MLWhiz Weekly AI/ML Newsletter # 4

2026-04-21 · Source: MLWhiz: Recs|ML|GenAI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Cloud Computing & IT Infrastructure · Depth: Advanced, medium

Summary

OpenAI has entered an unprecedented procurement deal with Cerebras, committing over $20 billion in chip spending through 2028 and securing 750 MW of capacity, expandable to 2 GW. In return, Cerebras provided OpenAI with a $1 billion loan at 6% interest and approximately 10% equity in Cerebras post-IPO. This "circular financing" model positions OpenAI as Cerebras's largest customer, creditor, and a significant shareholder, setting a new precedent for large-scale AI infrastructure deals. Additionally, the week saw NVIDIA release Nemotron 3 Super, an open-weight 120B hybrid Mamba-MoE model pre-trained in FP4, and Alibaba's Qwen 3.6-35B-A3B, optimized for agentic coding. Anthropic also launched Claude Design, its first standalone product beyond a chat interface, impacting Figma's stock.

Key takeaway

For CTOs and VPs of Engineering negotiating significant GPU contracts, the OpenAI-Cerebras deal establishes a new precedent: demand equity stakes, not just volume discounts. Additionally, diversify your inference architecture beyond NVIDIA, as the rapid closure of the US-China AI capability gap and new model releases from other vendors suggest a more competitive landscape. Be aware of rising RAM prices and potential supply shortages through 2030.

Key insights

AI buyers are becoming AI owners through equity-based procurement deals, reshaping supplier relationships.

Principles

• Customer equity in suppliers can reclaim margin.
• Agentic search can rival GraphRAG for efficiency.
• Quantization should be layer-specific for optimal results.

Method

Meta's SOLARIS uses speculative decoding for recommendation by precomputing user-item embeddings asynchronously. Meta's Cycle-Consistent Search trains agents without labels by verifying if retrieved documents reconstruct the original query.

In practice

• Negotiate for supplier equity in large chip contracts.
• Explore agentic search before building GraphRAG.
• Implement layer-specific FP8 quantization for ranking models.

Topics

• OpenAI-Cerebras Deal
• AI Infrastructure
• Circular Financing
• Agentic AI
• Retrieval-Augmented Generation

Best for: CTO, Investor, VP of Engineering/Data, AI Scientist, Director of AI/ML, Machine Learning Engineer

Related on AIssential

• not much happened today — The AI market is shifting towards inference economics, agentic platforms, and diverse hardware architectures beyond NVIDIA.
• OpenAI sidesteps Nvidia with unusually fast coding model on plate-sized chips — OpenAI is diversifying its hardware dependencies for AI inference, partnering with Cerebras and AMD to reduce Nvidia reliance.
• Nvidia has already committed $40B to equity AI deals this year — Nvidia is investing heavily in AI companies, including customers, to build a competitive moat.
• MLWhiz Weekly AI/ML Newsletter # 3 — The AI industry is rapidly fragmenting across infrastructure, models, and applications, moving beyond single-vendor dominance.
• The Nvidia Groq Acquisition Explained — Major tech firms use licensing and acquihire deals to absorb competitors and talent while bypassing antitrust regulations.
• MLWhiz Weekly Recsys/ML/GenAI Newsletter # 10 - The week AI infrastructure crossed from a technology story to a financial one — AI infrastructure financing is shifting from tech giants to private capital, impacting compute costs and accessibility.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MLWhiz: Recs|ML|GenAI.