Earnings Call Highlights: Astera Labs (ALAB) | Q3 FY25
✨ Summary
Driven by demand for AI rack-scale connectivity, Astera Labs hit another growth milestone:
Revenue of $230.6M (QoQ +20%, YoY +104%), non-GAAP gross margin of 76.4% and operating margin of 41.7%, a record high.
Scorpio P-Series continues to ramp; Scorpio X-Series starts initial production in Q4 and enters high volume in 2026;
The Aries PCIe 6 retimer is now the industry's only high-volume shipping solution, at >20% of quarterly revenue;
Taurus AEC 400G is strong, with 800G expanding in 2026;
Announced the acquisition of Xscale Photonics to build out Scale-Up Photonics (an optical Scorpio). Q4 guidance: revenue of $245–253M, gross margin of about 75% and EPS of $0.51.
💰 Financials (Non-GAAP)
Revenue: $230.6M (QoQ +20%; YoY +104%)
Gross margin: 76.4% (QoQ +40 bps)
Operating expenses (Opex): $80.0M (R&D $57.2M / S&M $10.0M / G&A $12.8M)
Operating income: $96.2M (operating margin 41.7%, QoQ +250 bps)
EPS: $0.49 (tax rate 18%, 180.6M fully diluted shares)
Operating cash flow: $78.2M; cash and investments: $1.13B
Outlook (Q4 FY25)
Revenue: $245–253M (QoQ +6–10%)
Gross margin: ~75% (higher Taurus hardware mix)
Opex: $85–90M (including Xscale acquisition costs)
Interest income: ~$11M; tax rate: ~15%
Fully diluted shares: ~183M; EPS: ~$0.51
🧩 Product Updates
1) Scorpio (Switch Fabric / Scale-Up)
P-Series: Continues to ramp at the lead customer, with multiple additional hyperscaler design wins in 2H 2025.
X-Series: Initial production starts in Q4, high volume in 2026; designed for a smooth evolution from PCIe-like → UALink, with hardware/software (COSMOS) features as an extensible foundation.
Higher content value: Scorpio X is the "anchor socket" of next-gen AI racks, significantly raising connectivity content per accelerator ($ per accelerator).
Annual mix: Management estimates Scorpio at about 10% of FY25 revenue, with a Q4 exit rate of ~20%; X overtakes P progressively in 2026.
2) Aries (PCIe Retimer / Smart Retimer)
PCIe 6 is now the industry's only high-volume shipping solution, at >20% of quarterly revenue.
With adoption by custom accelerators and new-generation merchant GPUs, growth continues in 2026.
3) Taurus (Ethernet AEC / Smart Cable Modules)
400G is ramping in AI scale-out deployments and is the main growth driver in Q4.
800G demand is starting; in 2026 it moves into volume production as customer qualifications complete. AEC has clear advantages in reach and signal integrity.
4) Leo (CXL Memory Expansion)
Focused on memory offload for AI inference (moving data from HBM to large DDR5 pools), extending into AI use cases beyond general-purpose compute.
5) Scale-Up Photonics (Optics)
Acquired Xscale Photonics (fiber-to-chip coupling / advanced packaging capabilities),
planning an optical Scorpio (CPO/LPO/NPO direction) to extend cross-rack reach and higher data rates.
Timeline: Revenue contribution expected as early as 2028–2029 (mainly rack-to-rack / longer-reach scale-up).
6) Standards and Ecosystem
Remains committed to an open standards strategy (PCIe / UALink / Ethernet)
UALink: Solutions available in H2'26, revenue contribution from 2027; working with >10 AI platforms (both merchant and custom accelerators).
🚀 Outlook
Driven by AI Infrastructure 2.0: Hyperscale AI racks are driving explosive demand for all-to-all ultra-low-latency scale-up fabrics.
Portfolio upgrade: Scorpio X becomes the anchor, pulling through attach of Aries/Taurus and other products; UALink is a new growth driver from 2027.
Protocol-neutral & software-first: In load-store (memory semantic) architectures, COSMOS software and management features improve efficiency and observability in large-scale deployments.
Long-term optics: As power and density bottlenecks bite, scale-up optics is a new growth curve for 2028+.
Organizational growth: >700 employees by end of 2025 (+60% since the start of the year), supporting multiple product lines in parallel.
🗣️ Q&A Highlights
PCIe vs. UALink: For now PCIe-like dominates; UALink offers higher line rates, lower latency and AI-traffic optimization, with meaningful contribution from 2027; the two are parallel and complementary.
Scorpio footprint: Being adopted on >10 AI platforms; P-Series keeps expanding in 2025, X-Series reaches high volume in 2026, with multi-generation platforms extending to 2029.
Aries growth: PCIe 6 penetration will expand with each generation of custom/merchant accelerators.
Taurus scope: 400G demand is strong across multiple customers; 800G begins ramping in 2026 (AEC solves reach/signal integrity).
China opportunity: With limited access to high-end IP, demand for PCIe-based scale-up is rising; more GPUs → higher connectivity value.
NVIDIA Rubin (cableless rack): Astera's opportunity comes from hyperscalers' customized implementations (not the reference platform itself).
Gross margin: Long-term model of ~70%; a rising hardware mix (Taurus) will dilute, but value-add from Scorpio X / UALink can sustain a high-margin range.
⚠️ Risks and Watch Points
Customer concentration: Hyperscalers carry heavy weight; project timing and procurement swings have significant impact.
Standards competition: Progress of alternatives such as Ethernet-based scale-up, and timing differences versus UALink.
Volume execution: Supply-chain and software (COSMOS) integration complexity as Scorpio X ramps in 2026.
Product-mix margin: Shifts in the Taurus hardware mix pull on overall gross margin.
Optical integration: Realizing benefits from the Xscale acquisition and the technical/commercial timeline of scale-up optics (2028–2029 at the earliest).
Regulation/geopolitics: Supply/demand uncertainty in certain regions.
🧭 Bottom Line
Astera Labs sits at the most critical hub of AI rack-scale connectivity:
with Scorpio (scale-up fabric) at the core and Aries (PCIe 6 retimer) and Taurus (Ethernet AEC) as the two wings,
backed by open standards (PCIe / UALink / Ethernet) and the COSMOS software stack,
and paced by Scorpio X high volume in 2026 and UALink commercialization in 2027, it has multiple waves of growth.
In the mid term, watch product-mix effects on margin and the pace of standards competition; in the long term, Scale-Up Photonics could open a new S-curve.


Comments