8 stories
Stratechery published an analysis of TSMC's latest earnings arguing that company leadership is not fully committed to the AI growth narrative; the full piece is behind a subscription paywall.
Intel CEO Lip-Bu Tan used the company's Q1 2026 earnings call to argue that inference and agentic workloads will shift compute ratios in favor of CPUs. Q1 revenue of $13.6 billion beat expectations, with AI-driven lines up 40 percent year-on-year.
Meta has signed a multi-year deal with AWS to deploy tens of millions of Graviton 5 CPU cores, making it one of the largest-ever consumers of Amazon's homegrown silicon, as the company diversifies its compute fleet away from x86.
The $599 base model M4 Mac mini is unavailable for delivery or in-store pickup on Apple's retail site, with eBay listings for the same configuration ranging from $700 for lightly used units to $979 for refurbished ones, according to TechCrunch.
NVIDIA NeMo RL applies FP8 precision across both generation and training phases of reinforcement learning, closing accuracy gaps via importance sampling and delivering approximately 48% total speedup when KV cache and attention are also quantized.
NVIDIA's developer blog details Blackwell platform support for DeepSeek-V4 on day zero, reporting initial GB200 NVL72 throughput numbers and covering deployment options via SGLang, vLLM, and hosted endpoints at build.nvidia.com.
NVIDIA has integrated the Muon higher-order optimizer into Megatron Core and NeMo, showing throughput within a small margin of AdamW on GB300 hardware — 1,080 vs 1,051 TFLOPs/s/GPU for Kimi K2 — while enabling training at thousands of GPUs.
The latest NVIDIA FLARE API splits federated learning adoption into two steps: a minimal client API that adds federation to existing training scripts without restructuring them, and portable job recipes that run unchanged from simulation to production.