A slower news cycle gives NLW room to defuse one of AI's most contentious fights — data centers — by actually checking the water and electricity numbers. Plus a Mythos/NSA correction, OpenAI's cyber escalation, another SpaceX compute deal, and a $200B Google wobble over two departing researchers.
The data center fight needs a middle path between rollover and pitchforks.
Both sides of the data center debate are wildly reductive. The scary water and electricity numbers mostly don't survive contact with context — Amazon's entire global data center fleet uses a day's worth of US golf course watering. But communities aren't powerless either. The real opportunity isn't 'let big tech do whatever' or 'stop it entirely' — it's negotiating for serious economic benefits, like the Louisiana parish where teachers got $50,000 bonuses funded by Meta's campus. Move past the binary into the how-everyone-wins phase.
— Theo Von, This Past Weekend podcast. Theo Von's podcast rant — warning of an emotional credit score and AI becoming a new god — packs a remarkable number of misperceptions into one paragraph. NLW argues Von reflects an emerging mainstream view rather than leading it.
Erin Brockovich launched a campaign against data centers, former Tea Party conservatives are planning nationwide protests, and comedian Charlie Berens called opposition 'the most bipartisan issue since beer.' The two recurring complaints: water use and electricity prices.
Amazon's global data centers used 2.5 billion gallons in 2025 (down 2% even as footprint grew). US golf courses use over 500 billion gallons a year — meaning Amazon's full year is roughly one day of US golf maintenance. California almonds use 1.2–1.8 trillion gallons, 5–8x all US data centers.
One study found the US loses 3.29 trillion gallons a year to leaky pipes — about 15 times total data center water use. In Indiana, the contested Amazon facility's 300 million gallons a year equals roughly 0.2% of the state's domestic water use.
Author Karen Hao's claim in Empire of AI that a Chile Google data center consumed 1,000x the surrounding population's water was wrong by a factor of 1,000 due to a unit mix-up. She acknowledged and corrected the error — but the book is still out there making the claim.
NLW's core diagnosis of the water debate: big numbers are politically potent because most people have no sense of how much water we actually consume. The astronomical-sounding figures grab attention regardless of context.
The Institute for Energy Research found prices in the top-ten data center states are virtually identical to the average elsewhere, with no significant relationship to faster rate increases. Most price pressure traces to upgrading an aging grid — a cost we'd face anyway.
A Bloomberg analysis of 25,000 grid nodes found wholesale prices rose as much as 276% since 2020 near data center clusters, with 70%+ of increases within 50 miles of significant activity. The Daily Economy's read: not evidence of incompatibility, but that grid infrastructure and cost-allocation rules haven't kept up.
Oregon's POWER Act requires the biggest users to bear the cost of infrastructure built specifically for them. The White House's Ratepayer Protection Pledge has companies committing to build/buy new power supply, pay for delivery upgrades, pay whether they use the power or not, and invest in local jobs and resilience.
NLW argues unions sit in both camps — sympathetic to community concerns that AI disproportionately benefits the rich, yet seeing firsthand how valuable the build-out is as demand for skilled blue-collar work climbs. They're positioned to broker the practical middle ground.
— The Information (Anne Davis Vaughn). The Information's reporting from rural data center projects found communities can extract far more financial concessions than they think — they're being taught the choice is roll over or grab pitchforks, while missing a massive middle path.
In rural Richland Parish, hundreds of teachers are set to receive unprecedented $50,000 bonuses funded by a surge in tax receipts tied to Meta's AI campus. An existing ordinance gives teachers a slice of sales taxes; bonuses quintupled from construction activity — a concrete example of the 'how everyone wins' phase NLW wants.
An Economist reporter issued a correction: Senator Warner misunderstood NSA Director Rudd. The 'hours not weeks' wording was real, but Mythos's use was a controlled red-team exercise testing internal networks — not a live cyberattack — and the agency's red teams no longer have access to it.
Per cybersecurity account Iris C2, gaining initial access to air-gapped classified systems remains the hard part. The real takeaway: once inside, Mythos makes exploit design and execution much faster, cutting the time defenders have to detect and contain an intruder.
As part of its Daybreak initiative — OpenAI's answer to Anthropic's Project Glasswing — OpenAI launched the full GPT 5.5 Cyber, its first cybersecurity-tuned model with reduced guardrails. It claims the model now overtakes Mythos on the CyberGym benchmark.
— OpenAI. OpenAI positioned Daybreak as more open than Glasswing, inviting smaller organizations to apply for access so defenders everywhere can find and fix vulnerabilities before attackers exploit them.
OpenAI launched Patch the Planet with Trail of Bits to secure open-source libraries. Initial testing of GPT 5.5 Cyber surfaced hundreds of bugs; 37 patches are deployed with many more in the pipeline and 30+ projects already enrolled.
— Trail of Bits. Trail of Bits argues the value has shifted to confirming findings, getting severity right, writing patches maintainers will accept, and coordinating disclosure — exactly the work that floods of AI-generated reports threaten to bury.
The intelligence alliance warned that frontier AI is rapidly transforming cyber risk and that assumptions can become outdated in months, not years. The UK NCSC bulletin urged businesses to integrate AI into security ops and treat cyber as 'a core business risk and leadership responsibility,' not a purely technical issue.
Two new executive orders direct federal agencies to partner with industry on a quantum computer for research (OSTP's Kratsios targets 2028), mandate migration to quantum-secure cryptography by 2031, and harden the quantum supply chain against adversaries.
Open-source startup Reflection AI agreed to rent Colossus II capacity through 2029 for $6.3B total, with three-month walk-away terms. It's notably smaller than the Anthropic and Google SpaceX deals running near $1B a month each.
— Reflection AI. Yet to ship its first frontier model, Reflection used the deal to argue that nations and enterprises increasingly recognize 'the risks and costs associated with exclusively depending on closed models.' It's working with the Pentagon and DOE's Genesis Mission.
— Swigs, Layton Space. Swigs of Layton Space argued no one is doing the math right: SpaceX has already recouped about half its Cursor investment via compute deals, with the rest covered if Composer 3 does well. No other company is simultaneously a leading lab and a Neo cloud where GPUs are concerned.
Google stock fell as much as 7.2% Monday — its largest intraday move since February — after Nobel laureate John Jumper left DeepMind for Anthropic and Noam Shazeer left for OpenAI. NLW thinks the market over-indexes on AI headlines, but a real narrative shift around Google's pace is underway.
— Gil Luria, D.A. Davidson. D.A. Davidson's Gil Luria argued Google held a state-of-the-art model for only weeks last year and has fallen off since, with these departures suggesting it is falling further behind on enterprise-essential agentic use cases.
Citing Prime Intellect's Florian, NLW notes the recurring cycle: 'peak Google is done for' is always followed by Google releasing a new model and everyone deciding no one can match its TPUs and data. For now it's for markets to debate which side is right.
