The American AI infrastructure race increasingly looks like industrial policy conducted at state level. Pennsylvania and Indiana are not competing to host a software office. They are competing for campuses that require land, large power connections, construction capacity and a workforce able to operate expensive digital infrastructure.
Amazon's recent commitments show how quickly the scale has changed. Pennsylvania announced a plan for at least $20 billion of investment, while Amazon has separately set out a further $15 billion programme in Northern Indiana on top of earlier spending there.
The project is as much about utilities as technology
A data-centre campus can be announced before all of its electricity is available. That makes the utility relationship central to delivery. Indiana's project is tied to 2.4GW of planned capacity and an agreement with NIPSCO, while Pennsylvania's announcement is accompanied by state workforce funding and multiple potential locations.
This is why a state with cheap land does not automatically win. Developers need enough electricity, realistic transmission schedules, fibre, contractors and a permitting process that can keep pace with the capital commitment.
States are building labour policy around infrastructure investment
Pennsylvania paired the Amazon announcement with a $10 million workforce initiative directed through schools, community colleges and union training rather than as a payment to Amazon. Indiana's programme also includes training for data-centre technicians and fibre work.
That distinction matters because permanent employment at a large data centre is much smaller than the construction workforce required to build it. A credible economic case therefore depends on supply-chain activity, training and whether the infrastructure attracts additional investment.
Announced dollars are not the same as commissioned capacity
State leaders understandably compete on investment totals, but the useful scorecard is physical. Which sites have land? Which have signed power arrangements? Which are under construction? Which are energised? How much of the planned capacity is carrying customer workloads?
American Commerce Review already tracks data-centre investment by state. Pennsylvania and Indiana are useful case studies because they show why project stage belongs beside every headline dollar figure.
What to watch next
The next competition will be for power and transmission as much as corporate incentives. States that can bring credible capacity online without shifting unreasonable network costs to existing customers will have the stronger long-term proposition.