The Subsidy Reflex in the Age of AI
A subsidy becomes policy the moment no one remembers what it was meant to correct.
Governments tend to treat new technologies the way tired parents treat precocious children. They reward enthusiasm first and ask about consequences later.
Tax credits become the bonding mechanism. The problem is not malice. It is drift. Once incentives are set, the system follows them with surprising obedience.
The AI buildout is arriving at the moment when public budgets are already strained. It appears that private ambition has learned how to lean on public capacity long before it learns how to pay for it.
OpenAI recently asked the federal government to expand the Advanced Manufacturing Investment Credit under the CHIPS Act to cover a broad AI infrastructure stack.
The AMIC was designed to support domestic semiconductor manufacturing by offering up to a 35% credit for qualified investments. OpenAI’s request would extend that treatment to data centers, steel grid components, and other inputs needed for large-scale AI operations.
The company framed its position as an effort to preserve US leadership in artificial intelligence. The factual implication is simple. OpenAI wants federal tax support for projects that already receive significant state and local subsidies.
Its proposed federal expansion would sit atop property tax abatements, discounted electricity arrangements, and accelerated permitting already common in states hosting major data center projects.
What changed is not the underlying law. It is the industry seeking to occupy a policy built for something else.
The request highlights a recurring cycle in industrial policy.
Once a credit exists, it attracts claimants who see themselves as adjacent to the original intent. Policymakers often accept these expansions because the infrastructure appears strategic, the jobs are marketed as high-value, and the political optics favor investment over hesitation.
Yet the record suggests something quieter.
When industries scale faster than public utilities, costs migrate—electricity demand spikes. Local grids adapt around private siting decisions. School districts and municipalities surrender future revenue in exchange for uncertain long-term economic development. The public pays for the ambition twice: once through incentives and again through the rising cost of accommodating the infrastructure itself.
This episode also signals the growing convergence between technology firms and utility economics.
AI is functionally an energy business wrapped in software language. Any attempt to subsidize it without acknowledging that reality invites misalignment.
Expect that credits designed for one industry will be repurposed by the next. The boundary between semiconductor manufacturing and AI infrastructure is already proving porous.
Treat local subsidies as part of the federal picture, since state abatements can create hidden leverage that a federal credit amplifies.
Large data infrastructure entails long-term utility obligations, with financial burdens beyond tax credits. Companies may offer social benefits, such as training or grid modernization, as negotiation tools rather than commitments.
Transparency is rarely voluntary; without mandated disclosure of overlapping incentives, communities can't assess actual public exposure.
Local officials see immediate construction and a handful of high-wage jobs. Companies see a subsidized foothold. Residents see rising utility bills and infrastructure designed around someone else’s priorities. Each group responds to the incentives directly in front of them.
It appears likely that AI infrastructure will continue pressing into policy spaces never designed for it. Congress may resist a broad credit expansion today, only to revisit the idea under a different name when the energy constraints tighten. Industrial policy tends to repeat itself, especially when the infrastructure at stake becomes politically symbolic.