Prometheus Raises $12 Billion — Building a "General Engineer" for the Physical World
Prometheus, an AI startup backed by Jeff Bezos, has announced a $12 billion (about ¥1.8 trillion) funding round. The round reportedly values the company at $41 billion. The company has set its sights on building an "Artificial General Engineer," aiming to automate engineering in the physical world, from heavy engineering to drug discovery.
The fact that this much capital is flowing into AI for the physical world suggests that investment in robotics and manufacturing AI is heating up in earnest. Money is spreading beyond traditional software-centric AI bets into domains with real-world impact.
Theker Raises $85 Million — Factory Robots That Don't Specialize
Also in robotics, Theker has raised $85 million. Unlike humanoid robots with fixed forms, such as those from Boston Dynamics, Theker's robots feature a reconfigurable design. Their design philosophy avoids specializing in any single task, letting the robots flexibly handle a wide variety of jobs on the factory floor.
It has also been reported that EngineAI, a Chinese humanoid robot maker, has filed for an IPO in Hong Kong — fundraising and listing activity across the robotics sector is clearly picking up.
Data Center Power Consumption Up 26% — Japan's Challenge Is "Transmission"
Gartner predicts that global data center power consumption will reach 565 TWh in 2026, a 26% increase year over year. Power demand is climbing rapidly as AI services scale.
On the situation in Japan, the firm noted that slow development of transmission infrastructure — rather than a shortage of generation capacity itself — is becoming a bottleneck for data center construction. A physical transmission infrastructure gap sits between demand centers and power supply regions, and closing it is seen as an urgent priority.
SpatialClaw — An Agent That Tackles Spatial Reasoning with Code
SpatialClaw, featured on Hugging Face's Daily Papers, proposes a new approach to spatial reasoning tasks. An agent backed by a VLM (vision-language model) performs spatial reasoning by writing Python code in a persistent kernel.
This method makes it possible to compose perception modules, inspect intermediate results, and revise strategy between steps. Without any additional training, it reportedly outperforms existing agent methods by an average of +11.2 points across 20 benchmarks, with consistent improvements confirmed across six VLM backbones.
The claim that "code is an appropriate action interface for spatial reasoning" points toward a research direction of applying LLM reasoning capabilities to physical and spatial problems.
MoVerse — Generating Real-Time 3D Environments from a Single Image
Also featured on Hugging Face, MoVerse is a video world model that generates interactively navigable 3D environments in real time from a single narrow-field-of-view image.
It uses a method that lifts topology-aware 360-degree panoramas onto 3D Gaussian scaffolds, achieving high-quality scene traversal at 8 FPS on a single NVIDIA RTX 4090. Applications are anticipated in games, simulation, and training environments for robotics.
Is AI Just Making Developers "Busier"?
An article published on LeadDev, "AI isn't making developers more productive – it's making them busier," is drawing attention on Hacker News. The argument: while AI coding tools have certainly increased developers' activity, that may not be improved productivity — developers might simply be "busier."
Drawing on voices from the field of developers buried in reviewing and fixing AI-generated code, the article highlights the importance of distinguishing "effort" from "outcomes" when measuring the impact of tool adoption. It's a topic that will keep fueling debate as the industry assesses the real-world effectiveness of AI development tools.