A U.S. Senate probe accuses major technology companies of misleading the public about AI data centers, stating that sites drawing 100 megawatts of power create only about 100 jobs while the operators secure lucrative tax breaks on GPU hardware.

The investigation, first reported by Tom's Hardware, centers on the gap between the facilities' scale and their employment footprint. Senators on the panel are questioning how many permanent positions the data centers actually create — and the companies involved have refused to divulge that data.

What does the probe allege?

The core accusation is one of misrepresentation. According to the Senate investigation, Big Tech firms have promoted AI data centers as major job creators while the facilities' actual permanent employment numbers tell a different story: roughly one job per megawatt of capacity at 100 MW-class sites.

The senators' specific line of inquiry targets permanent jobs. Data center construction generates temporary contracting work, but the operational phase requires a small permanent staff, and the probe puts that figure near 100 for a 100 MW installation.

Meanwhile, the same operators have secured tax breaks tied to GPU purchases — the accelerator hardware that powers AI training and inference. The senators are weighing whether the public subsidy per permanent job justifies the incentives currently on offer.

The takeaway: each permanent job at a large AI data center is backed by 1 MW of dedicated power capacity and a package of GPU-related tax relief.

Why won't companies release the numbers?

The Senate panel asked the companies to disclose permanent employment figures for their AI data center sites. The companies refused to divulge the data, according to the probe.

That refusal now sits at the center of the dispute. Without operator-supplied figures, the senators are relying on their own estimate of approximately 100 permanent jobs per 100 MW site — and the operators have not offered a competing count.

The takeaway: the companies contest the framing but will not provide the job numbers that could settle it.

What is at stake for the GPU tax breaks?

The tax incentives in question apply to GPU hardware, the most expensive component class in AI data centers. A 100 MW facility can house thousands of accelerators, and the associated tax relief scales with that hardware investment.

The probe's arithmetic sharpens the political question: if a 100 MW site yields about 100 permanent jobs, then the incentive cost per long-term position grows with every megawatt of expansion. Senators questioning the program are effectively asking whether the breaks were sold to the public on inflated employment projections.

The operators, for their part, have not disputed the 100-job figure publicly — they have declined to supply data at all.

The takeaway: the subsidy-per-job ratio, not the raw job count, is the number that could decide the fate of the GPU tax breaks.

What happens next?

The Senate probe has no binding outcome yet. Its immediate effect is pressure: continued refusal to disclose employment data keeps the companies' job-creation claims unsubstantiated, while the 100 MW-to-100-jobs ratio stands as the only public figure in the debate.

For state and federal legislators weighing future incentive packages, the investigation supplies a benchmark they can apply to any proposed AI data center deal.

The companies' answer so far is silence.