If you took time off over the holidays, good for you. AI did not. The first week of 2026 has delivered one of the most consequential news cycles in recent memory: NVIDIA unveiled Vera Rubin at CES, redefining what an AI platform even means. Meta closed a $2 billion acquisition of Manus, signaling that the race is now about agentic execution at scale, not just model performance. AMD positioned itself as the credible second source for enterprise inference. And the Model Context Protocol moved into the Linux Foundation, beginning the standardization of how agents interact with everything.
Meanwhile, two infrastructure stories emerged that will shape everything else: power constraints are becoming the primary bottleneck for AI scale, with data centers facing conditional access to the grid. And OpenAI formally conceded what security researchers have known for years: prompt injection is a permanent feature of the agentic landscape, not a bug to be patched.
Stepping back, three themes are defining 2026’s starting position: the hardware competition is shifting from chips to complete systems; agent security has entered a permanent arms race; and the winners will be whoever makes AI infrastructure boring, reliable, and governable.
Top Stories This Week
NVIDIA announced Vera Rubin at CES: a six-chip, extreme-codesigned AI platform promising 10x token cost reduction and positioning NVIDIA as a platform company, not a GPU company
Meta acquired Manus for $2B+, buying agentic harness technology that can finish work at scale across their massive platform surface area
AMD unveiled MI455 and MI440X at CES with OpenAI’s Greg Brockman on stage, positioning themselves as the chip for regulated industries and hybrid deployments
Microsoft partnered with MISO to modernize the Midwest power grid, marking hyperscalers becoming grid stakeholders rather than just customers
The “bring your own power” fight escalated as PJM and regional operators proposed requiring data centers to disconnect during peak demand
Anthropic donated MCP to the Linux Foundation’s new Agentic AI Foundation, with OpenAI, Google, Microsoft, and AWS joining as founding members
Google launched fully managed MCP servers for Maps, BigQuery, GCE, and GKE, turning tool use into managed infrastructure
OpenAI explicitly stated prompt injection is “unlikely to ever be fully solved,” conceding a core security truth about agents
Cursor acquired Graphite for well over its $290M valuation, betting that the winning AI dev platforms will own the entire SDLC loop
Lots to cover, so let’s get into it.
Listen to this episode with a 7-day free trial
Subscribe to Nate’s Substack to listen to this post and get 7 days of free access to the full post archives.
