AMD - Page 15

Microsoft unveils Maia 300 and says Maia will scale to multi-gigawatt deployments from 2027 as its main inference system. Maia 200 is in production.

Broadcom's Hock Tan and Intel's Lip-Bu Tan told a DBS forum in Singapore that capital, not chips, may limit AI as demand still outruns supply.

Contractor Willy Sun pleaded guilty to four counts in the Nvidia-server China case, Reuters and Bloomberg report. Co-defendants are presumed innocent.

A GPU price tracker says RunPod raised 12 of 29 Secure Cloud rates on 8 Oct; RunPod's pricing page now shows H100 SXM $3.99 and B200 $7.99 an hour. No RunPod announcement.

PitchBook reports Volta has launched a $5.012bn loan: a $1.3bn term loan C and a $3.712bn GPU term loan, price talk S+625-650. Launch-stage terms, not a closed deal.

Leakers say Nvidia has stopped allocating GB202 chips to the GeForce RTX 5090. Nvidia has not confirmed it. What is claimed, what is known and what is not.

Custom high-bandwidth memory puts a customer-designed base die under the memory stack, so the AI chip spends less of its own silicon talking to memory. What NVIDIA claims for its NVHBM, what SemiAnalysis estimates, and what is still unproven.

GlobalFoundries has signed a multi-year $2 billion agreement to make silicon interposers for TSMC's CoWoS advanced packaging at its New York plant, with volume production expected to ramp in the first half of 2028. TSMC also reported September revenue of about NT$511.86 billion. What an interposer is and what the numbers say.

A new benchmark, TasteVal, says the best AI model now reaches expert-level results on eight AI research tasks with 2.3 times less compute, though its authors list reasons that may overstate progress. Preference Model has open-sourced Karotte, a framework for training environments that are harder to game, and DatologyAI has released Zephon, a data loader that keeps experiments repeatable when the GPU count changes. What each claims, and who measured it.

Elon Musk says Tesla and SpaceX will build and run the Terafab AI chip complex in Texas, with TSMC at most subleasing part of it. Intel's CEO says Intel remains involved. What was said, what Terafab is, and the open questions.