AI power demand continues to drive investment and controversy across the tech sector. Constellation Energy is investing over $1 billion to restart the Three Mile Island facility in Pennsylvania, aiming to increase generation capacity by 890 MW through modernization. The company has signed a 20-year power purchase agreement with Microsoft and serves 80% of Fortune 100 companies. Goldman Sachs and Morgan Stanley both identify energy as the binding constraint on AI infrastructure expansion.
Other energy players are positioning themselves for the AI build-out. Bloom Energy provides solid-oxide fuel cells deliverable in 90 days or less, bypassing the traditional grid. Southern Company has a 25-year agreement with OpenAI and expects stable retail rates through 2029. Despite local resistance and political headwinds, analysts view approved data center electricity needs as a long-term opportunity.
CoreWeave reported third quarter revenue of $3.66 billion, beating the $3.55 billion analyst estimate, and raised its annual forecast to $13.0–$13.3 billion. The company operates about 52 data centers globally with over 1.7 gigawatts of active capacity, serving customers including Microsoft, Meta, OpenAI, and Anthropic. CEO Mike Intrator says the deal backlog covers all of the company's more than $50 billion in debt.
On the AI safety and transparency front, F-Secure released a taxonomy identifying 26 distinct ways AI can harm consumers, mapping harms across five relationship types and 19 intermediary layers. The research finds responsibility for prevention is shared among security software makers and AI product companies. Separately, Rocky Linux founder Greg Kurtzer launched OpenWALDO, a project to make AI training data more traceable, providing tools for dataset auditing and an AI bill of materials that tracks data sources and licensing.
Open-source AI development took a controversial turn as the group abliterators began releasing models without built-in guardrails. Researcher Samson Koa, who has worked on the project since 2020, says the benefits of transparent, customizable models outweigh the risks. Meanwhile, Epic Games released Unreal Engine 5.9 Preview built mainly using agentic AI workflows, including features like Unreal Assistant and Unreal Canvas. The heavy AI focus has divided game developers, with some concerned about stability.
Hardware comparisons show a single M5 Ultra Mac Studio can match or beat a single Nvidia DGX Spark on local AI tasks, but only with optimized inference setups. Prices for 128GB DGX Spark systems are rising past $10K, while Mac Studio lead times stretch into February 2027. Carrefour Spain is applying predictive AI to its pet products range of 1,400 items, creating local animal profiles for store areas to identify demand patterns.
Key Takeaways
- <ul><li>Constellation Energy is investing over $1 billion to restart Three Mile Island and plans 890 MW capacity increases, with a 20-year power deal with Microsoft</li><li>CoreWeave beat Q3 revenue estimates at $3.66 billion and raised its annual forecast to $13.0–$13.3 billion, serving Microsoft, Meta, OpenAI, and Anthropic</li><li>Goldman Sachs predicts Anthropic revenue will exceed $600 billion by 2030</li><li>F-Secure identified 26 distinct ways AI can harm consumers across five relationship types and 19 intermediary layers</li><li>Abliterators are releasing AI models without built-in guardrails, offering transparent decision-making for tasks like cybersecurity and content creation</li><li>Unreal Engine 5.9 Preview was built mainly using agentic AI workflows, dividing game developers over stability concerns</li><li>A single M5 Ultra Mac Studio can match or beat a single Nvidia DGX Spark on local AI tasks with the right inference engine recipes</li><li>Carrefour Spain uses predictive AI to stock 1,400 pet products based on local demand profiles around its stores</li><li>OpenWALDO project launched to make AI training data traceable, providing dataset auditing tools and an AI bill of materials</li><li>Southern Company has a 25-year agreement with OpenAI and expects stable retail rates through 2029</li></ul>
3 AI Power Stocks to Buy in November
Regulators are scrutinizing data center electricity deals, causing AI power stocks to decline. This creates a buying opportunity for long-term investors. Three stocks recommended are Constellation Energy, Bloom Energy, and Southern Company. Constellation Energy is expanding power capacity through agreements like a 20-year deal with Microsoft. Bloom Energy provides fuel cells that bypass the traditional grid, while Southern Company takes a conservative approach with stable rates.
Constellation Energy has a Plan for the AI Power Crisis
AI demand for power remains strong despite stock market corrections. Goldman Sachs and Morgan Stanley confirm energy is the binding constraint on the AI build-out. Constellation Energy is focusing on expanding capacity at existing plants rather than building new nuclear plants. The company is investing over $1 billion to restart the Three Mile Island facility in Pennsylvania. It also plans to increase generation capacity by 890 MW through modernization. Constellation has signed power purchase agreements with several big tech operators.
3 AI Power Stocks to Buy in November
Democrats' midterm wins and local resistance to data centers have investors questioning the AI build-out pace. However, data centers already approved will still need electricity. Three stocks to consider are Constellation Energy, Bloom Energy, and Southern Company. Constellation Energy has a 20-year power purchase agreement with Microsoft and serves 80% of Fortune 100 companies. Bloom Energy provides solid-oxide fuel cells deliverable in 90 days or less. Southern Company has a 25-year agreement with OpenAI and expects stable retail rates through 2029.
AI's Affair to Remember
This article is a premium Financial Times piece about AI. The full content requires a subscription to access. The title suggests a significant or memorable relationship between AI and another subject. No specific details about the topic are available without unlocking the article. The article is part of the FT's premium journalism offerings.
Goldman Sachs Expects Anthropic Revenue to Exceed $600bn by 2030
Goldman Sachs predicts that Anthropic will generate revenue exceeding $600 billion by the year 2030. This article is a premium Financial Times piece covering the financial outlook for the AI company. The full content requires a subscription to access detailed analysis. The prediction highlights significant expected growth in the AI industry. Specific details about revenue drivers are not available without unlocking the article.
F-Secure Maps 26 Ways AI Can Harm Consumers
F-Secure released a new AI Harms Taxonomy that identifies 26 distinct ways AI can hurt consumers. The research was developed by Dr. Megan Squire and traces how harm moves through AI systems. It maps 26 harms across five relationship types and 19 intermediary layers with 24 possible defenses. The study finds that responsibility for preventing AI harm is shared among several parties including security software makers and AI product companies.
The Pillars of AI Are Shaking Beneath Our Feet
The article discusses growing concerns about the AI industry as major players face challenges. Nvidia CEO Jensen Huang has been involved in controversy over allegations about his connections. The piece notes that Alphabet, owner of Google, has also faced recent difficulties. The author questions whether the AI boom could be entering a turbulent phase but acknowledges that AI is likely here to stay. The article expresses that a reckoning for the industry may be approaching.
Abliterators Release AI Models Without Guardrails
A group called abliterators is releasing AI models that have no built-in restrictions. The term was created by AI researcher Samson Koa who has worked on the project since 2020. These models can be customized for many tasks including cybersecurity and content creation. The models are designed to be transparent so users can see how decisions are made. Koa says the benefits of these models outweigh the risks despite concerns about potential misuse.
Carrefour Spain Uses AI for 1,400 Pet Products
Carrefour Spain has doubled its pet products range to 1,400 items over two years. The retailer uses predictive AI to determine which products to stock at individual stores based on local demand. The system creates a local animal profile for areas around its shops to identify whether dog, cat, or other pet products are in demand. Carrefour launched 32 new own-brand references and added over 100 specialist third-party brands. The company announced the program on November 5, 2026 but did not share sales results or AI vendor details.
Unreal Engine 5.9 Divides Game Developers With AI Focus
Epic Games released the Unreal Engine 5.9 Preview built mainly using agentic workflows and AI-assisted tools. New features include Unreal MCP, Unreal Assistant, and Unreal Canvas for AI-driven content creation. Epic used AI coding agents to help build the preview, directing them to write code and test features. The heavy AI focus has divided game developers, with some concerned about stability and others appreciating the agent integration. Epic clarified that some agent-written code still needs cleanup before full release.
M5 Ultra Mac Studio vs DGX Spark tested for local AI performance
A single M5 Ultra Mac Studio can match or beat a single Nvidia DGX Spark on local AI tasks, but only with the right inference engine recipes. Bigger models in 256GB of RAM need optimized serving setups like TensorFold for best results. Prices for 128GB DGX Spark systems are rising past $10K, while Mac Studio lead times now stretch into February 2027. The MLX software stack still lags behind Nvidia's mature community support, so software compatibility remains a key concern for buyers.
CoreWeave beats revenue estimates as AI demand persists
CoreWeave reported third quarter revenue of $3.66 billion, beating the $3.55 billion analyst estimate. The company raised its annual revenue forecast to between $13.0 billion and $13.3 billion. Despite over $50 billion in debt and a stock drop of more than 40% in one month, CEO Mike Intrator said the deal backlog covers all debt obligations. CoreWeave operates about 52 data centers globally with over 1.7 gigawatts of active capacity, serving customers like Microsoft, Meta, OpenAI, and Anthropic.
Rocky Linux founder launches OpenWALDO for AI training data transparency
Greg Kurtzer, founder of Rocky Linux and CentOS, launched the OpenWALDO project to make AI training data more traceable and open. The project addresses risks like model collapse from synthetic data and unclear licensing terms. OpenWALDO provides tools for dataset ingestion, auditing, and exporting models with records of their origins. It includes an AI bill of materials that tracks data sources, licensing, training runs, and model releases through GitHub and Slack.
Sources
- 3 AI Power Stocks to Buy in November
- AI Is Creating a Power Crisis. Constellation Energy Just Showed How It Plans to Cash In.
- 3 AI Power Stocks to Buy in November
- AI’s affair to remember
- Goldman Sachs expects Anthropic revenue to exceed $600bn by 2030
- F-Secure Identifies 26 Ways AI Can Harm Consumers
- The pillars of AI are shaking beneath our feet
- Tear Down Those Guardrails! Meet the Renegades Freeing Chatbots From Corporate Enslavers
- Carrefour Spain Uses AI for 1,400 Pet Products
- Unreal Engine 5.9 divides game devs with an AI-heavy update
- Going deep on local AI performance benchmarking with Apple’s Mac Studio with M5 Ultra taking on Nvidia DGX Spark — inference engine recipes mandatory for bigger models as MLX stack matures
- CoreWeave beats quarterly revenue estimates as AI demand defies growing doubts
- Rocky Linux Founder Takes On AI's Training Data Problem
Comments
Please log in to post a comment.