Earnings Highlights: Astera Labs (ALAB) | FY2026 Q2 — Scorpio Takes the Top Spot a Quarter Early as Connectivity Chips Become the Nervous System of the AI Rack

This article is for technology and industry trend analysis only and does not constitute investment advice.

• Quarterly revenue of $392.4M, up 104% YoY and 27% QoQ; PCIe 6.0 products jumped past half of total revenue (from just a third last quarter). This wasn't one product breaking out — Aries, Scorpio and Taurus all ramped together.

• Scorpio will become the company's largest product line in Q3, a full quarter earlier than expected. The high-radix Scorpio X-Series is in volume production, and management said it will "surpass Scorpio P-Series this quarter" — effectively announcing Astera's promotion from "signal conditioning and memory connectivity" to "AI Fabric (AI switching) supplier."

• Q3 guidance midpoint of $550M, another 40% QoQ jump, while non-GAAP gross margin is guided down from 73.7% to about 72%. Trading a bit of margin for growth is a deliberate choice the company is now very clear about.

---

1. Revenue and Financials: 104% YoY Growth — the Real Story Is in the "Basis" and the "Mix"

• Revenue of $392.4M, up +104% YoY and +27% QoQ — a quarterly record.

• Non-GAAP gross margin of 73.7%, down 2.3pp from 76.0% a year ago — one of the few numbers that went down this quarter; we'll explain why below.

• Non-GAAP operating income of $153.5M, operating margin 39.1%, up 290 basis points QoQ; operating income was up +104% YoY. Revenue doubled while operating expenses grew only 10% ($135.8M) — operating leverage straight to the bottom line.

• Non-GAAP net income of $145.8M (+87% YoY), non-GAAP diluted EPS of $0.80 (+82% YoY, up more than 30% QoQ).

• Cash, cash equivalents and marketable securities totaled $1.25B; tax rate 12%; interest and other income $12.1M.

There is one easily misread point STT wants to head off: this quarter's GAAP net income ($153.1M, GAAP EPS $0.83) was actually higher than non-GAAP, mainly because a one-time tax benefit lifted the GAAP bottom line. To judge Astera's "true earnings power," look at the non-GAAP/comparable basis rather than being carried away by the flattering GAAP EPS — which is why our earnings card and this article use non-GAAP throughout.

The real signal isn't any single number but the mix: the PCIe 6.0 generation went from a third of total revenue last quarter to more than half. That means Astera's core business has completed the generational handoff from PCIe 5 to PCIe 6, with pricier, higher-value parts following behind.

2. Technology and Business Highlights: Scorpio Arrives a Quarter Early as the Largest Product Line

The most important line of the quarter came from the CFO: Scorpio will become Astera's largest product family by revenue in Q3, one quarter earlier than previously expected.

Two lines to unpack:

Scorpio X-Series (high-radix AI Fabric switching) is in volume production. This is the key to Astera moving from connectivity sidekick to AI-switching lead. Management said Scorpio X alone will "surpass Scorpio P-Series revenue this quarter," and that "more than 10" customers are already running on Scorpio X, most entering pre-production or deeper qualification. The X-Series targets an approximately $20B merchant scale-up switching TAM, and its differentiation is not port count but the "intelligence layer" — hardware-accelerated in-network compute and Hyper-cast, claimed to improve collective-operation efficiency for large training/inference by nearly 2x, delivered through Astera's own COSMOS software platform.

Notably, COSMOS's role is being upgraded: from telemetry, diagnostics and optimization to part of the orchestration layer — XPUs now talk to COSMOS directly to schedule workloads. That "software stickiness" is the moat Astera wants to lock customers into the next generation with.

Scorpio P-Series goes for breadth. The P-Series targets standard PCIe switching — from storage and networking to some scale-out use cases — with many customers and varied applications. X goes deep, P goes wide: that's Astera's two-track design for the Scorpio family.

On the question of "who is competing with Astera for scale-up switching," there's no getting around incumbent switching giants like Broadcom — in Earnings Highlights: Broadcom (AVGO) | Q2 FY26 we broke down the order structure that extends its AI visibility to 2028. Read the two side by side and it becomes clearer how a handful of players are splitting the scale-up switching pie.

The other three product lines are no sideshow either:

• Aries (signal conditioning, retimer/gearbox) hit a quarterly record and was still the largest product line in Q2. Interestingly, the surprise came from "Gen 5 lasting longer than expected" — as inference took off, PCIe 5 retimers saw another wave of demand, while PCIe 6 (64 GT/s) brings higher attach rates and ASPs. In management's words, the retimer socket is "a socket that keeps on giving."

• Taurus entered pre-production at 100G/lane this quarter, supporting 800G links; the newly launched 200G/lane retimer/redriver lifts Taurus's market opportunity to more than double, to over $4B, by 2030.

• Leo (CXL memory controller) won a new design at a U.S. hyperscaler this quarter, and in 2027 will ship standard and custom Leo to two U.S. hyperscalers. After a long lull, CXL has momentum again, driven by tight memory supply and pricing plus the memory-capacity bottleneck of agentic AI.

3. Management Outlook: Q3 Midpoint Jumps Another 40% — Why Gross Margin Is Heading Down

Q3 FY2026 guidance (non-GAAP):

• Revenue of $540–560M, midpoint $550M, up about +40% QoQ.

• Non-GAAP gross margin of about 72% (73.7% this quarter, heading down).

• Non-GAAP operating margin of about 43%, up another 400 basis points QoQ.

• Non-GAAP diluted EPS of $1.16–1.21, on about 185 million diluted shares.

The key to reading this guidance is gross margin. The CFO was blunt: as the mix tilts toward switching (Scorpio), and with the revenue split between silicon and modules plus differences across use cases and lane counts, the gross margin range will widen, with the long-term target trending toward 70%. In other words, Astera is trading a bit of gross margin for big-ticket, high-ceiling business like scale-up switching — and operating margin is rising rather than falling (43%), which shows the math works: as scale grows, absolute gross-margin dollars still flow to the bottom line.

In tone, management spoke this quarter in terms of "confirmation" rather than "expectation" — Scorpio reaching the top early is already happening (in volume production), not a promise. That certainty deserves more attention than the guidance numbers themselves.

4. Supply Chain and Customer Clues: Ten Scorpio X Customers, UALink, and That Optical Roadmap

STT considers this the most valuable part, because it decides whether Astera is "a winner of this wave" or "a winner of this generation."

Dollar content per XPU is still climbing. At IPO, Astera's silicon content per accelerator was in the $50–100 range; now, for Scorpio X scale-up switching alone, management says "future generations could exceed $1,000 per XPU," and with signal conditioning and optics stacked on top, total content is heading toward "thousands of dollars per XPU." This rising content curve is the core of judging whether Astera can keep growing fast.

UALink is the next battleground. Management confirmed that the UALink 2.0 spec writes in-network compute and 200G SerDes — "directions Astera is already investing in" — into the standard, and previewed a UALink-capable Scorpio X scale-up switch in 2027. UALink's appeal to customers isn't just bandwidth but low latency — especially suited to inference. To understand how new scale-up standards like UALink and XPO position against each other, see What Is XPO? The Loudest New Scale-Up Standard at OFC 2026, and OIF's official map of AI interconnect, Three Networks, One pJ/bit Battleground — together they place what Astera said this quarter within the industry's full coordinate system.

Copper and optics will coexist, but optics' role keeps growing. Asked about optics, the COO was pragmatic: in-rack scale-up stays copper this generation and next; but as clusters grow from one or two racks to four, five or more, optics will come in, and Astera's strategy has three stages — NPO (near-packaged optics) first in 2027, CPO (co-packaged optics) after 2028, and ultimately "Scorpio switches with built-in optical engines." The key point: moving from copper to NPO, Astera's dollar content per XPU / per link goes up. We broke down the physical timeline of "the copper bottleneck retreating into the rack while optics moves in from outside" in Computex 2026: Marvell Spells Out the Physical Timetable for Optical Interconnect; and for whether the scale-up endgame goes SiPh interposer, compare with The Scale-Up Endgame Architecture After Marvell Acquires Celestial AI.

The customer base is broadening. Scorpio X is still driven mainly by a product transition at one lead hyperscaler, but management made a point of adding that "even at this lead hyperscaler, we're designed into multiple platforms, not just one." Add more than 10 customers in qualification, a new use case of neo-clouds running inference on add-in-card GPUs, and the China market (mainly open PCIe and Ethernet protocols, expected to stay in the single digits as a share) — Astera's customer concentration is loosening, a prerequisite for moving from "single-big-customer dependence" to "platform-level supplier."

5. Conclusion: Three Numbers to Watch Next Quarter

Astera made one thing clear this quarter: "connectivity," the least glamorous part of AI infrastructure, is becoming one of the highest-value, stickiest links in the chain. It holds the retimer socket that "keeps on giving" with Aries, pushes itself onto the AI Fabric table with Scorpio, and paves the 2027–2028 runway with Leo and its optical roadmap. When a connectivity chip company starts talking about "thousands of dollars of content per XPU" and "switches with built-in optical engines," it no longer wants to be a sidekick — it wants to be the nervous system of the rack.

Next quarter (Q3 FY2026), STT will be watching these three numbers:

1. Whether Scorpio really becomes the largest product line, and by how much Scorpio X overtakes P — the first check on whether the "AI Fabric supplier" narrative pays out.

2. Where gross margin lands: guidance is about 72% — will it come in lower if the switching share rises faster? Whether absolute gross-margin dollars keep up matters more than the percentage.

3. When the second and third Scorpio X volume customers appear — management says late this year to early next year. Whether customer concentration truly loosens decides if this is a repeatable growth model.

A reminder: this article is for technology and industry trend analysis only and does not constitute investment advice.

Related Reading

• Earnings Highlights: Broadcom (AVGO) | Q2 FY26: see clearly how a few players are splitting the scale-up switching pie.

• What Is XPO? The Loudest New Scale-Up Standard at OFC 2026: understand the scale-up standards fight Astera refers to.

• OIF Draws an Official Map of AI Interconnect: Three Networks, One pJ/bit Battleground: place this quarter's technical announcements within the full interconnect landscape.