Evening AI News Recap — Monday, March 23, 2026

Affiliate disclosure: We earn commissions when you shop through the links on this page, at no additional cost to you.

Maya ChenAI Researcher & Product Reviewer

Monday closes on a day that touched nearly every dimension of AI’s expanding footprint: its thirst for energy, its contested role in warfare, the strange philosophy it’s inspiring in unexpected places, and the urgent question of who — or what — deserves moral consideration as these systems grow more powerful. Here’s what stood out this evening, with the context that makes it matter.

1. Europe’s Power Grids Are Buckling Under the Weight of AI’s Appetite

A new deep-dive from Wired lays out the scale of a crisis that’s been building quietly but is now becoming impossible to ignore: Europe’s electricity grid infrastructure simply cannot keep up with the pace at which AI data centers want to connect to it. The numbers are staggering — National Grid, which operates England and Wales’s transmission network, currently has proposed data center projects representing more than 30 gigawatts of power demand waiting in the connection queue. That’s roughly two-thirds of Great Britain’s entire peak electricity demand, sitting on a waitlist.

The core problem isn’t generation — Europe is broadly on track to produce enough electricity, thanks to a rapid expansion of wind and solar capacity. The bottleneck is transmission: the physical infrastructure to move power from where it’s generated to where it needs to go simply hasn’t been built fast enough. Grid operators across Germany, the Netherlands, Ireland, and the UK are all grappling with versions of the same constraint, and the result is that some data center projects are quietly collapsing before they’re ever built. Amazon recently acknowledged that grid congestion had made certain European infrastructure projects unfeasible.

To cope, network operators are experimenting with novel approaches: “flexible connection” agreements that allow data centers to draw full power most of the time but accept curtailment during peak periods; incentive schemes that reward large customers for shifting load overnight; and accelerated permitting for new transmission lines. None of these are quick fixes. Building new high-voltage transmission corridors in densely populated Europe takes years and encounters fierce local opposition at nearly every stage.

Why it matters: The AI compute race has collided with physical reality in a way that quarterly earnings calls don’t capture. European governments are simultaneously pushing AI sovereignty strategies — the EU’s AI Factories initiative, the UK’s AI Opportunities Action Plan — and watching the enabling infrastructure fall behind. The irony is sharp: the regions most eager to compete with the US and China on AI are the ones where legacy grid architecture is most constraining. This story will define AI policy in Europe for the rest of the decade. The companies that figure out how to colocate renewable generation with compute — rather than relying on stressed transmission grids — will have a material advantage. Full story at Wired →

For developers building AI workloads who need reliable, cost-effective hosting in Europe, Contabo VPS offers robust European infrastructure that’s been quietly keeping pace with demand.

2. Project Maven’s Quiet Evolution: How the Pentagon Became a True Believer in AI Warfare

A book excerpt published in Wired today offers one of the most substantive accounts yet of how Project Maven — the Pentagon’s AI initiative that sparked a firestorm inside Google in 2018 — evolved from a contested experiment into a cornerstone of US military operations. Author Katrina Manson, drawing from years of reporting for her book Project Maven: A Marine Colonel, His Team, and the Dawn of AI Warfare, traces the transformation through the career of Marine Colonel Drew Cukor, the project’s founding leader.

The story begins with the original premise: using computer vision to process the enormous volumes of drone footage coming out of America’s overseas wars — footage that human analysts couldn’t review fast enough to be tactically useful. Google’s internal protest in 2018, which led the company to withdraw from the project, forced a reckoning inside the Pentagon. Rather than slowing the initiative, the controversy accelerated it: defence officials concluded they couldn’t rely on commercial tech companies with conflicted workforces, and began building more direct relationships with vendors willing to commit fully to the mission.

The result today is Maven Smart System — an AI platform now confirmed to be in active use in US operations against Iran. The tool has evolved well beyond computer vision: it integrates targeting intelligence, fuses data from multiple sensor types, and surfaces recommendations to human decision-makers in near-real-time. Former skeptics within the Pentagon, the excerpt reveals, have become some of its most vocal advocates — not because the ethical questions have been resolved, but because the operational advantages are simply too significant to step back from.

Why it matters: This story intersects directly with this morning’s coverage of the Anthropic-Pentagon dispute, but from a completely different angle. Where that story asked whether commercial AI vendors can be trusted by the military, this one asks a deeper question: how does an institution as hierarchical and risk-averse as the US military become, in less than a decade, structurally dependent on AI systems it barely understood when it started building them? The answer — pragmatic success overrides principled caution — has implications far beyond defence. It’s the pattern that plays out wherever AI delivers measurable operational wins: adoption outpaces governance, skeptics become converts, and the ethical debates get deferred. Read the excerpt at Wired →

3. The Sentient Futures Summit: Animal Welfare Advocates Want AI to Care About Animals — And Wonder If It Already Does

In a story that cuts against almost every other narrative in today’s AI news cycle, MIT Technology Review reports on a gathering that took place in early February at a San Francisco coworking space: the Sentient Futures Summit, hosted by an organisation of the same name, which believes that the future of animal welfare will be shaped — possibly determined — by how AI systems are trained to value non-human life.

The attendees weren’t the typical crowd. They included animal welfare advocates with long records in effective altruism, AI researchers, wildlife conservationists, and philosophers of mind debating whether insect sentience can tell us something about the inner life of a language model. The “Crustacean Room” saw a circle of people working through that exact question. Constance Li, Sentient Futures’ founder, framed the core thesis: if AI systems are going to make an increasing proportion of decisions in the world — allocating resources, optimising supply chains, shaping policy — then what those systems are trained to value will determine outcomes for billions of sentient animals who have no voice in that training process.

The movement is closely linked to effective altruism and has a decidedly long-termist orientation: its participants aren’t primarily focused on immediate animal welfare interventions but on ensuring that whatever post-AGI world emerges is one where animal suffering registers as something worth minimising. That means influencing how large models are trained, what reward signals they optimise for, and how interpretability research develops. There’s also a more immediate angle: AI is already being used to accelerate the development of cultivated meat, to monitor factory farming conditions, and to model animal cognition in ways that weren’t previously possible.

Why it matters: This is the AI alignment debate extended to its logical conclusion. If we’re worried about AI systems developing values misaligned with human flourishing, the same framework applies to the 70+ billion land animals raised in factory farming systems every year. The effective altruist framing is controversial — “maximising good” gets philosophically messy quickly — but the underlying concern is coherent: training data and reinforcement signals embed values, and no one is deliberately ensuring those values include animal welfare. As AI systems become more powerful, what they implicitly optimise for will matter more. The Sentient Futures crowd is placing an early bet that now is the time to shape that. Whether or not you find their conclusions persuasive, the conversation they’re having is one the field will eventually have to take seriously.

4. AI Drug Discovery Gets Another Serious Vote of Confidence: Converge Bio Raises $25M

AI drug discovery startup Converge Bio has closed a $25 million Series A, led by Bessemer Venture Partners, with additional participation from senior executives at Meta, OpenAI, and cybersecurity firm Wiz. The round is a notable signal of continued institutional confidence in the AI-for-drug-discovery thesis, even as the sector faces mounting pressure to demonstrate clinical results rather than just faster compound screening.

Converge Bio’s approach centres on using AI to identify novel drug targets and optimise compounds for therapeutic potential — the two phases of early drug development that have historically been the most time-intensive and failure-prone. The backing from OpenAI executives is worth noting: it suggests that some of the people closest to frontier AI model development believe the technology is genuinely mature enough to accelerate pharmaceutical research at a meaningful level, not just as a productivity tool but as a discovery engine.

The broader AI drug discovery landscape has had a complicated year. AlphaFold’s transformative impact on protein structure prediction has not yet translated cleanly into clinical successes, and several early-stage companies have quietly restructured after raising large rounds on the strength of in-silico results that didn’t survive contact with biology. Converge’s Series A, while relatively modest, suggests a recalibration toward credible near-term milestones rather than moonshot claims. For those building AI-powered research and automation pipelines in biotech, platforms like n8n are increasingly being used to connect data pipelines between research platforms, lab systems, and analysis tools.

Why it matters: Drug discovery is one of the few domains where AI could plausibly save millions of lives — if the models actually work in the messy reality of biological systems. The Converge Bio raise is a small data point, but the calibre of its backers (Bessemer has a strong track record in healthcare tech; the Meta and OpenAI executive angels bring credibility with the technical community) suggests more than just financial momentum. It reflects a genuine belief that the tooling has improved enough to justify serious capital deployment. Watch for the company’s first disclosed drug candidate announcement as the next real test of that conviction.

What to Watch Tomorrow

EU AI Act enforcement milestones: Tuesday marks a key implementation checkpoint for the EU AI Act’s prohibited practices provisions, which came into full force in February. Expect official commentary from the European AI Office on early compliance activity and any enforcement signals — particularly around biometric categorisation systems and social scoring.
UK AI Opportunities Action Plan updates: The UK government has been drip-feeding implementation details from its January AI action plan. With the grid capacity story breaking today, pressure on the government to fast-track transmission planning permissions for AI data centres is likely to intensify.
Project Maven book release: Katrina Manson’s full book hits shelves this week. The Wired excerpt today was a teaser — the complete account of how AI became embedded in US military targeting infrastructure is likely to generate significant coverage and congressional attention.
AI sentience research: The Sentient Futures story is a harbinger. With several interpretability research papers expected this quarter from Anthropic and others, the question of whether current AI systems have morally relevant inner states will get a fresh empirical angle. Keep an eye on the ACL 2026 paper calendar.

Monday’s evening digest covers the four stories that didn’t make the afternoon cut but arguably carry the longer arc: energy infrastructure, the military’s decade-long conversion to AI dependency, the philosophy of machine values, and the continued professionalisation of AI drug discovery. Different corners of the same expanding map.

Image: AI-generated

What to Read Next

Bookmark aistackdigest.com for daily AI tools, reviews, and workflow guides.

This article was produced with the assistance of AI tools and reviewed by the AIStackDigest editorial team.