Analytical briefings for engineers, PMs, and CTOs
Emerging hardware, semiconductors, chips, and AI infrastructure.
4 min readRISC-V has crossed 13 billion cores shipped and automotive giants have committed to it — but it is not displacing Arm. An analysis of where the open ISA actually wins in 2026.
5 min readCo-packaged optics moved from lab research to volume production in 2025. The bandwidth wall it solves is now the primary constraint on AI cluster scaling — not GPU availability.
6 min readSK Hynix controls 62% of global HBM shipments and sold out its entire 2026 production by mid-2025. Here is why High Bandwidth Memory — not TSMC node capacity or CoWoS packaging — has become the single binding constraint on the AI infrastructure build-out, and what Samsung’s 18-month qualification failure reveals about the difficulty of meeting NVIDIA’s thermal and power standards.
3 min readEvery major hyperscaler is running a significant fraction of AI inference on custom silicon — Google TPU v5, Meta MTIA, Microsoft Maia, Amazon Trainium. NVIDIA retains dominance in training, but the inference market is structurally bifurcating.
3 min readIntel’s Hala Point and IBM’s NorthPole deliver 22–100x energy efficiency gains over GPUs on specific sparse workloads — but no standard programming model, a thin software ecosystem, and narrow task compatibility have kept neuromorphic chips out of commercial AI deployments.