OpenAI wants Chips Act tax breaks to cover AI data centers and grid hardware

OpenAI has asked the US government to widen a key Chips Act style tax credit so it does not just apply to semiconductor plants, but also to AI data centers, AI server manufacturers, and even parts of the electrical grid. In a letter to White House Office of Science and Technology Policy director Michael Kratsios, OpenAI’s Chris Lehane argues that if the United States wants to stay ahead in AI, the incentives cannot stop at fabs and packaging lines. They need to reach the racks, transformers, and steel that actually run the models in production. :contentReference[oaicite:0]{index=0}

What OpenAI is actually asking for

The request focuses on the Advanced Manufacturing Investment Credit, a 35 percent tax credit that currently supports chipmaking and related semiconductor manufacturing investments under the Chips Act umbrella. OpenAI wants that credit expanded so it also covers:

AI data centers dedicated to training and inference.
Manufacturers of AI servers and accelerator systems.
Critical grid components like high voltage transformers, converters, and the specialized steel used to build them.

The logic is simple: the bottleneck in AI is no longer just GPU supply. It is also power, cooling, and the grid hardware that takes years to build. If the tax code only subsidises fabs, the compute stack stays lopsided. :contentReference[oaicite:1]{index=1}

How this fits with the “no bailout” line from Altman

The timing here is awkward for OpenAI’s messaging. Just as the letter surfaced, Sam Altman went on social media and said OpenAI does not want government guarantees or bailouts for its data centers, and that taxpayers should not be on the hook if AI companies overreach and fail. He framed any government role as focused on chip fabs and long term industrial capacity, not underwriting OpenAI’s own debt stack. :contentReference[oaicite:2]{index=2}

At the same time, the company is still lobbying to move AI infrastructure into the same policy lane as chip plants. The difference is where you point the support. Altman says no to guarantees that backstop a specific company’s loans. Lehane’s letter says yes to broad based tax credits and manufacturing incentives that lower the cost of AI hardware and the power gear around it for everyone.

Why AI data centers are not covered today

The Chips Act and the current Advanced Manufacturing Investment Credit were written with fabs in mind. Think lithography tools, clean rooms, packaging, and upstream materials. They do not automatically treat a hyperscale AI data center as “manufacturing” in the same way, even if that site consumes more power than a small town and houses tens of billions of dollars worth of processors.

OpenAI’s ask is essentially to redefine part of AI infrastructure as strategic manufacturing. In policy speak, it pushes the boundary from chips only to “chips plus the industrial stack needed to use them at scale”. That would be a big shift for tax law, because it opens the door to subsidising racks, power systems, and grid reinforcements that look a lot more like utility or real estate projects than a classic fab. :contentReference[oaicite:3]{index=3}

The scale problem OpenAI is trying to solve

OpenAI has talked about spending on the order of 1.4 trillion dollars over the next eight years on chips and data centers if its roadmap holds. That number is big enough that you have to sanity check it against national utility plans and grid build outs, not just tech budgets. Even if the eventual figure comes in lower, the direction is obvious. These clusters are beyond the scale that normal venture style funding can carry on its own. :contentReference[oaicite:4]{index=4}

To close that gap, OpenAI is already lining up long term deals with GPU vendors, cloud partners and power producers, and talking openly about selling compute as an “AI cloud” in its own right. Tax credits that cut the effective cost of servers and grid gear by a third change the model for everyone who has to buy this kit, not only OpenAI. That is exactly why the company is pushing for a policy change while trying to reassure people it is not asking for a personalised bailout.

Who else would benefit if the credit is expanded

If Washington says yes, the winners are not just OpenAI. You would expect:

Hyperscale cloud providers building AI regions in the US to claim credits on new AI focused data halls.
Server OEMs and ODMs making AI boxes and racks in the US to get more support for manufacturing lines.
Transformer and grid equipment makers to get help expanding plants and cutting lead times on the high voltage gear AI clusters need.
Colos and regional data center operators to position new builds as AI ready and lobby to fit within whatever “AI data center” definition Treasury writes.

In other words, this is an ecosystem play. The credit would indirectly flow into land, steel, transformers, switchgear, and concrete in the same way the original Chips Act flows into clean rooms and tool vendors today.

Why this is politically sensitive

On paper, the argument sounds like standard industrial policy: the US wants to stay ahead in AI, China is building AI megacenters in the desert, so you widen your manufacturing incentives to cover more of the stack. In practice, voters see “huge data centers for rich AI companies” and “bigger subsidies” in the same sentence and start asking who pays when the hype cycle breaks. :contentReference[oaicite:5]{index=5}

The Trump administration has already signalled there will be no explicit bailouts for AI companies that overextend on data centers. That draws a hard line for direct guarantees, but it does not block sector wide tax credits that apply to any qualifying project. OpenAI’s letter is trying to stay on the right side of that line: no government cheque if they blow up their balance sheet, but cheaper capital for anyone willing to build AI infrastructure on US soil.

Is this good policy, or just good lobbying

There is a reasonable case for updating incentives. AI workloads are stressing parts of the grid that were never designed for multi gigawatt clusters, and transformer lead times are measured in years. If you believe AI is strategic, subsidising the hard bits of that build out is not crazy. It also helps nudge projects to stay in the US instead of chasing cheaper power and looser rules elsewhere.

The flip side is that widening the tax credit too far turns it into a catch all subsidy for data centers and hardware that would have been built anyway. At that point you are back to the classic problem: socialising capex risk while privatising margins. The line between “targeted manufacturing support” and “corporate welfare for hyperscalers” is thin, and it will be Treasury guidance, not OpenAI’s letter, that decides where this lands.

Right now, OpenAI is trying to thread a needle. It wants Washington to treat AI like the next great industrial platform, without being tagged as the poster child for an AI bailout. The request to expand the Chips Act style tax credit to AI data centers and grid gear is one of the first real tests of how that balance will work in practice.