Recently, Forbes described physical AI as the next major industrial frontier of AI,  with a conversation about machines that can sense their surroundings, make decisions and act in the real world. Though many founders and CEOs talk passionately about physical AI, one of the more outspoken recently has been Nvidia CEO Jensen Huang, who touts physical AI and robotics as a $50 trillion market opportunity.

Huang is so hot on physical AI that he even ditched his signature leather jacket in Seoul while forging a multi-trillion-dollar deal with LG Group to build an AI factory focused on humanoid and physical AI. His vision is one in which the “50 million worker shortfall” is answered by an “entire manufacturing line operated by robots, managed by more robots, and where the entire factory is a robot.” 

For now, AI remains mostly “trapped” in data centers, but increasingly, it is moving to devices, with inference pushing AI to the edge – increasingly bypassing power and grid bottlenecks, and bifurcating compute into:

• On-Device AI (PC/Mobile): with PC chips powering AI natively on laptops and desktops, reducing CapEx and strain on industrial grids;
• Distributed ‘Mini’ Data Centers: As we wrote last week, Nvidia-backed companies like Span are looking to install extremely compact, liquid-cooled compute nodes onto houses and businesses to avoid traditional grid constraints.

As the discussion goes beyond GPUs for data centers and toward on-device AI, systems will increasingly perceive, reason, and act autonomously. That will usher in the era of robotics, drones, and autonomous vehicles, also moving bottlenecks from raw model training in cloud data centers to real-time execution at the edge.

Data center interconnect: Speaking during a recent RCR webinar, AFL’s Noah Taylor said that as hyperscalers expand AI deployments, the networks linking data centers are becoming increasingly important alongside the facilities themselves.

Alphabet further diversifies with Intel: Alphabet’s Google has placed an order with Intel to manufacture 6 million TPUs across 2027 and 2028. The order hinges on Intel hitting aggressive yield and timeline milestones on its 18A node.

View More News

Bus ducts are in demand: Bus ducts are increasingly a critical component of AI-era power infrastructure, with Fortune Business Insights projecting market size to grow from USD 14.53 billion in 2026 to 21.94 billion by 2034, a CAGR of 5.29%.

Data4 700MW DC in France : Data4 is planning a €5bn, 700MW AI data center in N Northern France, where a former steelworks site in the Hauts-de-France region will be home to a proposed five-building campus.

Kasi Cloud Nigerian campus: Kasi Cloud’s Lekki campus in Lagos, Nigeria – opened as West Africa’s first hyperscale-ready, AI-capable platform – recently commissioned the first phase of its $250 million, 100MW campus.

AIOnX $500M pivot to neocloud: Data center developer AiOnX acquired a majority stake in cryptomining firm Genesis Digital Assets for $500 million, shifting GDA’s 1.3GW power crypto footprint in the U.S. and Sweden neocloud infrastructure.

Explosive Revenue Growth: According to Q1/Q2 financial roundups, neocloud providers are rapidly growing, with Nebius reporting a 684% YoY surge and CoreWeave a 112% growth surge amid big deals with Meta and Microsoft.

European Expansion: French AI infrastructure company Policloud (part of Antimatter Group) signed a €580 million contract with CloudGrid Energy for 29,000 renewable-powered GPUs across France, Germany, Italy, Spain, and Sweden.

Data center pushback: More than 75 data center build-outs worth $130 billion have been successfully blocked or delayed by opposition groups in the first 3 months of 2026, with Data Center Watch claiming it’s the same volume blocked in all of 2025.

.