AI News Roundup: Meta Secures 6.6 GW Nuclear Power in Historic AI Energy Deal
Meta just made one of the largest corporate energy commitments in American history. The social media giant secured up to 6.6 gigawatts of nuclear power through 2035 - enough to power roughly 5 million homes - all to feed the insatiable appetite of AI data centers. Meanwhile, MIT’s annual breakthrough list confirms what developers already knew: AI is now writing a significant chunk of the world’s code.
Here’s everything that matters from January 11-12, 2026.
The Big Story: Meta’s Historic Nuclear Push for AI Dominance
Meta announced agreements with Vistra, TerraPower, and Oklo to secure up to 6.6 gigawatts of nuclear power by 2035, making it one of the largest corporate purchasers of nuclear energy in American history.
The deal package is massive in scope. It includes long-term purchases from existing reactors, funding for plant upgrades to extend operational life, and support for advanced reactor development. TerraPower will provide two Natrium units (690MW) by 2032, with rights to six more units totaling 2.1GW by 2035. Oklo’s Aurora powerhouses will add up to 1.2 GW of additional capacity.
The energy will power Meta’s “Prometheus” AI supercluster in Ohio - the infrastructure backbone for next-generation AI training. As CBS News reported, this move positions Meta to avoid the energy constraints that are increasingly becoming the limiting factor in AI development.
The timing isn’t coincidental. With memory sold out through 2026 and AI scaling hitting physical limits, securing reliable baseload power has become a strategic imperative. Meta’s bet on nuclear - including advanced reactors that won’t come online until the late 2020s - signals a long-term commitment to AI infrastructure that goes beyond the current hype cycle.
Sources: Meta Newsroom, CBS News
Today’s Top Stories
MIT Names Generative Coding a 2026 Breakthrough
MIT Technology Review released their annual 10 Breakthrough Technologies list, and AI dominates with two entries: Hyperscale AI Data Centers and Generative Coding.
The recognition of generative coding as a breakthrough technology reflects how fundamentally AI is changing software development. MIT called out both the scale of AI-assisted coding adoption and the massive engineering achievements behind hyperscale data centers, which now bundle hundreds of thousands of GPUs into single facilities.
The list validates what many developers have quietly known: the nature of software development is fundamentally changing, faster than most anticipated.
Source: MIT Technology Review
Anthropic Blocks Third-Party IDE Extensions, Sparks Developer Revolt
Anthropic deployed strict technical safeguards blocking subscription OAuth tokens from working outside their official Claude Code CLI. Users paying $100-200/month for premium plans are now forced to use Anthropic’s terminal-based tool exclusively.
The backlash has been swift and severe. Hacker News discussions generated 245 points with developers announcing subscription cancellations. DHH called the move “very customer hostile.” GitHub issues show 147+ reactions from frustrated users who integrated Claude into their preferred development environments.
The crackdown appears aimed at preventing third-party “harnesses” from accessing Claude’s capabilities, but it’s caught legitimate power users in the crossfire. Developers who built workflows around IDE integrations now face a choice: adapt to Anthropic’s CLI or leave the ecosystem entirely.
Sources: VentureBeat, Hacker News
Healthcare AI Race Heats Up: Anthropic Launches HIPAA-Compliant Features
Days after OpenAI debuted ChatGPT Health, Anthropic announced HIPAA-compliant healthcare features for Claude. The new capabilities allow users to share information from official medical records and fitness apps like Apple Health directly with the AI assistant.
The features are available now for Claude Pro and Max subscribers in the US. SiliconANGLE reports that the focus is on helping patients understand their own medical records - a use case that could democratize access to medical interpretation that typically requires expensive specialist consultations.
The timing signals that healthcare AI is becoming a key battleground. With 46% of all healthcare tech investments in 2025 going to AI and the JPMorgan Healthcare Conference kicking off today with AI as a central theme, expect the competition to intensify.
Sources: Bloomberg, SiliconANGLE
France Gets Its First Defense AI Unicorn
Harmattan AI raised $200 million in Series B funding led by Dassault Aviation, valuing the Paris-based defense AI startup at 1.4 billion euros. The investment makes Harmattan France’s first defense unicorn.
The partnership is strategic: Dassault will integrate Harmattan’s AI capabilities into combat aviation systems, including the future Rafale F5 fighter jet. French President Emmanuel Macron celebrated the announcement, highlighting it as a win for European defense sovereignty.
The funding comes as defense AI investment accelerates globally. Harmattan’s focus on autonomous systems for military applications represents the kind of high-stakes AI deployment that’s attracting both capital and regulatory attention.
Sources: TechCrunch, Bloomberg
AI Solves Multiple “Unsolvable” Math Problems
In what may be the most significant demonstration of AI reasoning capabilities to date, OpenAI’s GPT-5.2 has reportedly solved multiple previously unsolved Erdos mathematical problems autonomously.
Since late December, AI has cracked problems at an accelerating pace: #1077 (Dec 24), #333 (Dec 25), #728 (Jan 6), #720 (Jan 8), and #205 and #39 (Jan 10). Mathematician Terence Tao confirmed that Erdos problem #728 was “solved more or less autonomously by AI.”
Paul Erdos famously offered cash prizes for solving his problems, some dating back decades. The fact that AI is now solving these faster than human mathematicians raises profound questions about the future of mathematical research.
Sources: The Neuron Daily, Office Chai
Quick Hits
-
Windows 11 AI Update: Microsoft released the first Insider build of 2026 featuring Copilot-powered image descriptions for the Narrator accessibility feature.
-
Robotaxi Reboot: Motional announced an AI-first approach to its robotaxi plans, targeting Las Vegas driverless service by end of 2026.
-
JPMorgan Healthcare Conference: The 44th Annual JPMorgan Healthcare Conference kicked off in San Francisco with AI-driven drug discovery and diagnostic tools as central themes.
-
xAI Restricts Grok Images: Following content moderation concerns, xAI restricted Grok’s image generation to paid X subscribers only.
What This Means
Two themes dominate this weekend’s news: infrastructure is becoming the moat, and ecosystem control is intensifying. Meta’s nuclear deal isn’t virtue signaling - it’s securing the physical resources that will determine who can train the largest AI models, while Anthropic’s IDE crackdown and OpenAI’s contractor data practices show platforms getting aggressive about controlling access to their models. Meanwhile, AI solving decades-old Erdos problems “more or less autonomously” (per Fields medalist Terence Tao) is the kind of story that deserves far more attention - the question isn’t whether AI will transform research, but how quickly.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
Get Carly Today →
