One Fallen Power Line Exposes How Data Centers Can Disrupt the Grid

01One Fallen Power Line Exposed Data Centers’ Grid-Disruption Problem as EPA Weighs Limiting Public Input

A power line fell in Northern Virginia, and a routine grid failure became a warning for the region’s data centers. The disruption nearly spun out of control, according to TechCrunch, exposing how poorly the facilities respond when their electricity supply suddenly breaks.

The incident did not expose a shortage of generation. It exposed a failure in handling interruptions. Data centers can bring concentrated demand onto a grid, but their resilience plans must also account for the moment that power disappears unexpectedly.

That gap carries more weight as AI companies seek additional data-center capacity. Each new facility adds another operator whose equipment, backup systems, and response procedures must work during a grid disturbance. Local utilities also inherit more demand that cannot be treated like an ordinary commercial load.

Fixing the problem therefore extends beyond repairing the line that fell. Operators need disruption plans that account for abrupt grid failures. Utilities need to test how data centers respond under stress. Local governments need evidence that proposed facilities can manage outages before approving further construction.

Residents may have fewer opportunities to press for those answers. Ars Technica reports that a proposed rule from the Trump administration’s EPA would let states decide how much public input to permit, including whether to allow any.

That proposal would shift a central part of data-center oversight toward state policy. One state could preserve hearings and public comments. Another could restrict them, leaving neighboring communities with less access to the approval process as new facilities seek permission to build.

The timing connects two separate pressure points. Northern Virginia’s close call showed that physical infrastructure can fail before operators respond effectively. The EPA proposal could narrow one channel for examining those preparations while AI firms continue pursuing more sites.

For operators, resilience documentation may become as consequential as access to power. State and local officials will need to decide which outage scenarios applicants must address. Residents will need to track those state-level decisions before individual projects enter approval pipelines.

The next test will not be another corporate capacity announcement. It will be whether states preserve public review and whether operators can show that a failed line will not become a wider grid emergency.

0221 tech companies cite AI in layoffs; research finds little aggregate employment impact

TechCrunch has counted 21 major tech employers that cited AI as a factor in significant layoffs announced this year. The tally gives executives a crisp explanation: automation changed the staffing equation. Stanford researchers looking across the labor market reach a far narrower finding. AI’s effect on aggregate employment is likely small for now.

Those claims measure different things. A company can reorganize while adopting AI and name the technology in its announcement. That does not establish how many eliminated roles were directly replaced by models. Nor does a list of layoffs show whether displaced workers found jobs elsewhere. Corporate attribution is evidence of management decisions, not a national displacement rate.

The Stanford brief does not dismiss the threat. It describes firm adoption as faster but uneven across the economy. Research on worker productivity is mixed, though generally positive. Those gains can change hiring plans before they produce broad job losses. They can also arrive alongside budget cuts, weaker demand, or ordinary restructuring.

The clearest pressure point may be the entrance to the labor market. Stanford says the difficult market for new graduates may be partly due to AI, while warning that early evidence is not definitive. Aggregate employment can stay stable even if employers hire fewer junior workers. Existing employees and experienced applicants may conceal that damage in topline figures.

The scale gap is large. Anthropic CEO Dario Amodei has predicted AI could erase half of white-collar jobs and drive unemployment to 20%. The brief presents that forecast against current empirical evidence, which does not show upheaval near that magnitude. Today’s layoff statements therefore say more about individual company choices than economy-wide causation.

For job seekers, the distinction changes where to look. Graduate hiring rates, entry-level postings, and employment outcomes by age and occupation will test whether pressure is concentrating at the bottom rung. For managers, citing AI does not quantify substitution without disclosing which tasks disappeared, what systems absorbed them, and whether output held steady.

03Devin Is Buying Poke’s Personality as Midjourney Buys an Astrology App

Two AI acquisitions are shifting competition away from model capability alone. Cognition bought Poke to bring its conversational style and interaction model into Devin. Midjourney acquired Co-Star, a personalized astrology app with daily horoscopes and an existing consumer relationship.

The assets differ, but both sit between a model and its users. Poke offers a way of speaking and responding that Cognition can apply to a coding agent. Co-Star offers a recurring, personalized experience beyond Midjourney’s core image and video products. Neither acquisition primarily adds a new foundation model.

That distinction matters for products built on increasingly comparable technical inputs. Developers experience Devin through its requests, explanations, interruptions, and responses to failure. Poke gives Cognition an interaction model it can incorporate into those moments. Task performance still sets the floor, but conversational behavior can influence whether users keep an agent inside their daily workflow.

Co-Star gives Midjourney a different route. The free app already packages personalized output into a daily habit. Its subject is astrology, not media generation, which takes Midjourney into a consumer category where the continuing user relationship is the product. The acquisition adds an application and an established entry point rather than another generation format.

Together, the deals identify two assets that model releases cannot instantly reproduce: a recognizable interaction style and a direct user habit. One can make a work agent feel distinct even when rivals offer similar coding functions. The other can give an AI company regular contact with consumers outside its original product category.

The available reports do not disclose transaction prices or detailed integration schedules. Cognition’s next visible test is whether Poke’s conversational model changes how Devin behaves during coding work. For Midjourney, the key signal will be whether Co-Star stays a separate destination or connects to a broader personalized consumer product.