Steel ships, railroads, and data centers: A VC takes the Industrial Revolution analogy seriously

Notable Capital’s Hans Tung argues that AI leadership will depend on infrastructure and compute capacity, not just superior models or talent. China has about 24 gigawatts of working data-center compute capacity versus at least 47 gigawatts in the U.S., while the article highlights China’s lead in high-speed rail and shipbuilding as a warning about execution. A Columbia Business School professor estimates AI infrastructure investment could average 3.6% of U.S. GDP annually through 2032, with grid capacity and community opposition among the challenges.

Analysis

The investable implication is a shift in where AI rents accrue: from model novelty toward reliable access to power, grid connections, cooling, and deployable compute. That favors infrastructure suppliers if build schedules convert into orders and cash flow; it does not establish that every data-center project earns an adequate return. The key constraint may be power delivery and permitting, not headline capacity. Reported compute totals do not reveal utilization, chip mix, or how quickly capacity can be energized, so they are weak evidence of a durable U.S.–China advantage on their own.

Over 1–3 months, look for confirmation in hyperscaler capex guidance, utility interconnection timelines, equipment backlogs, and project cancellations—not investor narratives. Over 6–18 months, execution and utilization determine whether infrastructure spending supports durable earnings or produces stranded/underused assets. China’s buildout capability is a strategic risk, but translating industrial capacity into globally competitive AI services also depends on access to advanced chips, software, and customers.

Contrarian view: the infrastructure analogy can overstate scarcity and underweight capital-cycle risk. If compute becomes abundant faster than monetization grows, asset owners and leveraged developers may face pricing pressure even while AI adoption expands. The article supplies a framework, not fresh evidence of company-level earnings changes. ABNB and COIN are mentioned only as prior investments by the interviewee; there is no direct fundamental read-through to either.

Key Decisions for Investors

• Watch, rather than chase, a selective long in power and data-center electrical equipment—e.g., ETN, VRT, or GEV—if company disclosures confirm sustained orders/backlog and customers are securing power. Scale in around earnings or order updates; falsify the thesis on backlog deterioration, project deferrals, or weakening customer capex.
• Avoid treating data-center capacity announcements as proof of returns. Track utilization, power availability, financing terms, and customer concentration before taking exposure to data-center owners or developers; missing project-level economics make a broad long premature.
• Use hyperscaler capex and grid/interconnection developments as 1–3 month catalysts. If capex rises without corresponding utilization or monetization evidence, reduce infrastructure exposure rather than assuming spend benefits every supplier.
• No actionable change to ABNB or COIN on this article alone; reassess only if company-specific disclosures show material AI infrastructure spending, compute dependence, or revenue exposure.