Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteTechnology only creates broad economic power when infrastructure can deploy it at scale. That is the argument Hans Tung, managing partner at Notable Capital, made in a conversation reported by Fortune on October 13, 2026. Tung’s comparison links Britain’s steel-steamship era and the North’s Civil War railroad network to today’s data-center buildout: in each case, physical capacity determined how widely a breakthrough could be used.
His most concise formulation, as reported by Fortune, is: “The models that win are the ones that are available.” The analogy is useful for understanding why artificial-intelligence investment now reaches far beyond software. It is not, by itself, proof that any country will win a geopolitical contest or that every data-center project will earn attractive returns.
As an Amazon Associate I earn from qualifying purchases.
What the Industrial Revolution analogy is really saying
Steam engines, steel hulls and machine tools were inventions; shipyards, ports, fuel supply and transport links made them commercially useful. Tung applies the same distinction to AI. A highly capable model has limited value if customers cannot obtain inference capacity, if training runs are constrained by electricity, or if networks cannot move data reliably.
That makes infrastructure a bottleneck and a competitive advantage at the same time. Companies and countries need chips, servers, power connections, cooling, buildings and high-speed networks, plus the operational expertise to keep them running. The analogy therefore concerns an enabling system rather than a claim that data centers are literally equivalent to nineteenth-century railways or steamships.
#1 Best Overall
- Used Book in Good Condition
Britain’s steel ships: invention plus industrial capacity
Fortune reports that 80% of the world’s ships came from British shipyards in 1890. The figure is reported by Fortune; the underlying historical dataset was not independently verified in the material available for this article, so it should be read as an attributed historical claim rather than a newly established measurement.
In Tung’s framing, Britain’s advantage came from combining steel-ship technology with the industrial system that could manufacture, finance, supply and operate vessels at scale. The lesson for AI is practical: a technical lead can narrow or widen depending on who can build enough supporting capacity and make it available to users.
Rank #2
Railroads and the value of reach
The railroad comparison focuses on movement. During the U.S. Civil War, Tung contrasts the North’s railroad network with the ability to move people and resources across a large area. Rail lines did not create the underlying resources, but they changed how quickly and reliably those resources could reach the front.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Data centers play an analogous distribution role for computation. They aggregate expensive hardware and electricity, then expose that capacity through networks and cloud services. A model hosted in only a few constrained locations may be impressive but inaccessible. A model supported by geographically distributed facilities can serve more customers, tolerate local failures and reduce latency.
This is an analogy about reach and execution, not evidence that modern compute capacity has already produced a particular military or geopolitical outcome.
What the reported compute figures show—and what they do not
Fortune reports approximately 24 gigawatts of working data-center compute capacity for China and at least 47 gigawatts for the United States, using what it describes as conservative estimates. The figures are time-sensitive: capacity changes as facilities open, retire or are reclassified, and “working” capacity can depend on the definition used. They should therefore be treated as the estimates reported by Fortune, not as a permanent national balance sheet.
Rank #4
| Measure | Reported figure | Qualification |
|---|---|---|
| China working data-center compute capacity | About 24 GW | Estimate reported by Fortune in October 2026; recheck because capacity changes quickly. |
| U.S. data-center compute capacity | At least 47 GW | Conservative estimate reported by Fortune in October 2026; definition and date matter. |
Gigawatts describe power capacity, not a simple count of AI accelerators or the useful output of a model. Utilization, chip generation, networking, access to electricity and the share assigned to AI all affect how much computation customers can actually obtain. A larger headline number is not automatically a larger supply of useful AI services.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Four tests for judging the analogy
| Axis | Historical example | AI-infrastructure equivalent | Reader’s caution |
|---|---|---|---|
| Buildout scale | Shipyards, rail lines and industrial plant required sustained capital. | Data centers require servers, power, cooling, land and network connections. | Construction spending does not guarantee productive utilization. |
| Movement of resources | Railroads moved troops and supplies; ports moved ships and cargo. | Networks move data and expose remote compute to users. | Connectivity, latency and reliability matter as much as nameplate capacity. |
| Execution quality | Manufacturing and logistics determined whether equipment could be delivered and operated. | Procurement, permitting, grid interconnection and operations determine usable capacity. | Delays or power constraints can leave announced projects unavailable. |
| Reach and availability | Transport networks extended the practical reach of industrial power. | Cloud and colocated facilities make models available to more customers and regions. | Access can still be limited by price, export controls, outages or queueing. |
The economic scale—and an important uncertainty
Fortune also reports that a Columbia Business School professor estimated AI infrastructure could account for an average of 3.6% of U.S. gross domestic product each year through 2032. The feature material available here does not identify the professor or the underlying estimate, so the number should be presented as a reported projection, not an independently verified forecast.
For households and investors, the implication is that AI’s economic footprint may show up in power generation, construction, equipment supply, leasing and network services—not only in software revenue. The same breadth creates risk: demand forecasts can be wrong, electricity and permitting can delay projects, and rapidly changing hardware can make facilities or equipment less competitive.
How to read the claim without overreading it
- Separate invention from deployment. A better model matters only if users can obtain affordable, reliable access.
- Check the unit. Power capacity, installed servers, training capacity and customer-serving inference are different measures.
- Date every estimate. The 24 GW, 47 GW and 3.6% figures are reported in October 2026 and may change or depend on definitions.
- Look for execution evidence. Grid connections, permits, financing, delivery schedules and utilization tell more than announced megawatts.
- Keep history in proportion. The ship and railroad examples clarify a mechanism—technology needs distribution infrastructure—but do not establish a guaranteed geopolitical result.
Bottom line for the data-center “railroad” question
Data centers can function like railroads in the narrow sense that they extend the reach of a valuable capability: they turn concentrated computing resources into services that can be delivered across networks. Tung’s broader Industrial Revolution analogy is a warning against judging AI by algorithms alone. The durable advantage may belong to the systems that can finance, build, power, connect and operate enough capacity to make those algorithms available.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
