innovation, where the market is guaranteed but the technical execution is the primary risk and driver of value.

Quotes

“The inference market is going to be a $10 trillion market within 10 years.”

“I actually think the hardware is the place to be over the next 10 years.”

“I do think we will be in a post-Transformer world.”