🤖 Anthropic’s New Colossal Compute Deal

🤝 Anthropic’s New Colossal Compute Deal

🏠 Putting Mini AI Data Centers on Homes

📍 6 AI Stock Sectors

🛠️ 5 Trending Tools

🥪 4 Brief Bites

💰 3 Funding Frontlines

💼 4 Job Opportunities

For context, SpaceX acquired xAI on Feb. 2nd, 2026. This new parent-subsidiary arrangement, valued at roughly $1.25 trillion, is often referred to as SpaceXAI.

Because “Colossus 2,” the world’s first fully operational gigawatt-scale AI training cluster, is nearly functioning at full capacity, SpaceXAI doesn’t really need “Colossus 1” anymore. So, tech billionaire Elon Musk decided to rent it out.

Anthropic secured all the available compute capacity from “Colossus 1,” which is equipped with over 220,000 NVIDIA GPUs, including dense deployments of the H100, H200, and GB200 Grace Blackwell Superchip.

Anthropic gained immediate access to more than 300 MW of new computing power. On average, 1 MW of continuous power can support about 800 U.S. households, meaning Anthropic’s new compute capacity is equivalent to the power usage of around 240,000 U.S. households.

It’s already enabled Anthropic to raise API rate limits considerably, meaning you’re allowed to send more requests within a specific timeframe without receiving errors. In simple terms, Anthropic essentially added extra lanes to the highway, reducing traffic jams.

It’s also enabled Anthropic to remove the peak hours limit reduction on Claude Code for Pro and Max plans. In simple terms, Anthropic raised the speed limit on the highway.

The XFRA units are industrial-grade, liquid-cooled, and self-contained compute clusters packed with 16 NVIDIA RTX PRO 6000 Blackwell Server Editions, designed to deliver gigawatts of new compute capacity directly from suburban backyards. They’re already being installed within a handful of residential homes and commercial spaces.

Most residential homes have a 200-amp electrical service size. In the middle of the night, when only the fridge hums, homeowners are actively consuming no more than just 5% of their electricity. That extra 95% just sits there. SPAN can redirect it to XFRA units to help process NVIDIA’s AI inference.

For reference, 8,000 XFRA units can be installed 6x faster and at 20% of the cost of constructing a comparable 100 MW data center, bypassing the more intense local, state, and federal approval processes, including public zoning hearings, utility interconnections, and environmental filings.

It’s marketed as a “win-win-win” across the energy ecosystem. AI hyperscalers gain immediate access to “distributed data centers,” homeowners gain a cheap flat fee for Wi-Fi and electricity, and utility companies gain more efficient utilization of the U.S. power grid infrastructure.

We currently face a major bottleneck: the demand for computing power is growing faster than the world’s ability to supply the necessary electricity for it. In many regions, the U.S. power grid infrastructure isn’t designed to handle the sudden surge in energy consumption from AI. This distributed data center solution aims to mitigate the “grid strain.”

Astrada raised a $3.8M Seed Round to build the data layer for banks.

QuantWare closed a $178M Series B to pioneer the KiloQubit era.

Blitzy landed a $200M Growth Round to write code for weeks.

NVIDIA (Santa Clara, CA): Formal Verification Intern, Fall 2026

Meta (Paris, FR): Research Scientist, AI for Reasoning, Entry-Level

Articul8 (Dublin, CA): Applied AI Researcher, GenAI Stack, Mid-Level

Anthropic (Seattle, WA): Research Ops., External Artifacts, Senior-Level