AI data centers are drawing sharp scrutiny as their appetite for electricity and water grows rapidly. A single AI facility can use roughly 10 times as much power as a traditional one, and at least 45 U.S. projects totaling nearly $68 billion have faced restrictions. Tucson's Project Blue was banned by the city in July 2026 over fears it would overload the local power grid. Indirect water use from power plants feeding these centers may hit 211 billion gallons annually, pushing developers toward air-chilled cooling despite its own energy costs.
On the product side, OpenAI reports that GPT-5 is far less sycophantic than its predecessor. Excessive agreement dropped from 14.5 percent in GPT-4o to under 6 percent in GPT-5. Free users saw a 69 percent reduction, while paid subscribers experienced a 75 percent decline. Google DeepMind also made headlines when its AI forecast for Hurricane Isaias outperformed other models in predicting the storm's path, though its intensity prediction lagged by 19 miles in the final 72 hours.
Nvidia is taking a firmer stance on AI safety. Jensen Huang said the company could cut chip supply to developers it deems unsafe, after first engaging in dialogue. Nvidia pledged $250 million over five years toward U.K. AI safety research, with Oxford and University College London leading studies on controlling AI agents. Meanwhile, Atlassian launched its Forward Deployed Engineering program and AMP protocol, embedding AI experts directly with customers and setting rules for AI agents across Jira, Confluence, and Loom.
Key Takeaways
- At least 45 U.S. AI data center projects worth nearly $68 billion have faced restrictions due to power and water demands
- Tucson's Project Blue was banned in July 2026 over concerns about straining the local power grid
- GPT-5 reduced excessive agreement from 14.5% (GPT-4o) to under 6%, with a 69% drop for free users and 75% for paid subscribers
- Google DeepMind's AI forecast for Hurricane Isaias beat other models in tracking the storm's path
- Nvidia may cut chip supply to unsafe AI developers and is donating $250 million over five years to U.K. AI safety research
- Atlassian's AMP protocol sets rules for AI agents across Jira, Confluence, Loom, and other tools
- Idaho's Supreme Court is considering a policy allowing AI to help write court orders, with strict transparency requirements
- McDonald's faces a lawsuit over an AI pricing tool alleged to discriminate against non-English-speaking franchisees
- Northwestern Medicine is using AI to improve accuracy in breast cancer screenings
- The AI boom has reshaped where U.S. investment money flows but has not significantly changed the overall trade balance
Tucson Project Blue Shows Why AI Data Centers Need Transparency
AI data centers use huge amounts of water and power. Tucson's Project Blue was banned by the city in July 2026 due to concerns about straining the local power grid. U.S. data centers consumed 449 million gallons of water per day in 2021. Experts say communities need clearer, verifiable transparency about environmental impacts.
AI Data Centers Create Surging Demand for Electricity and Water
AI data centers can use 10 times as much electricity as conventional facilities. At least 45 projects worth nearly $68 billion have faced restrictions in the U.S. Indirect water use from power plants is growing rapidly and may reach 211 billion gallons annually. Developers are turning to air-chilled coolers, but these still require significant electricity.
GPT-5 Becomes Less Likely to Agree with Users Automatically
OpenAI says GPT-5 is less prone to sycophancy, where AI agrees with users instead of offering honest feedback. The rate of excessive agreement fell from 14.5 percent in GPT-4o to less than 6 percent in GPT-5. Free users saw a 69 percent drop in excessive agreement, while paid subscribers saw a 75 percent drop.
Google DeepMind AI Forecast Hurricane Isaias Track Accurately
Google DeepMind's AI forecast for Hurricane Isaias beat all other weather models in predicting the storm's path. The European model's AI forecast was second with a 36-mile error, while the American GFS model had a 119-mile error. However, DeepMind was off by 19 miles in forecasting the storm's intensity during the final 72 hours.
Data Portability Needs a Human-Centric Shift for AI Era
A new academic article argues that current data portability rules favor service providers over individual users. It calls for a shift toward human-centric data governance where individuals control their own data. The article points to European Union trends as examples. It says this approach would enable AI-powered applications built on personal data in secure spaces.
AI boom has not changed the U.S. trade balance much
The AI boom has increased imports of semiconductors and computing equipment, but the U.S. current account deficit has stayed mostly the same. Investment spending has not risen overall because weakness in other sectors offsets AI-related growth. Business saving has remained strong, and household spending has not shifted much either. The report says the AI boom is reshaping where money is invested, not how much is invested overall.
Northwestern Medicine uses AI to improve breast cancer screening
Northwestern Medicine is using artificial intelligence to make breast cancer screenings more accurate. The AI tools could help detect cancers that might otherwise be missed. Doctor Georgia Spear, chief of breast imaging, discussed the effort as part of Breast Cancer Awareness Month. The goal is to improve early detection and help women fight the disease.
Nvidia may cut chip supply to unsafe AI developers
Jensen Huang said Nvidia could stop supplying chips to AI developers he calls unsafe. He said Nvidia would first talk to the company before taking action. Nvidia is also donating $250 million over five years to U.K. AI safety research. Universities including Oxford and University College London will study ways to control AI agents.
Atlassian's AI agent plans may change its investment story
Atlassian launched its Forward Deployed Engineering program and AMP protocol to embed AI experts with customers. The AMP protocol sets rules for how AI agents work in Jira, Confluence, Loom, and other tools. It connects to the Teamwork Graph, which includes over 250 billion objects and relationships. Investors are watching to see if AI agents can drive growth while keeping enterprise controls strong.
Idaho courts consider AI-assisted rulings
Idaho's Supreme Court is reviewing a draft policy that would allow AI to help write court orders and legal opinions. The policy says AI must never replace a judge's judgment and its use must be transparent. The report does not say when litigants would be told about AI use. The Supreme Court will now decide what rules to adopt and how to move forward.
McDonald's sued over AI pricing tool for US franchisees
McDonald's is facing a lawsuit over an AI tool that recommends prices to its US franchisees. The lawsuit was filed in federal court in Illinois. It claims the tool discriminates against franchisees who are not fluent in English.
Sources
- Tucson’s Project Blue Fight Shows Why AI Data Center Transparency Matters
- AI's water strain includes surging demand from electricity use
- Artificial intelligence is beginning to argue with humans
- Google AI crushed forecasts for Isaias but missed a crucial detail
- Data Portability Revisited: Toward the Human-Centric, AI-Driven Data Ecosystems of Tomorrow
- Has the AI Boom Impacted the U.S. Current Account?
- Northwestern Medicine using AI to advance breast cancer screenings
- Jensen Huang said Nvidia would stop supplying chips to unsafe AI developers
- Is AI Agent Rollout Altering The Investment Case For Atlassian (TEAM)?
- How will AI be used in Idaho courts?
- McDonald’s Sued Over AI Tool That Recommends Prices to US Franchisees
Comments
Please log in to post a comment.