What Doesn’t Compute about “Compute Campuses”

Irina Raicu is the director of the Internet Ethics program (@IEthics) at the Markkula Center for Applied Ethics. Views are her own.

Two days after the midterm elections, OpenAI published an extensive document titled “AI Infrastructure That Works for America: Our Compute Campus Policy Blueprint.” “At its core,” argues the first line of the document, “the promise of AI is about expanding human agency.”

Of course, the promise of voting in a democratic society is also about expanding human agency. And one notable part of the recent election’s results is that out of 47 ballot measures focused on the deployment of data centers in communities across many states, the call to restrict such development had won 44 of them—with the votes in two of the remaining three still being counted as of November 12.

Public pressure around data centers had been building over the last year. When a colleague and I first started some related research in 2025 and looked for media coverage of community pushback, in California at least, we couldn’t find much. Then we started to see more and more reports. By June, polling showed that the American public had “swung 49 points against data centers in just nine months, underscoring the heightened political salience of the facilities and the AI industry that they embody.”

OpenAI is one of the leaders of that industry.

OpenAI’s new policy brief argues that “[t]he AI infrastructure buildout is the first major test of whether the benefits of AI will be broadly shared, and whether industry and government can work together to give American communities a meaningful voice in shaping their economic futures.” If so, the voting signaled that both industry and government have so far failed that test.

The policy brief also accurately flags the important fact that “AI training data centers [which, the brief claims, "are more accurately described as compute campuses"] are unlike most data centers that power internet searches, email or phone calls.” It’s an important distinction that many conversations about data centers ignore. However, in this case the distinction is presented as a positive: the scale of such setups, OpenAI writes, “creates massive demand for skilled construction and operations jobs. These dense computing systems, specialized power and cooling equipment, and recurring technology upgrades create demand for technical expertise to build, operate, maintain, and modernize them.”

This is an answer to a particular critique leveled by some against data centers: that, once built and functional, data centers bring few jobs into the local economy. According to the policy brief, the AI-specific data centers will be different, precisely because of their greater demands. Of course, however, superfund cleanup sites also create jobs—but other consequences still make most people reluctant to live near them. And most don’t see those jobs as evidence of the benefits of technology being “broadly shared.”

The policy brief does make several significant commitments, only one of which is “Good jobs for local residents”; the others are “Listening and learning” (to and from communities); “Clear accountability” (defined as “access to essential information about energy use, water resources, noise, and other local impacts, along with public reporting on progress on commitments”); and “Ratepayer protection and energy modernization.” It also includes useful details on the implementation of each of those commitments. And it urges lawmakers to enact some related laws: among them, “barring developers from requiring public officials to sign non-disclosure agreements, mandating publication of noise, water, energy and economic impacts, and protecting ratepayers.”

Unfortunately, those detailed recommendations come at a time when Americans are deeply skeptical about arguments and promises coming from big tech companies. And parts of the policy brief are bound to reinforce that skepticism. When OpenAI writes about “democratic AI” and explains that it is “AI that provides widely distributed economic opportunity, supports people’s individual agency, and prevents a concentration of benefits and power,” readers will be reminded that, so far, great technological advances in AI have failed to translate into any of those things.

Data centers, including some specifically designed and built for AI, are still needed. Processes that include community input and prioritize transparency and care in development are also very much needed. But what the policy brief lacks is a tone of humility, or at least acknowledgment of the ways in which data center development has mostly failed to engage, protect, and support communities so far.

On the very first page, the brief techsplains to readers that an AI data center “in fact isn’t [a] big box in a pasture to store data with few economic benefits for the community, but a materially different facility that powers people’s ability to scale their own ingenuity.” It’ll lose some readers right there—if it hadn’t lost them already at the headline, which first introduces the effort to rename the new generation of data centers as “compute campuses.” Even if the latter term more accurately describes what happens in the big boxes in a pasture or the multistory buildings in cities or the buildings proposed to replace endangered chestnut groves or the massive server-filled tents in some places, the renaming effort is likely to be seen as another attempt to deflect, rather than respect.

Image: Nadia Piet  & Archival Images of AI + AIxDESIGN / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/