Josh Smith, Senior Principal Engineer at SiFive, reveals how SiFive are able to deliver balanced performance CPU core IP into the area and power-conscious consumer and commercial market segments through unique microarchitectural innovation Click the video to watch now: na2.hubs.ly/H05NPMl0 #sifive #CPU #RISC-V #RVA23 #EdgeAI #AIIP #VectorProcessing #CustomAccelerators #TinyML #socdesign

$ALAB ALAB sharpened its position at the heart of what it refers to as “AI Infrastructure 2.0,” where hyperscalersare moving from box-level builds to rack-scale systems. Revenue was $230.6M, up 20% QoQ and 104% YoY, with record 76.4% non-GAAP gross margin and41.7% operating margin. Growth was broad-based across Aries PCIe retimers, Taurus smart cablemodules (AEC for Ethernet), and Scorpio switch fabrics. On the product front, PCIe Gen6 is now a real revenue driver: Gen6 products (Aries Scorpio P) exceeded20% of total Q3 revenue. Scorpio P (scale-out/PCIe-like) continued its initial volume ramp at a lead hyperscaler, with new winsacross multiple AI platforms spanning merchant GPUs (including NVIDIA GB300/B300) and customaccelerators. The company reiterated that Scorpio X (scale-up) begins shipping into production systems in Q4 with themeaningful, multi-customer ramps coming through 2026; management continues to frame Scorpio X asthe “anchor socket” in next-gen AI racks, pulling through additional Astera content (retimers, gearboxes,controllers, and smart cable modules) and materially lifting dollars per accelerator. Remember that last part of the last sentence – this will make or break ALAB’s 2026. Taurus rides 400G through 2026 while 800G ramps kick in from early 2026. Leo (CXL) is gaining traction ininference to shift memory off pricey HBM into big DDR5 pools, widening its footprint beyond generalcompute. But… … the bigger story is that Astera is wiring an open, software-defined rack-scale fabric just as Trainium andGoogle’s TPUs move from “inside baseball” to customer-facing accelerators. Recall just last week that Amazon has deployed 500,000 Trainium 2 chips for Anthropic and that numberwill reach over a 1,000,000 in the coming months. Layer in AMD’s MI450 (with OpenAI making a commitment) and the center of gravity tilts away fromsingle-vendor lock-in. NVIDIA’s demand stays effectively unlimited, but the architecture gets more openand multi-sourced, that’s the argument coming into focus. 2/2