NPI Engineer Intern

Northland Capital Markets downgraded Astera Labs to "Market Perform" from "Outperform" on October 9, citing valuation concerns, even though it expects earnings to rise. Analyst Gus Richard said shares had passed Northland's earlier target, but he still sees estimates climbing as the Scorpio X AI switch ramps up. Astera expects third-quarter revenue of $540 million to $560 million, up from a record $392.4 million in the second quarter. Non-GAAP earnings per share are guided to $1.16 to $1.21, against $0.80 in the previous quarter. The growth leans on one lead customer, though more than 10 customers are engaged on Scorpio X. Shares closed at $285.82 on October 16, about 43% below their June peak of $499.48. Fubon Securities recently began coverage with a "Buy" rating and a $500 price target.

Astera Labs' Orion rack supports Up to 160 accelerators. Astera Labs introduces Orion, a compute-agnostic rack-scale AI inference architecture supporting up to 160 accelerators per pod through a 320-lane PCIe fabric, modular reference designs, and COSMOS management software. Astera Labs (Nasdaq: ALAB) introduced Orion, a compute-agnostic rack-scale reference architecture designed to support AI inference deployments with up to 160 accelerators per pod. Announced at the 2026 Open Compute Project Global Summit in San Jose, Orion combines a high-radix, 320-lane PCIe switching fabric with modular compute, memory and connectivity subsystems. Operators can select GPUs, CPUs and other accelerators based on workload requirements, availability and cost rather than committing to a single compute platform. Orion comprises six production-grade reference designs covering the complete rack, compute trays, fabric trays, memory trays, link boards and compute boards. The designs aim to reduce integration and validation work for OEMs and ODMs while allowing different accelerator configurations within a common infrastructure. COSMOS Rack Manager and COSMOS Fabric Manager provide orchestration, telemetry, diagnostics and fabric management. Astera Labs will design and validate the reference architecture and subsystems rather than manufacture complete racks. OEM and ODM partners are expected to offer Orion-based systems beginning in Q1 2027. At OCP Global Summit October 12-15, Astera Labs is demonstrating a powered Orion rack with interchangeable, pre-validated compute trays. | ORION - KEY FACTS | | - Compute-agnostic architecture with interchangeable CPU/GPU options | | - Up to 160 accelerators per pod; 320-lane PCIe switch fabric | | - Six reference designs: rack, compute tray, fabric tray, memory tray, link board and compute board | | - COSMOS Rack Manager and Fabric Manager | | - UALink on roadmap; OEM/ODM availability targeted Q1 2027 | Analysis: Orion extends Astera Labs beyond individual connectivity components toward defining how those components are integrated into AI infrastructure. Its existing portfolio includes Scorpio Smart Fabric Switches, Aries signal conditioners, Leo Smart Memory Controllers, Taurus Ethernet connectivity and COSMOS management software. The announcement does not specify that every Orion configuration uses every product family. The PCIe fabric is intended to accommodate heterogeneous inference workloads. Actual throughput will depend on PCIe generation, topology, aggregate bandwidth, accelerator and memory configurations, software and workload characteristics. Astera Labs has not disclosed independently verified benchmarks establishing improvements in tokens per second or tokens per dollar against specific competing systems. Dedicated memory trays could help operators configure memory resources separately from compute, relevant to model weights and key-value caches. Benefits depend on memory access mechanisms, accelerator compatibility, latency and bandwidth. For neoclouds, modular infrastructure may reduce exposure to accelerator procurement constraints, although actual capital savings depend on configuration, utilization and integration costs. UALink is on Orion's future interconnect roadmap, not a confirmed feature of the initial release. Key commercial milestones include naming OEM and ODM partners, supported accelerators, customer deployments, pricing and measured inference performance. | ASTERA LABS | COMPANY PROFILE | | Headquarters | San Jose, California | | Ticker | NASDAQ: ALAB | | Products | Scorpio, Aries, Leo, Taurus, COSMOS, Orion | | Technology | PCIe, CXL, Ethernet, NVLink Fusion; UALink roadmap | | Markets | Hyperscalers, neoclouds, enterprise AI | | Website | asteralabs.com | | RELATED CONVERGE DIGEST COVERAGE | | For related reporting on Astera Labs PCIe, CXL, Scorpio, Aries and Leo technologies, see the Converge Digest archive: Astera Labs coverage. Individual article links from the draft have not been independently verified. | Converge Digest is tracking AI infrastructure developments. Follow ongoing coverage at Converge Digest AI Infrastructure. Help others discover this reporting. Editorial responsibility Jim Carroll. Editor & Publisher Every article published by Converge Digest is researched, curated, fact-checked and editorially reviewed by Jim Carroll, Editor & Publisher. AI-assisted drafting may be used to accelerate production, but all content is reviewed, refined and approved prior to publication. Builders & Visionaries The people who opened the doors to the networks Converge Digest run today. Which of these co-founded Astera Labs? On this day. October 12 From 25 years of the Converge Digest archive.

Astera Labs (Nasdaq: ALAB) has unveiled Orion, a modular rack-scale reference design that lets customers choose their own GPUs and CPUs. The designs target neoclouds and enterprises building AI inference capacity. Orion's PCIe switch fabric scales from 8 to 160 accelerators per pod. It includes six reference designs, covering the rack, compute tray, fabric tray, memory tray, link board, and compute board, all managed by Astera's COSMOS software. Astera Labs will not build the racks. Original design manufacturer and original equipment manufacturer partners will make and sell them, with availability expected from Q1 2027. The company will show a powered Orion rack at the Open Compute Project Global Summit in San Jose, October 12-15. It plans to extend Orion to other open interconnects, including UALink.