Date: 2026-02-22 | Reported: 2026-02-10 | Quarter: Q4 FY25 (Dec 2025) Long ALAB ~5.2%
Atlas scored this Conviction 3 (hold, not add). Key points: strong Q4 beat (+8%), structural GM compression, Amazon pricing risk, CFO departure, CRDO dramatically cheaper on growth-adjusted basis (~0.08x EV/Rev/$growth vs. ALAB ~0.35x). Market punished ~20% post-earnings.
My read: Atlas is directionally right on valuation vs. CRDO. But Atlas underweights the platform optionality of what Astera is actually building. That's where muji adds nuance.
Revenue trajectory: +269% → +619% → +207% → +179% → +144% → +150% → +104% → +92% (FY25)
| Q125 | Q225 | Q325 | Q425 | Q126 (guide) | |
|---|---|---|---|---|---|
| Mar-25 | Jun-25 | Sep-25 | Dec-25 | Mar-26 | |
| Revenue ($m) | 159.4 | 191.9 | 230.3 | 270.6 | 286–297 |
| YoY % | 144% | 150% | 104% | 92% | ~37% |
| QoQ % | +13% | +20% | +20% | +17.5% | ~+7.7% |
| GM [Non-GAAP] | 77.2% | 76.7% | 76.3% | 75.7% | ~74% |
| Op Margin [Non-GAAP] | 33.8% | 39.2% | 41.7% | 40.2% | — |
| EPS [Non-GAAP] | $0.33 | $0.44 | $0.49 | $0.58 | $0.53–0.54 |
| Beat vs. guide | +5.2% | +11.2% | +11.7% | +8.0% | — |
FY25 summary: Revenue $852.5m (+115%) | FCF $281.8m (33.1% margin) | Net cash $1,188.8m | Zero debt.
Rule of 40: 92 YoY + 40 op margin = 132 !! (extraordinary, even if forward normalizes to ~77)
Here's the thing — Astera is NOT a signal conditioning company. That's where they started (Ares PCIe retimers). What they're building is a full AI connectivity platform. Let me map the product stack:
Layer 1 — Signal Conditioning (Ares PCIe/CXL retimers) Dominant in hyperscale AI training. ~70% YoY growth FY25. Every AWS Trainium, every NVIDIA-based server cluster needs these. Building block — embedded in customer silicon designs, sticky. Largest revenue today.
Layer 2 — Rack-Scale AI Connectivity (Taurus ethernet cables/modules) 4x growth in FY25 !!. AI racks are getting bigger — more GPUs, more bandwidth per rack. Taurus is in the network fabric. Not commodity ethernet — purpose-built for AI.
Layer 3 — AI Fabric Switching (Scorpio P)
15% of FY25 revenue already. This is Astera entering switching — directly competing with Arista, Broadcom merchant silicon. Scorpio P is the proof-of-concept. The fact that it exceeded 15% of revenue IN YEAR ONE of ramp is !!.
Layer 4 — Scale-Up Networking (Scorpio X + UA Link) Initial Q4 FY25 shipments. Material ramp H2 2026. This is the mega-catalyst. UA Link is an open standard consortium (AWS, AMD, Intel, Meta, Microsoft, Google, Qualcomm) — the alternative to NVIDIA NVLink for scale-up GPU interconnects. AWS Phranyon Four (2027) and AMD MI 500 are committed. This TAM is $20B/yr by 2030. CEO said they're targeting "at least half." That's a $10B/yr opportunity from one product.
Layer 5 — CXL Memory (Leo) First production deployment — Azure M-series VMs (Microsoft, Intel, SAP workloads). Memory pooling is early innings. CXL is the future of disaggregated memory. This is a multi-year thesis.
Layer 6 — Software (COSMOS) This is what nobody's talking about. COSMOS is a software orchestration layer that sits above all the silicon. If Astera can monetize COSMOS as SaaS on top of hardware sales, GM structure changes entirely. This is the option value that's not in any model.
NUANCE: The GM compression from 77% → 74% is Taurus/Scorpio hardware mix. As Astera sells more ethernet cables and switches, those products carry lower GMs than pure silicon (Ares). BUT: (1) Scorpio will improve as it scales (fab economics), (2) COSMOS software layer is GM-accretive, (3) Leo at scale should carry 80%+ GM (semiconductor IP). The 74% floor might not be the floor at all. It's the cyclical trough of hardware-heavy product mix early in ramp cycles.
Platform verdict: Building block !!. Cloud-native architecture. Self-serve customer qualification (hyperscalers qualify in-house). TAM expanding 10x over 5 years. This is exactly "scale in platform."
The 8-K is worth understanding carefully:
Bull read: This is a demand floor. Amazon locked in Astera supply. DING DING DING for supply security.
Bear read: Warrants = pricing discounts. Amazon negotiated volume discounts baked into the warrant structure. Structural GM compression might be partly Amazon pricing, not just mix.
My read: The Amazon relationship is net positive. $6.5B committed over 8 years is extraordinary visibility. The warrant dilution (~1.8% of shares) is manageable. BUT I want to watch whether the "Amazon effect" shows up in Q1 FY26 GM — if 74% guided and they come in at 73%, that's a yellow flag.
Mike Tate → Desmond Lynch (ex-Rambus CFO) effective March 2, 2026.
Rambus is a semiconductor IP company with a royalty licensing model — Lynch understands semiconductor revenue recognition, IP monetization, and complex customer contracts. This could actually be ADDITIVE for the COSMOS software/IP licensing story. Rambus is also known for their memory interface chips work — Leo/CXL expertise alignment.
My take: Not alarmed. Timing is awkward (mid-hypergrowth) but the successor background fits Astera's next chapter.
Broadcom: The elephant in the room. Broadcom dominates merchant semiconductor for networking. If they decide to go after retimers or AI fabric switching with full force, Astera faces a 10x-funded competitor. Counter: Astera is customer-led (hyperscalers design them in), and switching costs once designed-in are high. But this is the #1 long-term risk.
Marvell: Active in AI networking. Watching.
NVIDIA NVLink: The competing scale-up standard. UA Link is the "open" alternative but NVLink has NVIDIA's moat. If hyperscalers keep buying NVIDIA hardware and NVLink dominates, Scorpio X/UA Link ramp gets delayed. This is the #1 near-term Scorpio X risk.
| Period | YoY Growth | Tier |
|---|---|---|
| FY24 | +242% | Tier 1 |
| FY25 | +115% | Tier 1 |
| Q4 FY25 | +92% | Tier 1 (decelerating) |
| FY26 implied | ~37% | Tier 2 transition |
Current tier: Tier 1/2 transition. The deceleration from 150% → 92% YoY is real. Guidance implies further YoY compression to ~37% in Q1 FY26. But QoQ has been remarkably consistent: +13% → +20% → +20% → +17.5%, with Q1 FY26 guided +7.7% (soft — seasonality + pre-ramp).
NUANCE: The YoY decel is almost entirely a base effect problem. FY24 was +242% — impossible comparables. Sequential momentum (QoQ) has been accelerating through FY25. If Scorpio X ramp materializes in H2 FY26, I'd expect YoY to reaccelerate in FY27. This is the key watch.
Market cap ~$28.7B | Net cash 1.19B|Enterprisevalue 27.5B TTM revenue 852.5m → EV/Rev 32xFY26implied 1.16B → Forward EV/Rev ~24x
CRDO comparison (Atlas flagged this): CRDO at ~22x TTM with 3x higher YoY growth. Growth-adjusted, CRDO is dramatically cheaper.
My read: ALAB deserves a platform premium over CRDO (deeper product stack, software layer, broader customer relationships, $6.5B Amazon commitment). But 32x TTM is rich when growth is decelerating to ~37% forward. I wouldn't add here. I wouldn't trim either — the Scorpio X catalyst is real and H2 2026 is close enough to matter.
| Commitment | Made | Status |
|---|---|---|
| Scorpio P >15% FY25 revenue | Q2 FY25 | ✅ Confirmed (>15%) |
| Scorpio X initial shipments Q4 FY25 | Q3 FY25 | ✅ Done |
| Scorpio X material ramp H2 2026 | Q4 FY25 | 🔄 Watching |
| Leo production deployment 2026 | Q3 FY25 | ✅ Azure M-series done |
| UA Link AWS Phranyon Four (2027) | Q4 FY25 | 🔄 Multi-year |
| Amazon $6.5B cumulative through 2033 | Q4 FY25 8-K | 🔄 Long-term |
| SAM $25B in 5 years | Q4 FY25 | 🔄 Multi-year |
The market's reaction (−20% post-earnings) is pricing in: deceleration, GM compression, CFO departure. It's correct on all three. But it's missing three things:
Scorpio X option value. A $10B+ TAM opportunity with committed customers (AWS 2027, AMD MI 500) isn't in a 32x sales multiple in any serious way. It's actually cheap optionality if you believe UA Link wins.
COSMOS software. Zero analysts model COSMOS SaaS revenue. If COSMOS becomes a $50–100m/yr software line in FY27, that alone re-rates the GM structure.
Design-in stickiness. Once Astera silicon is in an AWS server design, it's in for 3–5 years minimum. The $6.5B Amazon commitment validates this. The switching costs are infrastructure-level.
My stance: Hold at ~5%. Not trimming — the Scorpio X/UA Link thesis is real and H2 2026 is the test. Not adding until I see (1) GM stabilization at or above 74%, and (2) Scorpio X revenue confirmation Q2 FY26. If both confirm, ALAB re-enters the Tier 1 consideration at a materially higher forward multiple justification.
-muji Long ALAB ~5.2%