Category intelligence
24 current items analyzed and ranked.
Executive synthesis
Meta signed a deal worth up to $100B with AMD for AI chips, potentially acquiring a 10% equity stake — one of the largest AI infrastructure deals ever and a major challenge to Nvidia's dominance. This comes as US tech companies are projected to spend $660B on AI assets this year, even as datacenter construction faces growing delays from energy shortages, tariffs, and community opposition.
Primary evidence
Meta has signed a deal reportedly worth up to $100B with AMD for AI chips, just a week after committing to millions of Nvidia chips. The deal could result in Meta owning 10% of AMD and signals a major diversification of AI chip supply away from Nvidia dominance.
By Michael Acton, Financial Times
Meta's multi-billion dollar AMD chip deal could lead to a 10% equity stake in the chipmaker. AMD CEO Lisa Su noted each gigawatt of compute is worth 'double-digit billions,' with total capacity of 6 gigawatts.
By Aisha Down
The Guardian reports the Meta-AMD deal at $60B, noting it comes amid growing AI bubble fears. US tech companies are expected to spend $660B on AI assets this year, and the deal may signal a broader pivot in Meta's AI strategy.
By Last Week in AI
Building on yesterday's Reddit buzz, Anthropic released Claude Sonnet 4.6 just 12 days after Opus 4.6, making it the default for Free and Pro tiers. The model debuts a 1 million-token context window (4x previous Sonnet), with significant gains in coding and instruction-following. Gemini 3.1 Pro also released.
Continuing our coverage from yesterday, Anthropic revealed three 'industrial-scale' distillation campaigns by overseas labs that generated over 16 million exchanges using ~24,000 deceptive accounts to extract Claude's capabilities. The campaigns aimed to train competing models on Claude's high-quality outputs.
By Unknown
Latent Space provides detailed analysis of Anthropic's distillation accusations against DeepSeek, Moonshot, and MiniMax. MiniMax conducted the largest-scale extraction, while DeepSeek's was comparatively smaller. Anthropic caught MiniMax during an active distillation run.
By Asif Razzaq
Alibaba's Qwen team released the Qwen 3.5 Medium model series, including models where a 35B parameter model surpasses a 235B model through architectural efficiency and reinforcement learning. The series demonstrates frontier-level intelligence at significantly lower compute costs.
By Dashveenjit Kaur
First spotted on Reddit, now making mainstream headlines, Anthropic's Claude Code tool for COBOL modernization caused IBM's worst single-day stock drop in 25+ years (13% decline). The tool threatens IBM's lucrative legacy consulting business by automating what once required 'armies of consultants spending years.'
By Asif Razzaq
Google DeepMind used AlphaEvolve, an LLM-powered evolutionary coding agent, to automatically discover novel multi-agent reinforcement learning algorithms. The system treats source code as a genome, inventing new symbolic logic rather than just tuning hyperparameters.
By Esther Shittu
AI Business analyzes the Anthropic distillation controversy, noting that while distillation is common in AI, the industrial scale at which Chinese vendors allegedly operated is unprecedented and poses enterprise risks.
By Tom Perkins
US datacenter construction faces increasing cancellations and delays due to supply chain problems, energy shortages, tariff constraints, and growing grassroots community opposition. Some investors are also pulling back amid AI bubble fears.
OpenAI is partnering with major consulting firms as part of a broader enterprise AI push to gain traction with business customers.
