Wednesday, May 6, 2026

Start of archive·May 16

Top story

Platform WatchThe Decoder

Anthropic taps SpaceX's Colossus-1 data center for 220,000 GPUs to power Claude

Anthropic is making a massive infrastructure bet to scale Claude's training and inference capacity, signaling aggressive competition in the compute arms race and directly challenging OpenAI's infrastructure dominance.

220,000 NVIDIA GPUs allocated to Anthropic

The briefs

OpenAI and five hardware giants (AMD, Broadcom, Intel, Microsoft, NVIDIA) built MRC, an open-source networking protocol that slashes GPU interconnect complexity and power consumption. It's already live on Stargate—signaling a fundamental shift in how AI infrastructure scales beyond 100K GPUs.

MRC protocol reduces switch layers from 3-4 to 2 for 100K+ GPU clusters

Musk is building AI compute sovereignty outside the traditional foundry model. A $119B Terafab factory signals how serious players are about owning their entire chip-to-inference stack—and what it costs to compete with TSMC.

Terafab facility in Texas

DeepSeek's ascent to $45B valuation signals China's strategic push in frontier AI and raises questions about geopolitical AI competition and state-backed consolidation in the Chinese AI ecosystem.

DeepSeek valuation: ~$45 billion

SpaceX/xAI is making a massive vertical play into semiconductor manufacturing, signaling either extreme compute confidence or extreme compute desperation—either way, it reshapes the AI infrastructure race and threatens traditional chip supply chains.

SpaceX filing indicates $55B initial investment with potential total of $119B for 'Terafab' chip factory

Anthropic is shipping a production capability that lets Claude agents retain memory across sessions and self-correct, moving agentic AI from stateless to stateful systems. This is a major product differentiation move in the managed agents race.

Claude Managed Agents now have 'dreaming' capability

Anthropic is outsourcing inference to SpaceX's cutting-edge supercomputer infrastructure, a strategic move to reduce capex while scaling Claude deployment. This signals a shift in how frontier labs are leveraging third-party compute capacity rather than building entirely in-house.

Anthropic partnering with SpaceX for inference on Colossus 1 supercomputer

A major strategic partnership between two AI-forward platforms collapsed, signaling either shifting priorities at Snap or stalled momentum at Perplexity. The deal's failure raises questions about the viability of tight AI integrations and whether search engines can command premium valuations for platform exclusivity.

Snap-Perplexity deal was worth $400M

Arm is moving beyond IP licensing into direct semiconductor manufacturing for AI, signaling a major shift in the chip supply chain and intensifying competition with NVIDIA and custom silicon makers.

$2bn projected sales from new AI chip (starting next year)

Nvidia is securing critical bottleneck infrastructure for AI data centers. By partnering with Corning to build three new U.S. factories for optical networking components, Nvidia is directly addressing supply chain constraints that limit AI cluster deployment and competitive positioning.

Nvidia partners with Corning on optical network component production