Senate Passes Energy Permitting Bill 62-33 and Makes Data Centers Pay for Their Power

Image
Photo: blickpixel / Pixabay November 21, 2026 · Energy The Senate passed a bipartisan bill on Thursday, November 19, to speed up federal permits for power plants, transmission lines and pipelines. The vote was 62–33. The bill also requires data centers to pay for the energy they use. It is the first major action by Congress since the midterm elections, and it now goes to the House. What the bill does Faster reviews: it shortens the timeline to start large energy projects. Environmental reviews under current law often take five years or more. Data centers: it is written to make sure data centers cover their own energy costs. Anger over rising electricity bills was a major issue in the elections. Law changes: it revises the National Environmental Policy Act, the Clean Water Act and the Endangered Species Act, and limits court challenges to approved projects. Name: the Bipartisan American Affordability and Jobs Act of 2026. The vote For: 62 senators, incl...

Why AI Robotics Is Stalled: The Battle Between VLA and World Models

While generative AI and Large Language Models (LLMs) like ChatGPT have scaled at an astronomical pace, physical AI—robotics—seems to be hitting a frustrating plateau. Despite billions of dollars in venture capital funding, we have yet to see autonomous humanoids seamlessly integrated into our daily lives.

What is holding physical AI back? The answer lies in a fundamental technical divide. Robotics developers are currently split into two main camps, each trying to build a viable brain for physical machines, but both facing severe bottlenecks.

1. VLA (Vision-Language-Action) Models: Inheriting LLM Intelligence

The first major approach is the VLA model. This strategy essentially takes an existing, highly capable language model—the same kind that powers advanced chatbots or coding agents—and customizes it. By training it further on physical and robotic data, developers attempt to bridge the gap between digital reasoning and physical execution.

  • The Vision: Start with a baseline that is already inherently "intelligent." Since the model already knows how to reason, plan, and make sense of abstract concepts, the goal is to translate that digital intelligence directly into physical actions.

  • Key Players: Tech giants like Nvidia and high-profile startups like Physical Intelligence are heavily backing the VLA approach.

2. World Models: Teaching AI the Laws of Physics

The second approach gaining massive traction in Silicon Valley is the World Model. Unlike a VLA, a world model focuses purely on achieving a deep, intuitive understanding of physics.

Instead of reading text, world models are trained on massive amounts of video data collected from the internet—videos showing people interacting with objects and environments changing over time. By learning to predict the next frame of a video, the AI ideally figures out how the physical world works.

  • How Robots Use It: A world model can either serve as the robot's direct brain (predicting what physical actions to take next) or be used to generate hyper-realistic, AI-driven simulated environments where robots can train safely before entering the real world.

  • Key Players: Video AI pioneers like Runway and Luma, humanoid developers like 1X, and Fei-Fei Li's startup World Labs are prominent figures in this camp.

3. The Core Bottlenecks: Why Robotics is Stalled

The fierce debate between VLA and World Models exists primarily because neither approach works reliably well yet. Physical AI faces distinct, punishing challenges that digital AI never had to deal with:

  • The Data Scarcity Crisis: LLMs became powerful because they had the entire internet's worth of text waiting to be scraped. Robotics doesn't have an "internet of physical data." Roboticists must painstakingly collect physical interaction data manually, resulting in a severe data shortage.

  • High Stakes of Failure: If a chatbot hallucinates or makes a mistake, the consequences are usually minor. But if a robot hallucinates a physical reality, it can trip, break its incredibly expensive parts, or worse, injure a human nearby.

  • The Missing "Force" in Video Data: World models rely heavily on video, but video has severe limitations. When a robot watches a video of someone picking up a glass, the video cannot convey "force data"—how hard the person had to squeeze to lift it. Video also suffers from obstructions and blind spots, hiding crucial physical data.

4. Shifting Hype and Alternative Paths

Because VLA models have yet to achieve their definitive "ChatGPT moment," market enthusiasm and venture capital have recently been shifting toward World Models. However, World Models have yet to produce concrete, game-changing results.

Frustrated by this divide, some roboticists are looking outside both camps. A growing alternative involves using LLM coding agents (like Anthropic's Claude) to write robotic control software directly rather than building a single, all-encompassing physical AI brain. Surprisingly, many developers are finding immediate, practical success with this method.

Conclusion: Waiting for the Breakthrough

Whether the future of robotics belongs to language-infused VLA models, physics-driven World Models, or an entirely new hybrid approach remains to be seen. What is certain is that the race to build the ultimate brain for physical AI is highly funded, fiercely contested, and serves as the next true frontier for global technology.

Comments

Popular posts from this blog

Why European Oil Companies Are Outperforming America’s: The Secret World of Energy Trading

Inside Anthropic’s Slack Takeover: What 'Claude Tag' Means for Enterprise AI

Buying the Dip: Jim Cramer’s Top Stock Picks Amid Market Rotation