Oracle to Offer Cerebras Systems Computers in AI Cloud Service
THE SO WHAT
Oracle adding Cerebras systems to its AI cloud is another data point that hyperscale-style providers want alternatives to a single GPU vendor—heterogeneous accelerators are becoming a strategic hedge. If you’re building AI workloads at scale, design for portability across different compute backends rather than hardwiring to one chip ecosystem.
READ THE SOURCE
MORE FROM THE WIRE
The secret succession rankings that Anthropic shareholders will never see (unless the company is ordered)
Quiet, pre-baked succession lists inside a special voting LLC show how frontier labs are trying to hard-code governance continuity before public-market pressure arrives. If you’re building critical AI infra, board design and control mechanics are now as strategic as model architecture.
Applied AIFrontier AI labs say their models refuted the Exponential Time Hypothesis. What does that mean?
If large models are genuinely helping probe — or refute — complexity-theory conjectures like ETH, they’re becoming tools for discovering new algorithms, not just applying known ones. For operators, the upside is step-change improvements in core primitives like search, optimization, and crypto — but also a moving target for any security or pricing model that assumes today’s hardness results.
Applied AI'This is a wow moment': T-Mobile has made the world's first AI-RAN call, and it could mean huge changes for the next generation of staying in touch
AI-managed RAN is about squeezing more capacity and efficiency out of fixed spectrum and capex — not just a marketing label on a phone call. Network-dependent products should assume more dynamic, variable-quality links and start testing for behavior under AI-optimized, rather than statically engineered, radio conditions.
Applied AIA Visit to the AI Model Retirement Home
“Model retirement homes” are a symptom of the long tail of legacy AI — old checkpoints still drawing power, maintenance, and sometimes revenue. Teams should inventory which models they’re actually using versus just hosting, and treat decommissioning as a cost and risk lever, not an afterthought.