AI Race is Like an Ultra Marathon (Plus Video)
AI Infrastructure, Energy, Tech and the Road Ahead
On May 21, 2026, following the US India Chamber of Commerce AI tech conference in Dallas, I step back to recap energy and AI infrastructure to attempt to see ahead. Two years earlier, I spoke about energy at the inaugural Seeking Alpha summit in June 2024 in New York City. At that point, we were at the tail end of the big energy transition conversations, and data center demand was just starting to emerge, with AI and regular workloads being distinguished and forecast. Few were analyzing the infrastructure layer then, but in 2025 more think tanks, consultancies and the likes of Goldman and S&P became more granular in their analysis.
Regarding the recent energy landscape, in 2024, global oil demand was around 103 million barrels per day and was forecast to grow to 104–105 million barrels—roughly where we are now. The U.S. EIA had a solid forecast at the time, and the projections I’ve written about in recent articles have tracked fairly well.
Today, given the Iran situation, $100 oil is likely the new normal at least through year-end, maybe lower if a deal holds and the best case scenario emerges. Much has been documented here on this Substack and on an investor platform. How markets readjust will matter for the global economy at large and the near-term digital infrastructure buildout.
The AI tech conference featured leaders working in the applications area but also chips and infrastructure layers.
The Five-Layer Cake: Jensen Huang’s Framework
At the Dallas conference, NVIDIA, Amazon, and several AI-focused private firms were represented. An NVIDIA executive referenced Jensen Huang’s “five-layer cake” framework for the AI buildout:
1. Energy
2. Chips
3. Infrastructure — data centers, AI factories
4. Models — Anthropic/Claude, Gemini, OpenAI/ChatGPT, and many others; highly competitive
5. Applications — how firms will deploy and benefit from AI
This serves as a simple framework to understand how the pieces fit together. Within each layer, it is anything but simple.
Disruption, Competition, and the Next Four to Five Years
Since May 2023, I’ve been analyzing NVIDIA and chips through an energy lens and tracking how this technology is diffusing. While it had largely been an NVIDIA story in the markets, competition is increasing and the focus is shifting toward execution: how firms implement AI, a broad area, and capture value from it.
Conference participants highlighted that the next four to five years will be a period of substantial disruption. Looking from 2026 to 2030, the growth in data center demand and power consumption is enormous—and it will be constrained by physical realities: converting hydrocarbons to electrons, building out wind and solar, and upgrading grids.
This is precisely why the NextEra–Dominion merger matters: the power sector will have to become more efficient, just as oil and gas did. Midstream players have already been agile—making deals and connecting with hyperscalers and data centers as new customer segments.
Backgrounder:
Agentic AI, Enterprise Adoption, and Cultural Resistance
A theme that has emerged more visibly over the past year—and that I wrote about in earlier AI infrastructure pieces—is agentic AI. Training was once the dominant story; now agentic AI (AI actively using its capabilities to accomplish tasks) is driving a considerable portion of real-world demand and attention.
In contextualizing one’s approach, one tech leader drew a useful distinction: are you “AI native” or are you “working with AI”? Tech firms on the front lines can exploit AI capabilities more fully at present. For broader enterprise adoption, context matters—and model makers and hyperscalers need proximity to the enterprises they serve.
For example, I attended an Anthropic event in Dallas—clearly an effort to establish presence in a major enterprise market. Differentiation in how firms create and capture AI demand will grow, and education will be a large part of the process.
Regarding the efficient use of AI: one leader made the practical point that AI should solve a problem people will pay to have solved. Tokens are not cheap, and resource requirements are real. The related strategic questions—relevance, competitive differentiation, and cultural readiness—are what will separate winners from laggards. Another leader noted that cultural resistance within organizations is a meaningful drag on adoption and diffusion.