32,768 cores and 74 petaflops: AMD’s monster new rack for AI agents could be too much for traditional data centers
THE SO WHAT
Agentic workloads are getting their own hardware form factor—AMD’s Helion rack with 32,768 cores and 74 petaflops assumes dense, tightly coupled inference at scale. Data center operators and large AI users should be modeling power, cooling, and network upgrades now if they expect to run this class of hardware in production.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIThe collapse of web coding gigs is getting ugly — AI takes over routine website development jobs, and business owners are very happy with its job so far
Low-end web dev is getting unbundled from humans—AI site builders are now good enough that small businesses prefer self-serve over freelancers. If you sell generic brochureware or template work, you either move up-stack into custom systems and integration or you watch your lead flow evaporate.
Applied AIAfter Muse Launch, U.S. Banks Are Saying Hell No to AI Agents Moving Money
Banks drawing a hard line on AI agents moving funds means agentic finance will be gated by existing KYC, auth, and ops controls for a while. If you’re building in fintech, assume AI as copilot for humans—not autonomous transactor—until regulators and risk teams rewrite the playbook.
Applied AISources: Anthropic pitched its own AI safety bill to US lawmakers; the bill would limit the circumstances in which the government could block a model's release
Model providers are no longer just reacting to regulation—they’re drafting the templates for it, including limits on when the state can block releases. If you operate on top of these models, your future compliance posture will be shaped as much by vendor-written statutes as by traditional tech policy groups.
Anthropic polled 1,334 of its employees. They're more sure than most of us that AI will take over jobs.
When 1,334 people building frontier models say AI can do 80% of their own jobs by 2030, that’s an internal forecast of aggressive automation, not just external disruption. Operators should treat this as a planning input—map which roles in your org look most like theirs and start designing supervision, not replacement, paths.