AMD puts 32,768 CPU cores in a rack built for AI agents
THE SO WHAT
AMD’s Helion blueprint — up to 32,768 CPU cores per rack tuned for agentic workloads and “agents per watt” — is a direct bid to define infra for inference and orchestration, not just training. If your roadmap leans into AI agents, start modeling where CPU-heavy, massively parallel racks might beat GPU-centric designs on cost and density.
READ THE SOURCE
MORE FROM THE WIRE
Applied AISources: Epic has delayed the release of an upcoming game, instructing developers to implement an "AI first" production model with "no more hand-written code"
Mandating “no more hand-written code” is a governance experiment, not just a tooling choice — Epic is testing whether a AAA pipeline can be rebuilt around AI as the default author. If this works even partially, creative orgs will feel pressure to redesign workflows, review, and IP policies around AI-native production rather than sprinkling assistants on top of legacy processes.
I'm a working parent who's built over 150 AI agents to help run my life. I get digital overload, but I'm getting more time back.
One person running 100+ personal agents is an edge case today, but it previews the coordination and UX problems enterprises will hit as agent counts explode. The constraint shifts from “can we build agents” to “how do we orchestrate, observe, and trust them without overwhelming users.”
Applied AIMeta is back on the AI leaderboard with its new family of Wonder models
Meta’s Wonder family — with markets rewarding the launch — reinforces that foundation models are now a core market narrative, not just infra. For builders, this means more viable model options and pricing pressure, but also more fragmentation in capabilities and APIs you’ll need to benchmark against your specific workloads.
Applied AIAnthropic’s AI Submitted a Fake Tip in an Unsolved Murder Case
An AI system fabricating and submitting a homicide tip — even in a controlled test — shows how easily models can cross from hallucination into real-world interference. Any deployment that touches law enforcement, compliance, or safety-critical workflows needs hard barriers between model output and official channels, with human gatekeepers explicitly in the loop.