America was right about railroads. Northern Pacific still went bankrupt.
The technology can be real and the demand can arrive, and the company building it can still fail. For Northern Pacific, that happened because it had to pay for the track before freight traffic generated enough revenue to cover the cost.
AI infrastructure has a similar shape: data center developers have to borrow to secure power and chips before tenants have signed leases for that capacity.
For lenders and insurers backing data centers, the question isn’t whether AI is real. That seems settled, just as demand for rail was clear while Northern Pacific rose and fell. The question is which promise is being backed, and whose future revenue has to arrive for that promise to be kept.
How Northern Pacific's railroad finance chain worked
Northern Pacific’s line across the West depended on many parties, and each made a different promise.
Miners and steelmakers promised to deliver iron, coal, and rail. They were paid when they delivered their commodities to Northern Pacific.
Contractors promised to lay track and build bridges. They were paid as the work progressed on Northern Pacific's rail network.
Northern Pacific promised to pay all of them before the finished line carried enough traffic to cover its costs.
Jay Cooke & Co., a leading bank at the time, covered that gap by selling Northern Pacific bonds to investors. As costs rose and investors cooled on railroad bonds, Cooke couldn’t raise enough. His firm failed in 1873, and Northern Pacific entered bankruptcy in 1875.
The railroads were eventually built and the West was eventually settled. The overall bet was right. What failed was the timing, and the damage fell on whoever had promised to pay before the revenue arrived.
Mapping the railroad boom to our new AI boom
Power and equipment → construction → data center → compute → AI services → end customer
| Role | Railroad | AI | What must happen for them to be paid |
|---|---|---|---|
| Supplier | Miners and steelmakers | Power and equipment suppliers | The buyer pays on delivery. |
| Builder | Track and bridge contractors | Data center and grid contractors | The owner keeps funding the work. |
| Owner/operator | Northern Pacific | Data center owners, compute providers | Customers use and pay for the capacity for years. |
| Customer | Farmers and manufacturers | AI developers and enterprises | Their own products earn enough to cover the bill. |
| Financier | Cooke & Co. and bondholders | Lenders and investors | Every link above holds for the life of the loan. |
Read the last column from top to bottom. Each step down depends on more of the future. A supplier is betting on next month. A lender is betting on the whole buildout.
Strong company, different promise
A credit rating describes a company. It doesn’t describe each promise the company makes.
Take a well-established contractor building a data center. Its finances may be sound across a diversified business. But whether it can finish this project depends partly on terms set in the contract: how often it gets paid, how much is held back until completion, and whether the owner’s funding is secured or depends on tenants signing leases. Two projects with the same contractor can carry very different exposure to the AI buildout.
That exposure is set out in the agreements:
- The commercial contract defines what must be delivered, by whom, and by when.
- The bond defines which failure the surety covers.
- Guarantees and security define what stands behind the claim if that failure happens.
The test
Before accepting a credit in the AI buildout, an underwriter should answer three questions:
- What exactly is promised? Delivery, completion, payment, or years of performance?
- When does the promise come due? Before the asset earns revenue, or after?
- Whose revenue pays for it? The obligor’s existing business, or customers who haven’t paid yet?
Cooke’s bond buyers weren’t wrong about railroads. They misjudged what their bonds depended on, and how well secured they were. The same holds today: an underwriter doesn’t need a view on whether AI will pay off, only on whether the obligation in front of them depends on it.
Sources
Harvard Business School, “1873: Off the Rails” — Northern Pacific’s bond financing and Jay Cooke’s failure.
Minnesota Historical Society, Northern Pacific treasurer records — Cooke’s role in funding construction and equipment.
PBS, Story of the Northern Pacific — Northern Pacific’s 1875 bankruptcy.
Got a deal that needs collateral? Comity can help.
Comity supports a wide range of collateral posting needs in the power and digital infrastructure space. Emailing us or booking a 30-minute call is the fastest way to find out how we can help.