A detailed look at Jalapeño, OpenAI's ASIC developed with Broadcom in 16 months, which beat Nvidia, AMD, and Google chips on multiple top open-source models
THE SO WHAT
An in-house ASIC that beats leading GPUs on TCO and throughput for key open models shows hyperscalers are serious about verticalizing inference. If you’re building AI infra or large-scale apps, assume the economics will tilt toward custom silicon tied to specific clouds rather than general-purpose GPUs.
READ THE SOURCE
MORE FROM THE WIRE
Deep & Emerging TechNew Windows malware lays dormant until a custom command activates it like a sleeper agent
Dormant Windows malware that waits for a custom command before activating is another step toward low‑noise, long‑dwell intrusions. Security teams need to treat “clean” baselines with more skepticism—focus on unusual command‑and‑control patterns and tighten egress monitoring, not just signature scans.
Deep & Emerging TechMeaco fans sold out completely during the last heatwave — but the company just launched two new tower models, and they're both in stock right now
Fans selling out during heatwaves then returning with new tower models is a small but clear sign that climate volatility is now a demand driver for home hardware. Retailers and OEMs should treat cooling and air-quality gear like seasonal infrastructure, not niche appliances—stock and supply need to match weather, not the calendar.
Deep & Emerging TechNSF Extends Quantum Sensing Program With Five-Year $37.5M Award
A $37.5M, five-year NSF renewal for quantum sensing in biophysics is a bet that measurement tech will unlock the next wave of bio insight. If you’re building in diagnostics or bioinstrumentation, assume quantum-enhanced sensing will move from lab curiosity to competitive edge over this decade.
Deep & Emerging TechUniversity of Washington and Seattle biotech win $245M patent judgment against Guardant
A $245M judgment plus royalties through 2033 shows that core sequencing IP still has real teeth—and long tails. If your product depends on foundational bioinformatics or sequencing methods, budget for FTO analysis and potential licensing as a primary risk, not an afterthought.