Anthropic’s new compute deal with SpaceX is the kind of move that shows just how weird – and how physical – the AI race has become. At the same time, it drops straight into a growing backlash against AI data centers that mixes very real local grievances with some truly out-there conspiracy theories.

Anthropic is, on paper, one of the “careful” AI labs – the one that talks constantly about safety, guardrails, and responsible deployment. Yet the company just signed on to tap the full power of Colossus 1, a massive supercomputer complex in Memphis originally built for Elon Musk’s AI operation. According to announcements from SpaceX/xAI, the deal gives Anthropic access to more than 300 megawatts of capacity and over 220,000 NVIDIA GPUs, including H100, H200 and next-generation GB200 accelerators, within roughly a month. That sort of firepower is what powers Anthropic’s flagship Claude models and, more concretely for paying customers, is supposed to ease the rate limits and “sorry, try again later” messages that have plagued Claude Pro, Claude Max, and heavy API users.

This is the new normal in AI: rivals renting each other’s infrastructure because there simply is not enough top-tier compute to go around. Colossus 1 was built to fuel Musk’s own Grok models, but as SpaceX focuses on a second-generation Colossus 2 cluster, the first campus is effectively being leased wholesale to a direct competitor. Musk himself has tried to frame the move as a kind of ideological due diligence: he posted that he agreed to lease the system after talking with Anthropic leaders about how they keep Claude “good for humanity.” Underneath the spin, though, is a simpler reality: whoever controls massive, reliable compute wins in the short term.

That scramble for compute is why we are suddenly talking about gigawatt-scale AI campuses in places that, until recently, were better known for farms, warehouses, or flat empty land. These complexes devour electricity, require access to high-voltage transmission lines, and often draw huge amounts of water for cooling. Companies like Anthropic, OpenAI, Microsoft, Google, Meta, and Oracle are competing to secure land, grid capacity, and construction timelines fast enough to keep their models on track. Fortune recently estimated that big tech firms are on pace to spend close to $700 billion on AI this year alone, much of that going into hardware and physical infrastructure rather than the glossy apps we actually see.

The human side of that buildout can look a lot less like the future and a lot more like an old-fashioned land fight. In Saline Township, a small Michigan farming community near Ann Arbor, residents packed into a 200-year-old township hall last year to debate whether 575 acres of farmland should be rezoned to make way for a sprawling OpenAI–Oracle data center. Locals raised worries that will sound familiar to anyone who has watched a controversial project move in: industrial noise, environmental stress, new demands on emergency services, and the loss of “prime farmland” that many assumed would stay agricultural. The planning commission and township board initially denied the rezoning, saying it conflicted with the town’s master plan.

Then the developer sued. Within weeks, faced with the risk of millions of dollars in potential damages, the township quietly settled, effectively clearing the way for the very project residents thought they had just defeated. Construction moved ahead less than two months after the original no vote. That whiplash transformed what looked like a straightforward zoning decision into something more like a civics lesson in power: who actually gets to say no when a multibillion-dollar AI project comes knocking? Some residents have since launched recall efforts against local officials, arguing that the town caved too easily and never truly represented community sentiment.

Zoom out, and Saline is one dot in a map that now stretches across Texas, Arizona, Louisiana, Michigan, and far beyond. Reporters who have gone town to town find the same tension repeating: data centers promise tax revenue, jobs, and “future-proof” infrastructure, but they also bring semi trucks, substation expansions, water rights questions, and land prices that leap out of reach of longtime residents. In some places, locals embrace the change as the best shot at fresh investment; in others, they see it as an industrial invasion happening on a timeline they did not choose. Either way, the AI boom is no longer invisible cloud magic – it is rows of concrete pads, towers of HVAC equipment, and fenced-off compounds breaking up farmland and desert scrub.

Into that already fraught mix, social media has poured gasoline. Community Facebook groups that started as organizing hubs for questions about traffic, noise, or well water have morphed into feeds where posts calling AI data centers “surveillance centers,” “military bases,” “killing machines,” or tools for “population control” rack up likes and shares. Some posts go further, claiming that officials are deliberately siting facilities on farmland so locals will lose the ability to grow food, or that tech firms are quietly preparing for some kind of digitally enforced social control. In one especially bizarre example documented by reporters, commenters accused Nvidia of installing “mini AI data centers” outside new homes as a step toward literally “implanting” people.

Once you move from zoning codes to microchips-under-the-skin, the line between legitimate skepticism and full-blown conspiracy gets blurry. Robert F. Kennedy Jr., who has spent years promoting unfounded claims about vaccines and wireless radiation, has now linked AI data centers to his long-running narrative about electromagnetic radiation harming human health. Mainstream scientific bodies and public health agencies say those broader radiation fears are not backed by solid evidence, especially at the exposure levels involved in typical telecom and data infrastructure. But in Facebook comment threads and TikTok explainers, that nuance tends to disappear; the story collapses into a vague sense that something dangerous is humming just beyond the fence line, and that nobody in power is telling the truth.

The result is a strange convergence: people worried about very concrete issues – tree lines, aquifers, traffic safety, tax abatements – are organizing in the same digital spaces where others are convinced that data centers are secretly weapons or mind-control facilities. For residents and activists trying to push for better terms or more transparency, the conspiracy content can actually get in the way; it gives developers and officials an easy way to dismiss online opposition as fringe, even when many people just want better answers about noise, water, or property values. And for companies like Anthropic or OpenAI, the temptation is strong to write off all backlash as “misinformation” rather than engaging with the very real ways these projects reshape local life.

At the core of all this is a trust problem. Big, technically complex projects are being proposed and approved on extremely aggressive timelines, often described in jargon that even policy experts need to stop and decode. Negotiations around incentives, tax breaks, and grid upgrades frequently happen behind closed doors, then get presented to the public as essentially done deals. When residents feel that decisions are being made far away, by executives and officials who assume communities will simply adapt, it creates exactly the kind of information vacuum where rumors, half-truths, and full conspiracies thrive.

Anthropic’s SpaceX deal lands squarely in that trust vacuum. On one level, it is just a business arrangement: Anthropic needs compute, SpaceX has a giant facility that just freed up, and money changed hands. On another level, it is a symbol of how fast AI companies are willing to move – even partnering with outspoken critics or ideological opposites – to secure the underlying infrastructure they believe they need. For people already anxious about an AI gold rush happening in their backyard, seeing marquee labs cut massive deals with Musk’s orbiting empire of rockets, satellites, and data centers can reinforce a sense that AI is something being done to them, not with them.

So what would it look like to actually handle this buildout differently? Some local governments have started pushing for slower, more deliberate processes, including temporary moratoriums on new data center proposals while they study long-term impacts on land, water, and the grid. Others are rewriting zoning codes to better distinguish between conventional server farms and these new, power-hungry AI campuses that can rival major industrial plants in their electricity use. On the industry side, there is growing talk about new cooling technologies, on-site renewables, and even, in Musk’s world, potential off-planet data centers as a way to reduce the strain on earthly communities and grids – though the space-based ideas are still speculative.

None of that solves the immediate challenge: people see bulldozers and substation upgrades now, not theoretical green tech later. If AI companies genuinely believe these facilities are “critical infrastructure of the future,” as executives like to say, then they cannot treat community engagement as an afterthought or a PR box to check once a deal is already structured. That means sharing more specifics earlier: actual numbers on water use, realistic job counts, contingency plans for grid stress, and clear answers on what happens if a project is abandoned or sold. It also means being honest about tradeoffs – that cheaper, faster AI for enterprise customers may come with visible costs in the towns hosting the underlying hardware.

The flip side is that communities also need better tools to separate signal from noise. It is absolutely reasonable to worry about air quality, groundwater, or tax fairness when a huge project shows up; it is less helpful to jump straight to theories about secret weapons, implants, or mass mind control. Local groups that want to be taken seriously can lean on independent experts, public-interest technologists, and environmental researchers to challenge company claims without drifting into fantasy. That does not mean giving AI firms a free pass; it means fighting with facts, which tends to work better in courtrooms and regulatory hearings than screenshots of Facebook memes.

In the background, the AI race rolls on. Companies report rising returns on AI initiatives – one recent survey of 10,000 businesses found that around 60 percent now see measurable ROI from AI projects, even though only a tiny fraction feel their data infrastructure is truly ready. That gap between ambition and plumbing is exactly why projects like Colossus 1 exist, and why billion-dollar bets on GPUs and power contracts are suddenly as important as clever model architectures. The risk is that the physical footprint of these bets grows faster than our ability to govern them, explain them, or earn public consent for where and how they are built.