The Right Way to Deal With AI Data Centers

June 23, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
June 23, 2026

HEADLINES

Mythos/NSA Story Gets a Correction

The viral claim that Mythos had broken into NSA systems turns out to be a misreading of congressional testimony — and Shashank Joshi, the Economist reporter at the center of it, has now confirmed as much. A US official told him that Senator Warner misunderstood NSA director General Rudd: the "hours, not weeks" language referred to a controlled red team exercise, not a live cyberattack. A cybersecurity account called Iris C2 further clarified that red teams begin with pre-established access to classified air-gapped networks — so Mythos didn't "hack" anything from the outside. What it does mean is that if an attacker gains initial access, AI tools like Mythos make the subsequent exploit dramatically faster, shrinking the window for detection. The real buried lede: the NSA's red teams no longer have access to Mythos, because their authority for using it was tied to Project Glasswing.

OpenAI Expands Daybreak, Launches "Patch the Planet"

OpenAI has rolled out the full version of GPT-5.5 Cyber and announced a new initiative called Patch the Planet, in partnership with security research firm Trail of Bits. The model has reduced guardrails for professional security work — similar to Mythos — and OpenAI is claiming it now outperforms Mythos on the CyberGym benchmark. Trail of Bits says the introduction of strong AI security models has fundamentally shifted what the expensive part of security work actually is: no longer finding bugs, but confirming findings, writing acceptable patches, and managing disclosure. Over 30 open source projects have already joined Patch the Planet. All of this landed alongside a rare Five Eyes public bulletin warning that AI-driven cyber risk is now a core business issue, not just a technical one — framing the threat in language clearly aimed at enterprise leadership rather than security teams.

Trump Signs Quantum Computing Executive Orders

President Trump signed a pair of executive orders calling for a powerful US quantum computer by 2028 and directing federal agencies to migrate to quantum-secure cryptography by 2031. The orders cover both building the technology — through public-private collaboration led by the Energy Department — and protecting it, with provisions on intellectual property and supply chain security. Quantum is a genuinely contested space: credible scientists remain deeply skeptical of the technology's scalability, while others point to recent Google papers becoming notably less detailed as a possible sign of meaningful progress being kept close. With this administration's track record of taking stakes in sectors it then promotes via executive action, it's difficult to know how much signal to read into the policy direction.

SpaceX Signs $6.3B Compute Deal with Reflection AI

Open source AI startup Reflection AI has agreed to pay $150 million a month for capacity at SpaceX's Colossus II data center — a deal running from July through 2029 and worth up to $6.3 billion in total, though either party can walk away on three months' notice. That puts it meaningfully below the Anthropic and Google deals, each running at around a billion dollars a month. Reflection used the announcement to pitch itself as a domestic open source alternative, citing "recent events" around export controls and closed model risk. The company has yet to release a frontier model, though it has been working with government partners including the Pentagon and the Department of Energy's Genesis Mission. On the SpaceX side, the deal extends a now-substantial compute-leasing business — one observer noted that SpaceX has already recouped roughly half its investment in Cursor through compute deals alone.

Google Loses $200B in Market Cap Over Two Departures

Google stock fell as much as 7.2% on Monday — its largest intraday drop since February — following the departures of Nobel laureate John Jumper to Anthropic and Noam Shazeer to OpenAI. That move effectively erased over $200 billion in market cap attributable to two individuals leaving. The market reaction may be an overreaction in the DeepSeek mold, but there's also a more substantive narrative at work: Google had briefly become the largest company in the world on the back of Gemini's strong run, and these departures feel like confirmation of a view that's already been forming — that OpenAI and Anthropic are pulling away on enterprise agentic use cases. Gil Luria of D.A. Davidson put it plainly: Google had the state-of-the-art model for a few weeks and has fallen off since. The bull case remains that Google has unmatched TPU infrastructure and data, and that peak-pessimism moments tend to precede big releases.

MAIN STORY

The Right Way to Deal with AI Data Centers

Opposition to AI data centers has become one of the strangest bipartisan coalitions in American politics — Erin Brockovich on one side, former Tea Party conservatives on the other, with the New York Times quoting a Milwaukee comedian calling it "the most bipartisan issue since beer." The concerns tend to cluster around two things: water use and electricity prices. Both are worth taking seriously, and both are more manageable than the opposition suggests — but the real opportunity isn't winning a debunking war. It's that communities sitting in the path of this buildout have far more negotiating leverage than they realize, and most of them haven't tried to use it.

DEBUNKING THE DATA CENTER MYTHS

The Water Myth
The numbers sound enormous — until you compare them to anything.
Amazon recently published full water use figures for the first time: 2.5 billion gallons across all its global data centers in 2025, down 2% year over year despite a significant expansion in footprint. That sounds vast, but the 16,000 golf courses in the US consume over 500 billion gallons a year — meaning Amazon's entire annual water use amounts to roughly one day of US golf course maintenance. California almonds use between 1.2 and 1.8 trillion gallons annually, five to eight times all US data centers combined. The US also loses 3.29 trillion gallons a year to leaky pipes — about 15 times data center consumption. One key source of the water myth: Karen Hao's book Empire of AI claimed a Google data center in Chile consumed a thousand times more water than the surrounding population — a figure she later acknowledged was off by a factor of 1,000 due to a unit mix-up. The correction was issued; the original claim is still in the book. In Indiana, where residents were alarmed by Amazon's 300-million-gallon figure, that number represents 0.2% of the state's annual domestic water use. The companies are also working to reduce consumption further: NVIDIA recently touted a new liquid cooling approach that could bring data center water use to near zero.

The Electricity Price Question
The aggregate data doesn't show a correlation — but the local picture is messier.
The Institute for Energy Research analyzed state-level data and found no statistically significant correlation between data center concentration and either current electricity prices or the rate at which prices are rising. Most electricity price pressure, the research suggests, comes from the cost of upgrading aging grid infrastructure — a problem that exists with or without the data center buildout. That said, The Daily Economy's examination of the same research found the local picture is more complicated: a Bloomberg analysis of wholesale electricity prices across 25,000 grid nodes found prices up as much as 276% since 2020 near major data center clusters, with more than 70% of nodes recording price increases located within 50 miles of significant data center activity. The conclusion from that analysis: concentrated price spikes are evidence that grid infrastructure and cost allocation rules haven't kept up — not that data centers are inherently incompatible with affordable electricity. Policy is beginning to catch up: Oregon's Power Act requires the biggest electricity users to bear the cost of infrastructure built specifically for them, and the White House's Ratepayer Protection Pledge commits data center operators to building new power supply, paying for delivery infrastructure upgrades, and covering costs whether they use the power or not.

The Middle Path: Negotiate
Communities have more leverage than they're using.
The real argument from The Information's Anne Davis Vaughn — whose piece on debunking data center myths is worth reading in full — is that the binary framing of "roll over" versus "get your pitchforks" is causing communities to miss a massive middle path. Data center operators need community buy-in, and that creates real negotiating leverage. In rural Richland Parish, Louisiana, hundreds of teachers are set to receive unprecedented $50,000 bonuses this year, funded by a surge in tax receipts tied to Meta's AI campus construction — the result of an existing ordinance requiring teachers to get a share of sales taxes. Labor unions may be the most important actors for finding this middle ground: they're sympathetic to community concerns while also representing the skilled workers who stand to benefit most directly from the blue-collar jobs boom the buildout is generating. The conversation needs to move from debunking to deal-making.