A yearlong Senate investigation says some of the biggest AI infrastructure builders are misleading the public about what data centers cost and what they deliver. The findings, first shared with TIME on October 8, 2026, come from the offices of Senators Elizabeth Warren (D-Mass.), Chris Van Hollen (D-Md.) and Richard Blumenthal (D-Conn.), who looked at seven developers: Amazon, Google, Meta, Microsoft, CoreWeave, Digital Realty and Equinix.
This is a report from Democratic senators, not a ruling, and it creates no new rules. But it is the most detailed look yet at the economics behind the AI build-out, and it lands as data centers have become a campaign-trail issue. This post covers what the investigators say they found, which claims are the senators' own, what the companies said back, and why it matters for anyone following AI infrastructure. For the wider local-opposition picture, see our AI data center backlash map.
What the Senate investigation is, in one table
| Question | Short answer |
|---|---|
| Who ran it? | Warren's office, with Van Hollen and Blumenthal |
| What was examined? | Seven developers: Amazon, Google, Meta, Microsoft, CoreWeave, Digital Realty, Equinix |
| How? | Staff requested information from each company and interviewed employees |
| Main charge | Some firms mislead the public about the real costs and benefits of data centers |
| Status | A report with findings. No law, fine or enforcement action |
| What next? | Expected to feed a broader congressional effort on data center guardrails in the coming year |
The inquiry did not start this week. As Data Center Dynamics reported, the three senators sent letters to hyperscalers in December 2025 asking about the effect of data centers on US electricity prices; Bloomberg covered that opening move at the time. The October report is the result of that yearlong effort. TIME notes the findings "largely match reporting from various news outlets over the past year", so the value is less in surprise than in a consolidated, document-backed version.
Finding 1: jobs claims are thin once construction ends
Developers routinely cite construction employment when asking for incentives. Investigators say that when staff asked for comprehensive information about permanent employment, several companies refused. Some told investigators that permanent staffing runs at roughly one worker per megawatt of power demand.
TIME turns that ratio into a concrete example: a 100-megawatt data center could use about as much electricity as 100,000 homes while employing about 100 people after construction. That is the report's framing of the companies' own ratio, not an independently audited figure, but it explains why local officials keep asking what permanent benefit a site brings. Our earlier piece on a water claim by Sam Altman shows how quickly resource numbers around AI get contested.
Finding 2: the biggest subsidy may be sales tax on chips
Most public debate focuses on property-tax abatements. The investigators found that the most lucrative incentives are not necessarily those headline-grabbing property-tax breaks, but sales-tax exemptions on computer equipment. The logic is simple arithmetic: AI facilities need huge quantities of specialized hardware, and GPUs account for an estimated 39% of spending at an average 1-gigawatt AI data center, according to the report. A sales-tax exemption on that equipment is worth far more than a break on the building.
That matters because the hardware bill is exploding. If you want the scale of the spending behind it, see our coverage of OpenAI's projected compute spend through 2030. Every dollar of equipment exempted is revenue that a state or county does not collect.
Finding 3: nondisclosure agreements limit scrutiny
All four large technology companies in the probe, Amazon, Google, Meta and Microsoft, have routinely sought nondisclosure agreements during data center development, per the report, and some acknowledged asking government officials to sign them. Data center negotiations do involve commercially sensitive information, but the report argues NDAs have also restricted public scrutiny of deals that involve tax dollars, utility rates and public infrastructure.
The companies responded differently:
- Microsoft told investigators it will stop seeking NDAs with local governments, though it will keep using them with state agencies, public utility commissions and utilities.
- Amazon has announced a similar policy.
- Google and Meta declined to commit to ending the practice with local governments.
Finding 4: nobody would agree to pay for what they cause
The longest part of the report asks who pays for the power. The investigators found that none of the seven companies would agree to a standard requiring them to pay for building new power infrastructure that would not have been needed but for their data centers.
The companies' position, as TIME summarizes it, is that they will pay the direct costs of serving their facilities, but that larger grid investments, such as new power plants and transmission lines, can benefit other customers and should not automatically be charged to the company that created the need. That is a real argument, not a dodge: grid upgrades often do serve many customers. The disagreement is about how to apportion costs when one very large new customer triggers the build.
The Louisiana example
The report points to a power plant in Richland Parish, Louisiana, that the utility Entergy has sought to buy. Analysts argue the purchase was driven mainly by Meta's planned $50 billion data center, expected to draw 4,500 megawatts, about four times the peak electricity demand of the entire city of New Orleans. Estimates cited in the report suggest the purchase could raise bills for the average Entergy customer by $8 to $13 per month. Meta has disputed that its project is responsible for those costs. Treat the dollar range as an estimate and the causation as contested.
Server blocks with green airflow lines, standing for the grid and cooling load of a large AI data center
Why this lands now
Three things make the timing notable.
- Public opinion. TIME cites a recent Economist/YouGov poll in which about half of Americans say data center construction is bad for the country.
- Both parties are using the issue. Even as President Trump warns that opposition could push the AI buildout overseas, politicians on both sides are campaigning on it.
- A stalled bill. In September the House passed the bipartisan Ratepayer Protection Act 417-3. It would direct states to consider standards requiring large-load customers such as data centers to cover the incremental generation, transmission and distribution costs of serving them. In the Senate it stalled: Democrats argued it was toothless because it did not require states to push any costs onto the companies, and the Senate voted 57-43 against advancing it.
Warren has called for a national moratorium on new AI data centers until developers agree to cover the full costs. Van Hollen says Congress needs a bill that ensures the corporations "cover the costs of the energy they need, instead of pushing those costs onto the backs of consumers."
Because electricity rates are mostly set by state regulators, TIME notes the biggest fights will keep playing out in statehouses and public utility commissions. That is already visible in local actions: San Francisco's 45-day data center moratorium, the New York moratorium push, protests over Amazon's site in Thane, India, and California's data center bills.
What is verified and what is not
Being clear about provenance matters in a politically charged story.
Reported by TIME from the senators' report: the seven companies examined; the refusal by several to provide permanent-job data; the one-worker-per-megawatt ratio some companies cited; the NDA practices and Microsoft and Amazon's partial reversals; the 39% GPU share; the refusal of all seven to accept a pay-for-what-you-cause standard; the Entergy example.
Claims by the investigators, not independent findings: that the firms are "misleading" the public. That is the authors' characterization. The report was produced by Democratic senators with a stated policy goal, and TIME says its conclusions largely echo prior press reporting rather than revealing new hidden facts.
Disputed: Meta's responsibility for the Louisiana-related costs. Companies also argue grid investments benefit other ratepayers.
Not yet confirmed by us: we have read TIME's account, not the full report text, so figures here come through that reporting. When the full document is available, check it directly.
What it means for people who build with AI
If you use AI APIs or run products on them, this is a cost-side story, even if it feels distant. Compute is the largest input to model pricing, and the political fight is over who absorbs the cost of the power behind it. Three practical takeaways:
- Siting risk is rising. More moratoriums, NDAs under scrutiny and ratepayer rules mean slower, costlier capacity. Our look at China's 24 GW versus the US 56 GW compute shows the competitive stakes that companies cite when asking for speed.
- Self-supply is a likely response. Companies are already signing their own generation deals, such as Google's nuclear uprates, partly to blunt the argument that they raise bills for neighbors.
- Transparency expectations are shifting. Microsoft and Amazon's NDA moves suggest disclosure to local governments is becoming a baseline expectation.
What to watch next
- Whether the report's findings turn into a bill when Congress returns after the midterm elections.
- Whether Google and Meta change their NDA stance with local governments.
- State regulators' rulings on large-load tariffs, which decide who actually pays.
- Any company rebuttals published against the report, especially on the Louisiana estimate.
We will update this post if the full report or formal company responses appear.
FAQ
See the structured questions above for quick answers on the scope, the companies involved, and the status of the report.