SuperML
← All issues

The Silicon Decoupling: Meta's 1GW MTIA, OpenAI's $20B Cerebras Deal, and AI's Quiet Escape From Nvidia

Meta's 1-gigawatt Broadcom MTIA deal, OpenAI's $20B Cerebras contract, and Perplexity's Personal Computer on Mac — three stories, one pattern: AI compute is decoupling from Nvidia and from the cloud.

Hi there,

This week the AI-infrastructure story finally stopped being about Nvidia. Meta extended its Broadcom partnership through 2029 with a one-gigawatt MTIA commitment, OpenAI signed a $20B+ three-year deal with Cerebras, and Perplexity shipped an agent that runs on your Mac instead of in the cloud. Different companies, same shift — the monolithic "one GPU, one cloud" stack is unbundling into workload-specific silicon and on-device compute.

🔥 Featured Post

The Silicon Decoupling: Meta's 1GW MTIA, OpenAI's $20B Cerebras Deal, and AI's Quiet Escape From Nvidia

  • Meta and Broadcom extended their MTIA partnership through 2029 with an initial 1-gigawatt deployment — Meta's chip is the first AI ASIC on a 2nm process node, optimized for inference rather than general-purpose workloads.
  • OpenAI committed to pay Cerebras over $20B across three years for wafer-scale compute, directly diversifying away from sole Nvidia dependency.
  • Perplexity's Personal Computer for Mac flipped the script — instead of shipping your data to the cloud, the agent orchestrates local files, native apps, and the browser on-device, keeping sensitive context off the network.
  • PwC's 2026 AI Performance Study found 20% of companies are capturing 74% of AI's economic value — and the leaders are redesigning workflows, not just buying bigger GPUs.
  • The real pattern is workload segmentation: custom ASICs absorb high-volume inference, Nvidia GPUs retain training, and edge devices eat the long tail — which rewrites AI unit economics from the hardware up.

Read the full post →

📚 In Case You Missed It

Human-Led, AI-Accelerated: Why the Winning Stack in 2026 Isn't Fully Autonomous — Gartner expects 40% of agentic-AI projects cancelled by 2027 and production agent reliability still sits near 25% failure — but the 'human-led, AI-accelerated' stack is quietly winning across coding, research, ops, and content. Here's the pattern, the evidence, and how to design for it.

AI as a Research Partner: AlphaEvolve Cracks Math, Machine-Learned Physics Goes 10,000× Faster, and Frontier Models Get Cheap — AlphaEvolve breaks Strassen's 56-year matrix-multiplication ceiling, machine-learned force fields promise 10,000× faster chemistry, and Gemini 3.1 Flash-Lite launches at $0.25/M tokens.

The Agent Stack Grows Up: Opus 4.7, MCP Becomes a Standard, and a $50B Infrastructure Bet — Claude Opus 4.7, MCP hitting 97M installs under Linux Foundation governance, and Oracle's $50B AI-infra bet — how the agent stack is industrialising.

More posts dropping every day. Stay curious.

— Bhanu @ superml.dev