Astera Labs: Scorpio Fabric Switches to Become Largest Product Line a Quarter Early as Content Per XPU Heads Toward "Multiple Thousands of Dollars"
Q2 2026 earnings call, August 4, 2026
Astera Labs delivered a quarter that validates the bull case on AI connectivity spend accelerating faster than even optimistic estimates had assumed. Revenue hit a record $392.4 million, up 27% sequentially and 104% year-over-year, and management guided Q3 to $540 million to $560 million, implying roughly 40% sequential growth at the midpoint. The standout disclosure wasn't the beat itself but the acceleration timeline: CEO Jitendra Mohan confirmed that the Scorpio fabric switch family will become the company's largest product line by revenue in the third quarter, a full quarter ahead of prior guidance. "We expect Scorpio to become our largest product family in Q3, which is 1 quarter ahead of our prior expectations, marking an important strategic milestone," Mohan told investors.
Scorpio X pulls forward and content per XPU keeps climbing
The Scorpio X-Series, the high-radix scale-up switch family, has entered volume production and President and COO Sanjay Gajendra said the company now has more than 10 customers engaged on Scorpio X, with a lead hyperscaler customer expected to ramp to high-volume production in Q3 and actually surpass Scorpio P-Series revenue this quarter. Multiple platforms are designed in even within that single lead customer, a nuance Gajendra flagged directly: "Even with the lead hyperscaler we have, there are multiple platforms where we are designed in. It's not like there is just one." Additional customers are expected to ramp to production toward year-end and into early 2027.
The more important number for modeling purposes is dollar content per XPU, which management said is now moving well beyond $1,000 for the Scorpio X-Series alone, addressing what the company sizes as a $20 billion merchant scale-up switching TAM. That figure has grown from roughly $50 to $100 per accelerator at IPO. Gajendra was explicit that this is an early stage of the ramp, not a ceiling: "This is already starting to happen... but at the same time, this is just the beginning of it because when you start adding several other features that are important in terms of optical, for example, or adding more signal conditioning because of the higher speed and the higher attach rate, we do believe that this number will continue to scale up." Management's stated target is multiple thousands of dollars of content per XPU as optical, UALink, and higher-radix switching layer on top of existing copper connectivity.
Optical roadmap gets a concrete timeline
Astera Labs laid out a three-phase optical strategy that gives investors the clearest timeline yet on when the aiXscale Photonics acquisition starts contributing to revenue. Phase one is high-density fiber attach solutions enabling up to 50-meter fiber connectivity, with an initial Tier 1 AI platform customer expected to enter volume production in 2027. Phase two is near-packaged optics (NPO) chipsets for multi-rack scale-up clusters, also targeted for 2027. Phase three, co-packaged optics (CPO) integrated directly into Scorpio X-Series switches, is a 2028-and-beyond opportunity. Mohan estimated the optical TAM unlock at "several tens of billions of dollars" beyond the company's existing copper footprint, and importantly said dollar content per XPU or per link increases as customers move from copper to NPO, meaning the transition is margin- and content-accretive rather than cannibalistic. He was clear that copper and optical will coexist rather than one displacing the other within the rack: "We do fully expect copper and optical to coexist... we do expect that for this generation and the next generation the scale-up within a rack will continue to be copper."
CXL finally shows commercial traction
After several quarters of CXL being a story about potential rather than revenue, Astera Labs closed a new design win for its standard Leo memory controller with a U.S. hyperscaler in the quarter and now expects to ship both standard and custom Leo CXL controllers in volume to two U.S. hyperscalers in 2027. Gajendra attributed the renewed momentum to three converging use cases: memory-constrained general-purpose servers seeking alternatives to costly HBM and DRAM, GPU/accelerator-attached memory paired with the company's KV Cache acceleration function to improve inference latency, and traditional high-memory-intensity workloads like SAP. His framing was candid about the years of patience the bet required: "I think it was about time, isn't it. I think it's finally great to see some traction back on CXL."
Gross margin trajectory and the software moat
Non-GAAP gross margin came in at 73.7% in Q2, above the 73% guide, but CFO Desmond Lynch guided Q3 down to approximately 72%, attributing the mix shift to Scorpio's wider margin range depending on lane count and silicon-versus-module use case. Lynch reiterated a long-term gross margin target of 78%, implying the company expects margin to trend up over time as switching volume scales despite near-term mix noise. Operating margin, by contrast, is expanding sharply on operating leverage, guided to approximately 43% in Q3, up 400 basis points sequentially, after landing at 39.1% in Q2.
Beyond the hardware ramp, management spent meaningful time describing how its COSMOS software platform is evolving from fleet management and telemetry into an orchestration layer that XPUs talk to directly, a shift Mohan said materially increases switching costs for customers. "Now we are becoming part of the orchestration layer. And the XPUs are talking directly to this -- our COSMOS platform to run the workload, and this drives a lot more stickiness into our solution," he said, adding that customers adopting the current generation are the most likely to carry that investment into the next one. The hardware-enabled, software-delivered features -- Hypercast and In-Network Compute -- are said to deliver up to a 2x improvement in collective performance for large-scale training and inference workloads.
UALink, China, and where the risks sit
On UALink, Mohan pointed to the consortium's 2.0 specification as validation of Astera's architectural bets on in-network compute, advanced RAS, and 200-gig SerDes, and said the company plans to intercept 2027-vintage XPUs with UALink-enabled Scorpio X switches. China was flagged as a real but bounded opportunity: Mohan said absolute China revenue will keep growing, driven by open protocols like PCIe, Ethernet, and increasingly UALink, but expects the rest of the world to grow faster, keeping China revenue in the single digits as a percentage of total sales. That is a modest but notable disclosure for investors weighing geopolitical exposure in the semiconductor supply chain. The company did not flag any customer concentration deterioration, and Lynch noted the growth has been genuinely broad-based across Aries, Scorpio, Taurus, and Leo rather than reliant on any single product ramp, which reduces single-point-of-failure risk in the model even as Scorpio's share of revenue mix rises.