Q1 guide: revenue up 6-10%, EPS down 7-9%. The OpEx ramp hits before the OpEx payback.
Thesis: Astera owns the only PCIe Gen6 switch in volume production while the scale-up protocol war (UALink vs ESON vs proprietary) plays out, and both Amazon and NVIDIA independently selected them for NVLink Fusion with one-per-accelerator attach. But the street is being asked to underwrite 2027 catalysts today while Q2 EPS compresses ~20% under OpEx ramp plus a 200bp warrant charge. The structural positioning is real; the near-term earnings trajectory is not.
Verdict: HOLD — Conviction: MEDIUM
Catalyst: Scorpio X Series volume ramp at lead customer in back half 2026, plus Amazon warrant tranches beginning to vest against revenue milestones in Q2 2026.
Key Risk: Every major catalyst (Scorpio X volume, two new hyperscaler design wins, UALink, optical) produces revenue in 2027 or later. 2026 rests on the existing Amazon Scorpio P ramp, Ares Gen6, and Taurus 400G while OpEx expands 17-22% QoQ.
The Tell: Jitendra Mohan volunteered: 'We have very minimal opportunity with the reference design itself' when asked about Astera's content in NVIDIA Vera Rubin. This reveals their entire revenue thesis depends on hyperscalers choosing custom deployments over NVIDIA's reference racks. If reference designs win, the Scorpio TAM assumption breaks.
Friction Level: MODERATE_FRICTION — Both sides agree Q4 is clean and the structural story is real. Disagreement is whether ~370bp of margin compression (hardware mix + 200bp warrant charge) in two quarters matters when the 2027-2028 catalysts (Scorpio X volume, UALink, optical) are still 12-36 months out.
2025Q4 · Feb 10, 2026
4 of 5 readings agreed4 HOLD · 1 SHORTRead the full call
EPS Down 8% While Revenue Rises 8%, And Management Calls It Investment
Astera owns the only PCIe Gen6 switch in volume production while the scale-up protocol war (UALink vs ESON vs proprietary) plays out, and both Amazon and NVIDIA independently selected them for NVLink Fusion with one-per-accelerator attach. But the street is being asked to underwrite 2027 catalysts today while Q2 EPS compresses ~20% under OpEx ramp plus a 200bp warrant charge. The structural positioning is real; the near-term earnings trajectory is not.
Jitendra Mohan volunteered: 'We have very minimal opportunity with the reference design itself' when asked about Astera's content in NVIDIA Vera Rubin. This reveals their entire revenue thesis depends on hyperscalers choosing custom deployments over NVIDIA's reference racks. If reference designs win, the Scorpio TAM assumption breaks.
Q4 revenue $270.6M beat by 17% QoQ and 92% YoY. Q1 guide $286-297M extends the streak. Scorpio P exceeded 15% of revenue against 10% bogey. FY2025 revenue $852.5M up 115%.
Revenue guides up 6-10% in Q1 but EPS guides down 7-9% ($0.58 to $0.53-0.54). OpEx jumps 17-22% QoQ against single-digit revenue growth. Management says 'now is the time to invest' with 18-24 month payback.
Gross margin guides down 170bp to 74% on hardware mix shift, then another 200bp warrant charge from Q2. Total ~370bp compression in two quarters. Bear flags this as masking structural margin pressure.
The $6.5B Amazon warrant is milestone-based, earned as purchases are achieved. It signals intent rather than firm non-cancellable commitment. Scorpio X ships preproduction only. Volume ramps are 2027.
Both Amazon and NVIDIA independently selected Astera for NVLink Fusion, attaching one-per-accelerator matching full switch revenue content. Software-defined fabrics create integration depth across protocol transitions.
OpEx steps from $96M to $112-118M QoQ, deploying against Israel acquihire, optical development, and fabric expansion. Management states they are building optical capabilities 'that other companies are paying billions to get externally.'
New TAM quantifications this quarter: $25B served market in 5 years and $20B scale-up switching by 2030. Both are decade-long payoff windows against near-term EPS compression. Optical not until 2028.
-25
CAUTIOUS"Management proactively disclosed negative datapoints: EPS declining despite revenue growth, 18-24 month OpEx payback, optical revenue not until 2028, warrant charge at 200bp/quarter from Q2. Hedging language: 'on the earlier side' for payback, 'at some point' for product crossover."
Specific guidance ranges, OpEx bridge, margin waterfall, and warrant accounting disclosed in detail. NVLink Fusion business model deflected under NDA. Scorpio X hyperscaler identities not disclosed.
Scorpio X Series volume ramp at lead customer in back half 2026, plus Amazon warrant tranches beginning to vest against revenue milestones in Q2 2026.
Every major catalyst (Scorpio X volume, two new hyperscaler design wins, UALink, optical) produces revenue in 2027 or later. 2026 rests on the existing Amazon Scorpio P ramp, Ares Gen6, and Taurus 400G while OpEx expands 17-22% QoQ.
“Astera Labs, Inc. Common Stock delivered strong results in Q4 with revenue at $270.6 million, up 17% from the prior quarter and up 92% versus Q4 of last year.”
“We expect Q1 revenues to increase to within a range of $286 million to $297 million, up roughly 6% to 10% from the fourth quarter levels.”
Figures are extracted from the transcripts and checked in code: every quote appears word for word. Descriptive only: none of this feeds the verdict.
Q4 revenue $270.6M up 92% YoY with clean cash conversion ($95.3M OCF vs $105M non-GAAP NI). Scorpio P beat its 10% bogey at 15% and is the only PCIe Gen6 switch in volume. Both Amazon and NVIDIA selected Astera for NVLink Fusion with one-per-accelerator attach matching full switch content. Two new US hyperscaler Scorpio P design wins confirmed. AWS and AMD publicly endorsed UALink for 2027. Taurus grew 4x, Ares grew 70%, all five product lines expanding simultaneously. The $6.5B Amazon warrant vests through 2033 against revenue milestones across switches, signal conditioning, and optics. This is the only connectivity company shipping in every socket across scale-up, scale-out, and memory.
Q1 guides revenue up 6-10% but EPS down 7-9% as OpEx jumps 17-22% QoQ. Gross margin drops 170bp to 74% on hardware mix, then another 200bp warrant charge from Q2. Combined ~370bp compression. Every major catalyst produces revenue in 2027+: Scorpio X volume, two new hyperscaler wins, UALink, NVLink Fusion deployment. Optical not until 2028. OpEx payback is 18-24 months minimum. The warrant is milestone-based, earned as purchases occur, so it is intent signal not committed backlog. 3.3M warrant shares dilute a 181M base. CFO transition March 2 during spending step-up. 2026 depends entirely on existing Amazon Scorpio P ramp plus Ares Gen6 and Taurus 400G while OpEx runs ahead of revenue.
Say/Do Gap
Revenue guides up 6-10% while EPS guides down 7-9% sequentially. OpEx rises 17-22% QoQ against single-digit revenue growth. Management explicitly states payback is 18-24 months at earliest.
Mix Shift Spin
170bp hardware mix compression plus 200bp non-cash warrant charge equals ~370bp stacked over two quarters. The non-cash classification means the P&L optics worsen while narrative says it does not matter.
Backlog Quality
$6.5B Amazon warrant is earned against milestones as purchases occur. It signals intent through 2033 rather than firm non-cancellable commitment. Scorpio X and new hyperscaler wins all produce revenue in 2027.
Unsustainable Trend Confidence
$25B TAM and $20B scale-up switching claims are new this quarter with decade-long payoff windows, deployed alongside a 17-22% OpEx step-up.
HOLD. The structural positioning is genuinely differentiated: only PCIe Gen6 switch in volume, dual NVLink Fusion selection from Amazon and NVIDIA, UALink tailwind confirmed by AWS and AMD. But 2026 EPS compresses under ~370bp margin headwind while every material catalyst lands 2027+. The bear's tactical short on Q2 EPS trajectory when the 200bp warrant charge hits has merit, but the stock is not structurally broken. Wait for Q2 print to see how the market digests the warrant charge and whether the Scorpio X ramp begins tracking to back-half 2026. Entry point improves on post-Q2 weakness if the structural story remains intact.