Earnings Highlights: Astera Labs (ALAB) | FY2026 Q3 — Revenue Up 154%, Scorpio Tops a $1B Run Rate, and the 2027 Handoff Starts in the Second Half
This article is for technology and industry analysis only and is not investment advice.
1. The numbers: a 6% beat, and net income needs two lenses
The STT earnings card uses GAAP with a non-GAAP reference, while management talks mostly in non-GAAP on the call and in guidance. So we show both.
Item | GAAP | YoY | Non-GAAP | YoY |
Revenue | $584.9M | +153.7% | Same | Same |
Gross margin | 72.3% | -4.0pp | 72.4% | -4.0pp |
Operating income | $179.1M | +223.2% | $261.6M | +172.1% |
Operating margin | 30.6% | +6.6pp | 44.7% | +3.0pp |
Net income | $182.9M | +100.8% | $243.4M | +175.9% |
Diluted EPS | $0.99 | +98.0% | $1.32 | +169.4% |
▲ Astera Labs FY2026 Q3 core financials (GAAP, with non-GAAP reference) and Q4 guidance. (Image source: Simple Tech Trend original; data from the company's 8-K press release and earnings call)
Against last quarter's guidance, all four metrics came in ahead:
Revenue: $584.9M versus $540–560M, $24.9M above the high end.
Non-GAAP gross margin: 72.4% versus about 72%.
Non-GAAP operating margin: 44.7% versus about 43%.
Non-GAAP EPS: $1.32 versus $1.16–1.21.
One caveat on the bottom line. GAAP operating income rose 223% YoY while GAAP net income rose only 101%, and the gap is tax: a $24.3M income tax benefit a year ago versus a $9.1M provision this quarter. Tax items also pushed GAAP net income above non-GAAP last quarter, so operating income is the better trend line.
The balance sheet is starting to show the ramp. Receivables grew from $83.2M at the end of last year to $284.3M, and inventory from $59.0M to $200.9M, or 3.4 times. Operating cash flow hit a record $179.2M this quarter, and cash plus marketable securities reached $1.38B. One more line to watch: warrant contra revenue on the cash flow statement reached $30.4M in the first nine months, up from $4.0M a year earlier, so the cost of customer warrants is scaling with shipments.
2. Scorpio: past a $1B run rate, the pitch moves from ports to token economics
Last quarter, in Astera Labs FY2026 Q2 Earnings Highlights, we set three things to track. Here is how they landed:
Would Scorpio become the largest product family? Yes, and X-Series revenue passed P-Series.
Where would gross margin land? 72.4%, in line with guidance.
Second and third Scorpio X customers? Progress, broken down in the next section.
The company says the Scorpio family exceeded a $1B annualized run rate, meaning at least $250M in the quarter. By our math, Scorpio was at least 43% of revenue. The exact figure is undisclosed, so treat that as a floor.
X-Series shipped to two customers across multiple platforms. P-Series also spans multiple platforms, including two US hyperscalers, grew double digits this quarter, and the CFO expects a similar pace in Q4. He was also plain about the mix: with its higher ASP, X-Series will be the dominant growth driver in 2027.
What changed this quarter is the sales language. Jitendra Mohan's premise is that wafers, power and memory are all tight, so customers now optimize for two things: faster time to first token and higher accelerator utilization. All three Scorpio selling points hang on that:
Radix: 320 lanes allow a flat all-to-all topology with fewer hops and lower collective communication latency.
Intelligence: Hypercast and in-network compute offload all-reduce and broadcast from the XPU into the switch, for up to 2x collective performance. With Leo X-Series handling KV cache offload, the company's internal tests show up to 62% faster time to first token and up to 22% more tokens per second.
Software: COSMOS has grown from device diagnostics to rack and fabric management, and it also configures Hypercast and in-network compute, so Astera now takes part in the workload running on the XPU.
Connectivity silicon used to be valued on bandwidth and reach. Astera now frames it as how many more tokens the same accelerators can produce. If customers accept that framing, the pricing basis moves up with it. These performance numbers are company-measured or company-stated and still need production data from customers.
3. The timeline: the most important line came in Q&A
Two analysts asked versions of the same question: after such a steep ramp, could Scorpio X plateau before next-generation platforms arrive? One was more direct, saying expectations are for the lead customer's current-generation build to peak around Q1 2027, with the next generation not ramping until closer to 2028.
The CFO gave the same answer both times: the initial Scorpio X platforms have completed much of their initial deployment ramps and will keep contributing at scale, with growth coming from new platforms and new customers, and no comment on end-customer volumes. He neither denied nor confirmed the premise, and simply pointed to new platforms. The sequential growth rate falling from 49% to 27%, and the quarterly revenue add falling from $192.5M to about $155M at the guidance midpoint, fits that message.
Putting management's scattered dates in one place:
Timing | Event | Certainty in company wording |
Q4 2026 | Scorpio X and P keep expanding; Aries PCIe 6 and Taurus 800G ramp; Aries 7 and NPO parts begin sampling | In guidance |
Q4 2026 shipments | Scorpio X backlog from multiple new customers, then development and qualification in 1H27 | Backlog in hand |
2027 | Second hyperscaler's Scorpio X (pre-production since Q3) starts contributing | "Should" |
Mid-2027 | Leo 2 high-volume ramp | Projected |
2H27 | New-customer Scorpio X revenue; NPO TIA and laser driver revenue (about six months of qualification after sampling) | Expected |
2027 | Orion reaches market through OEMs and ODMs and starts contributing; numbers promised next call | Not yet quantified |
Late 2027 to 2028 | First custom silicon shipments, volume in 2028 (10+ engagements) | Expected |
2028 to 2029 | CPO volume | "Probably" |
Our read: most new customers and new products converge on the second half of 2027. Growth in the first half leans more on existing Scorpio platforms at scale, P-Series, Aries and Taurus. The company did not say the first half will stall, and the CFO stressed that Aries, Leo and optical all layer into the model during 2027. But the table does frame the question to ask: what is the slope in the first half, before the second-half handoff?
4. Content: how "doubling per XPU" actually works
Astera kept repeating that silicon content per XPU doubles each generation, and an analyst asked for the definition on the spot. Jitendra's answer had two steps:
First doubling: some highly loaded systems today already exceed $1,000 per XPU across fabric switches and copper signal conditioning, and next-generation products aim to add $1,000+ from switching alone. Note the condition, "highly loaded": this is not a fleet average.
Each generation after that: add NPO optics and Leo memory controllers, plus higher attach from larger clusters and higher ASPs from faster data rates.
ASPs come with real numbers: each data-rate doubling lifts Aries ASP by about 20–30%, and Smart Cable Modules sell above chip-down parts. Taurus 800G began volume shipments this quarter, and 3.2T retimers are sampling to multiple customers.
Orion is a different calculation. This compute-agnostic rack-scale reference architecture holds up to 10 compute trays with 8 accelerators each, or 80 per rack, and the company puts Astera content at about $100,000 per rack. By our math that is about $1,250 per accelerator, consistent with the "over $1,000" claim. It monetizes three ways: baseboards, COSMOS software, and royalties from OEM and ODM partners. It is also the first time COSMOS gets its own price tag.
The COO says Orion will be accretive to margins. We would put a question mark on that: baseboards are hardware and usually carry thinner margins than chips, and whether software and royalties close the gap has to wait for numbers on the next call. Orion's enterprise and inference cloud buyers also differ from Astera's usual direct hyperscaler model, and this path runs through OEM and ODM channels.
On spending, non-GAAP opex was $162M, up 19% QoQ, with Q4 guided to $194–200M, another 22% or so at the midpoint, and headcount has doubled in a year. The CFO's long-term anchor is opex at about 30% of revenue, versus about 27.7% this quarter. An analyst cited a 70% long-term gross margin target, which by our math implies a long-term operating margin near 40%, below today's 44.7% and the roughly 45% guided for Q4. Some of the investment in four new product vectors will be paid for out of today's excess margin.
5. Optics: a scale-up route without lasers
This is the section for optical readers. Astera gave its most complete optical roadmap so far:
NPO stage: start with TIAs and laser drivers inside the module. Sampling this quarter to select hyperscalers and OEMs, about six months of qualification, revenue in 2H27, and hundreds of dollars per accelerator.
CPO stage: more complex EICs, with volume "probably" in 2028 and 2029.
Optical engines: silicon photonics and connectors were shown at OCP. The goal is fully integrated engines and, eventually, an optically enabled Scorpio X.
Market size: the company estimates a $40B scale-up optical interconnect market in 2030. That is Astera's own definition, and the number can vary widely with scope.
Asked about the Marvell-Lumentum merger (an all-stock deal announced on November 2), Jitendra declined to judge their strategy and described his own: focus on scale-up and build the EICs, silicon photonics and connectors up to the optical engine, while hyperscalers keep sourcing lasers from their existing vendors.
That is a clean dividing line. Marvell is bringing lasers in-house for co-design, and Astera is leaving the light source outside and concentrating on the electro-optical interface. On Marvell's side, Marvell's 2026 Investor Day showed scale-up optics starting from zero, with multi-hundred-million-dollar revenue from FY2028, so the two companies are not far apart on when scale-up optics starts to matter. For the light source itself, see Lumentum's CPO and NPO optical sources talk at OCP, which also covers the optical PCIe demo of 1060 nm VCSELs with Scorpio alongside Astera and Corning.
The risk sits here too. Jitendra stressed on this call that optical links carry far more dollars per link than copper, and that some links that needed no retimer before will now need content. But as we noted in our breakdown of Astera's own OCP talk, its slide states that CPO needs no additional signal conditioning in the end-to-end link, and it draws the hybrid of in-rack copper and rack-to-rack NPO up to 576 accelerators. So Astera's math rests on a premise: the NPO stage has to last long enough, and lost signal conditioning has to be replaced by EIC and optical engine content. CPO volume in 2028 and 2029 is the test.
On protocols, management says little has changed. Customers on memory-semantic protocols like PCIe will migrate to UALink over time, and Ethernet users stay on Ethernet, such as Meta's Tahan scale-up switch. Astera has gone from PCIe to NVLink Fusion, is developing UALink, and says it will address other protocols as customers ask. The goal is the whole scale-up switching market, which also means meeting rivals like Broadcom head-on in the Ethernet camp sooner or later.
6. Q4 guidance
Item | GAAP | Non-GAAP |
Revenue | $730–750M | Same |
Gross margin | About 72% | About 72% |
Operating expenses | $273–279M | $194–200M |
Tax rate | About 20% | About 12% |
Diluted EPS | $1.13–1.17 | $1.62–1.66 |
Non-GAAP operating margin is guided to about 45% on about 187 million diluted shares. The $740M midpoint implies about 173% YoY growth (Q4 FY2025 was $270.6M). Growth comes from continued Scorpio X and P expansion, with Aries PCIe 6 and Taurus 800G driving signal conditioning.
7. Takeaway
Astera has shed the supporting-actor label: $585M in a quarter, a Scorpio run rate past $1B, and four new product vectors in motion at once. It no longer talks about a single retimer but about the whole rack, from switching and memory to optics and software.
But what to read in this report is the timeline, not the numbers. The first Scorpio X wave is mostly deployed, and the next wave of customers, optics and Orion sits in the second half of 2027 or later. The product story has never been wider, and the near-term slope has never leaned harder on a few platforms.
For the next call in early 2027, we are watching three things:
The sequential growth implied by Q1 2027 guidance: the most direct answer to the plateau question.
When the second hyperscaler's Scorpio X moves from pre-production to volume, and whether new-customer backlog grows.
The first Orion numbers: promised for the next call, and a check on whether it really helps margins.
A reminder: this article is for technology and industry analysis only and is not investment advice.
Comments