铜连接真不行了？Astera Labs季报后大跌！

01

—

ALAB

摩根士丹利：

Astera delivered stronger than previewed top line results and outlook, with strong growth drivers in 2h around scale-up; this was offset by a material step up in opex to focus on significant longer term growth opportunities. Stay OW.

Key Takeaways

Very strong top line performance, with March guidance more than 10% above consensus, in what we had viewed as a transitional phase before Scorpio X ramp 2h

Opex for CY26 looks about 20% above our prior expectations, amid heavy investment in next generation technologies including optical scale up

CFO transition to advisory role is a negative at the margin

Best growth opportunities still to come; stay OW

Astera 的营收结果和业绩展望均强于我们此前预期，2026 年下半年围绕 scale-up 的增长驱动因素强劲；但这被经营费用（opex）的大幅上升所部分抵消，公司将投入用于更具长期意义的增长机会。维持 OW（超配）评级。

要点

营收表现非常强劲，3 月季度指引较一致预期高出超过 10%，而此前我们认为在 2026 年下半年 Scorpio X 放量前这应属于一个过渡阶段

2026 年 opex 看起来较我们此前预期高约 20%，原因是对包括 optical scale up 在内的下一代技术进行重投入

CFO 转任顾问角色在边际上是一个负面因素

最好的增长机会仍在后面；维持 OW

② 业务进展：Scorpio / Aries / Taurus / Leo + 标准之争

Strong execution with Scorpio momentum building and scale-up opportunity ahead. Astera delivered another strong quarter and outlook, with growth broad-based across Scorpio, Aries, and Taurus. While near-term results continue to be supported by signal conditioning and active cable products, the longer-term story around scale-up fabrics and Scorpio is increasingly firming up. Management highlighted that Scorpio P exceeded 15% of FY25 revenue, well ahead of their original 10% target. Right now the contribution is almost entirely driven by the P-Series (scale-out), where they announced two additional hyperscaler customers this quarter. They continue to expect a material ramp in Scorpio-X (scale-up) in the 2H of 2026 into 2027. Eventually, Scorpio-X will be their largest product lines, though the other products are still expected to grow.

ALAB is also performing well outside of switches - Aries PCIe grew nearly 70% y/y, Taurus AEC was the strongest grower in Q4 and grew more than 4x y/y in 2025, and Leo CXL, while in early stages, continues to see progress with initial production volumes expected in 2H 2026.

Scorpio 动能持续增强、scale-up 机会在前，执行表现强劲。Astera 再次交出强劲的季度业绩与指引，增长覆盖 Scorpio、Aries 与 Taurus，呈现广泛驱动。尽管短期业绩仍主要由信号调理与有源线缆产品支撑，但围绕 scale-up fabrics 和 Scorpio 的长期逻辑正在进一步夯实。管理层强调，Scorpio P 在 FY25 的收入占比已超过 15%，显著高于最初设定的 10% 目标。目前的贡献几乎完全由 P-Series（scale-out）驱动，公司本季度宣布新增两家超大规模客户。公司仍预计 Scorpio-X（scale-up）将在 2026 年下半年开始显著放量并延续至 2027 年。最终，Scorpio-X 将成为其最大的产品线之一，尽管其他产品也仍预计会继续增长。

除交换机之外，ALAB 的表现也很强：Aries PCIe 同比增长接近 70%；Taurus AEC 是 Q4 增长最快的业务，并在 2025 年实现同比超过 4 倍增长；Leo CXL 虽仍处早期阶段，但持续取得进展，预计将在 2026 年下半年出现初始量产出货。

③ UALink / NVLink Fusion 细节 + 光互连投入 + Opex 上调

The debate around scale-up standards remains noisy - spanning UALink, NVLink Fusion, Ethernet/ESUN, and proprietary approaches - but Astera maintains they are well positioned across all of these vectors. Public endorsements of UALink from AWS (Trainium 4) and AMD (MI500), alongside Astera’s role in NVLink Fusion, position ALAB well for this multi-protocol greenfield opportunity. On NVLink Fusion, they described a revenue opportunity roughly comparable to their Scorpio switch deployments, though architectures differ. While a native Scorpio switch is typically shared across multiple accelerators, the NVLink Fusion solution attaches one-to-one to each XPU, resulting in a higher attach rate that offsets a potentially lower ASP per device. They also emphasized that NVLink Fusion is additive, opening access to an ecosystem where Astera previously had no content. They also clarified that hybrid racks are custom ASICs that use NVLink for scale-up, and they don't expect mixed racks that use both GPUs and ASICs.

A meaningful portion of the incremental OpEx is also being directed toward optical connectivity for scale-up, which is going to be expensive - and eventually, for the winners, lucrative. While copper is expected to remain the preferred medium for as long as feasible, the company believes optical engines will become increasingly important as bandwidth and reach requirements exceed copper’s limits, with initial scale-up deployments likely around 2028. ALAB views this transition to optics as market expanding, and emphasized that its investments in silicon photonics, optical engines, and related IP are being made in close alignment with customer roadmaps. Management suggested that optical could ultimately represent a larger opportunity in today’s signal conditioning business, reinforcing the rationale for stepping up R&D spend now to secure long-term leadership in next-generation scale-up fabrics.

While the step-up in opex pressures earnings, our sense is they are attacking a market that MRVL just paid over $5bn to enter, and think it's worth the spend to be part of this opportunity. ALAB is doing this mostly organically, in addition to their recent acquires and X-Scale acquisition, which is expensive and should pressure operating margins for the foreseeable future. We are modeling OPEX to increase to $485mn in CY26 (vs. our previous model of $393mn), up from $312mn in CY25. Similar to what we have seen in peers investing in AI - notably AMD - its probably fair to consider that revenue upside will drive offsetting opex upside as well

围绕 scale-up 标准的争论依然嘈杂——涵盖 UALink、NVLink Fusion、Ethernet/ESUN 以及各类专有方案——但 Astera 坚称其在所有这些方向上都具备良好布局。AWS（Trainium 4）和 AMD（MI500）对 UALink 的公开背书，再加上 Astera 在 NVLink Fusion 中的角色，使 ALAB 在这一多协议的全新（greenfield）机会中具备良好定位。关于 NVLink Fusion，公司描述其收入机会大体可与 Scorpio 交换机部署相当，尽管架构不同。原生 Scorpio 交换机通常在多张加速卡之间共享，而 NVLink Fusion 方案以一对一方式连接到每个 XPU，从而带来更高的 attach rate，可抵消单设备 ASP 可能更低的影响。公司还强调 NVLink Fusion 具有增量属性（additive），使其得以进入一个此前 Astera 没有内容（content）的生态系统。公司同时澄清，混合机架（hybrid racks）是使用 NVLink 做 scale-up 的定制 ASIC，并且他们不预期会出现同时使用 GPU 和 ASIC 的混合机架。

新增 opex 的相当一部分也被投入到 scale-up 的光互连（optical connectivity）方向，这会很昂贵——但最终对赢家而言也会很赚钱。尽管只要条件允许，铜仍预计会尽可能长时间作为首选介质，但公司认为当带宽与传输距离需求超过铜的能力边界时，光引擎将变得越来越重要，初始 scale-up 的光部署可能在 2028 年左右出现。ALAB 将这一向光的转变视为“扩容市场”（market expanding），并强调其在硅光子、光引擎及相关 IP 上的投资与客户路线图保持紧密一致。管理层表示，光最终可能在当今的信号调理业务中代表一个更大的机会，这进一步强化了当下提高研发投入、以获取下一代 scale-up fabrics 长期领导力的理由。

虽然 opex 的上升对盈利构成压力，但我们的感觉是，公司正在进攻一个 MRVL 刚刚花费超过 50 亿美元才进入的市场，并认为为了参与这一机会而投入是值得的。ALAB 主要以有机方式推进这一布局，同时叠加其近期的收购以及 X-Scale 收购；后者成本高昂，预计将在可预见的未来对经营利润率形成压力。我们模型预计 ALAB 的 2026 年 OPEX 将升至 4.85 亿美元（此前模型为 3.93 亿美元），高于 2025 年的 3.12 亿美元。类似我们在其他加大 AI 投资的同业（尤其是 AMD）身上看到的情况，可能需要公平地认为：收入上行也会带来相应的 opex 上行。

④ 财务细节 + 指引 + 预测调整 + 股价看法与估值

Details on the quarter: Revenue of $260.6mn (up 17.4% q/q and 91.8% y/y) was above the Street and our estimate of $250mn. Gross margin of 75.7% (down 71bps q/q) was 60bps above the Street at 75.1% and 70bps above our estimate of 75.0%. Non-GAAP EPS of $0.58 was also ahead of the Street and our estimate of $0.52.

Details on guidance: The company expects MarQ revenue of $291.5mn at the midpoint (up 7.7% q/q and 82.8% y/y), above the Street and our estimate of $260mn. Gross margin was guided to 74.0%, below the Street at 77.2% and above our estimate of 72.3%. EPS guide of $0.54 was ahead of the Street at $0.52 and our estimate of $0.51.

Changes to estimates: We had previously modeled 41% and 27% top-line growth in 2026 and 2027, respectively. We now model 54% and 28%, reflecting the stronger ramp of all products and the contribution of Scorpio-X in the 2H of this year. We are slightly lowering GM in CY2026 from 71.8% to 71.3% and modeling a decline to 69.1% in CY27. Our Non-GAAP EPS increases from $2.18 to $2.29 in CY2026 and we are modeling EPS of $3.09 in CY27.

Thoughts on the stock: We have viewed 1h26 as a transitional phase, with the importance of the Scorpio X ramp with Amazon's Trainium 3 in 2h, so the continued upside is welcome here. Astera offers one of the best growth rates in our coverage, with multiple product cycles and a very strong position in key AI technology - with visibility into multiple product cycles that have yet to even start ramping, but the stock is quite expensive, with an opex ramp that is limiting EPS upside on strong revenue upside, but we see strong persistent growth. We are keeping our PT at $210. The stock remains expensive, thought the multiple comes down as estimates go up; we are now valuing ALAB on 0.65x CY27 EV/S/G on 1.7bn revenue, assuming 34% 2026-2028 CAGR. This is roughly 21x CY27 sales.

季度细节：收入为 2.606 亿美元（环比增长 17.4%，同比增长 91.8%），高于市场一致预期以及我们 2.50 亿美元的估计。毛利率为 75.7%（环比下降 71 个基点），比市场一致预期的 75.1% 高 60 个基点，也比我们 75.0% 的估计高 70 个基点。Non-GAAP EPS 为 0.58 美元，同样高于市场一致预期以及我们 0.52 美元的估计。

指引细节：公司预计 3 月季度收入中枢为 2.915 亿美元（环比增长 7.7%，同比增长 82.8%），高于市场一致预期以及我们 2.60 亿美元的估计。毛利率指引为 74.0%，低于市场一致预期的 77.2%，但高于我们 72.3% 的估计。EPS 指引为 0.54 美元，高于市场一致预期的 0.52 美元以及我们 0.51 美元的估计。

预测调整：我们此前对 2026 年和 2027 年的收入增速分别建模为 41% 和 27%。我们现在将其调整为 54% 和 28%，反映出所有产品更强的放量节奏以及今年下半年 Scorpio-X 的贡献。我们将 2026 年毛利率从 71.8% 小幅下调至 71.3%，并预计 2027 年毛利率下滑至 69.1%。我们的 2026 年 Non-GAAP EPS 从 2.18 美元上调至 2.29 美元，并预计 2027 年 EPS 为 3.09 美元。

对股价的看法：我们此前将 2026 年上半年视作一个过渡阶段，核心在于 2026 年下半年随亚马逊 Trainium 3 推动 Scorpio X 放量的重要性，因此当前继续出现的上行是令人欢迎的。Astera 提供了我们覆盖范围内最好的增长率之一，具备多轮产品周期，并在关键 AI 技术中拥有非常强的位置——且对多个尚未开始放量的产品周期具有可见性；但该股估值非常贵，opex 的上升在强劲收入上行的同时限制了 EPS 的上行空间，不过我们仍看到强劲而持续的增长。我们维持目标价 210 美元。该股仍然昂贵，尽管随着盈利预期上修，估值倍数会下移；我们目前基于 2027 年 17 亿美元收入、假设 2026–2028 年 CAGR 为 34%，用 0.65x 的 2027 年 EV/S/G 对 ALAB 进行估值，这大致相当于 2027 年收入的约 21 倍。