AI agent frenzy fuels interconnect upgrades! From "connecting chips" to "connecting entire racks," Morgan Stanley loudly calls for overweighting Astera Labs (ALAB.US)

Title context: AI agent frenzy fuels interconnect upgrades! From "connecting chips" to "connecting entire racks," Morgan Stanley loudly calls for overweighting Astera Labs (ALAB.US) Text: After Astera Labs (ALAB.US), the semiconductor giant focused on intelligent connectivity solutions for AI data centers, reported its results, Wall Street financial giant Morgan Stanley immediately issued a bullish research report. The institution's analyst team said that Astera Labs is converting AI interconnect demand into revenue and profit, but whether the stock price can shift into a new bull-market trajectory depends on whether the market sufficiently believes that 2027 growth can span a single customer and a single product cycle. The Morgan Stanley analyst team led by Joseph Moore wrote in the report that it maintains an "Overweight" rating on Astera Labs and an "Attractive" investment strategy judgment for the industry, and raised the target price from $382 to $389, implying 74.2% potential upside relative to the November 4 closing price. The Morgan Stanley analyst team's overall bullish logic and the AI infrastructure expansion narrative run through the entire research report: Trainium 3 drives volume ramp of Scorpio switch chips, followed by customer and platform expansion, higher value per accelerator, and subsequent growth from CXL memory, optical interconnect, and platform solutions, positioning the company to continue outperforming the overall AI infrastructure investment growth rate. Astera Labs' third-quarter revenue rose 153.7% year over year to $584.9 million, and the midpoint of management's fourth-quarter revenue guidance range was about 12.8% above market expectations. However, the strong results still did not reverse the pressure of a sharp stock price decline amid valuation contraction in the semiconductor sector, surging AI financing costs, and significant cash flow pressure on cloud computing giants. The team led by Morgan Stanley analyst Joseph Moore emphasized in the report that it has high confidence in the continuity of the company's 2027 growth: as Scorpio switch chips ramp, customers and application platforms expand, and memory connectivity, optical interconnect, and rack-level solutions gradually contribute revenue, Astera Labs is expected to grow further from a high-speed signal conditioning vendor into an AI interconnect platform supplier. The key support for the Morgan Stanley analyst team's bullish judgment is that the company, by increasing the value of products per AI accelerator, is turning AI infrastructure expansion into sustained growth that can span a single customer's procurement cycle. From "connecting chips" to "connecting entire racks": where is the 2027 growth relay? What Morgan Stanley values most is that Astera Labs' scope of value capture in large AI computing cluster systems is expanding. At the time of its IPO, Astera Labs mainly solved high-speed signal transmission issues; today, Scorpio has entered the switching network that determines the communication efficiency of accelerator clusters. The report attributes recent growth mainly to volume ramp of Amazon.com, Inc.'s Trainium 3, and believes that evolution of next-generation platforms toward UALink and NVLink Fusion is expected to further increase the value of switch chips. Astera Labs' earnings call provided a clearer revenue base: the annualized revenue of the entire Scorpio family has exceeded $1 billion, with X-series revenue exceeding P-series; here the company's own framing should be used, and for now the full $1 billion cannot be attributed entirely to the X-series, where demand is stronger. The company's management said that in existing high-configuration systems, the combined value of switch chips and copper interconnect signal conditioning can exceed $1,000 per accelerator, and in the next generation switch chips alone are expected to reach this level. Morgan Stanley said revenue growth therefore comes simultaneously from an increase in the number of accelerators and from more Astera Labs products per accelerator. Morgan Stanley's report highlights the shift from "connecting chips" to "connecting entire racks," specifically reflected in Astera Labs' capabilities extending from signal assurance for a single high-speed link to coordination of compute, switching, and memory resources across an entire rack. Morgan Stanley said that early products represented by Aries retimers mainly addressed signal integrity issues in high-speed PCIe/CXL transmission; today, Scorpio switch chips connect multiple accelerators, Leo memory controllers expand available memory, and COSMOS software extends management scope from a single device to the entire switching network and AI server racks. On this basis, Astera Labs' Orion series integrates chips, boards, compute trays, switch trays, and memory trays into a rack-level reference architecture, delivered through OEM/ODM partners to help customers reduce duplicate design, validation, and system integration work. The company's business model has also expanded accordingly: in addition to chip and board sales, the company has begun generating revenue through software and reference design licensing; management estimates that a fully configured single rack with 80 GPUs can carry about $100,000 of Astera Labs-related value. With Orion expected to enter the market in 2027, the company is expected to simultaneously increase product content and revenue sources per AI system, turning interconnect device advantages into rack-level platform growth opportunities. The basis for the Morgan Stanley analyst team's view that "strong growth in 2027 will not stall" is that multiple product and customer curves will follow in succession, rather than repeating this year's explosive sequential growth rate. A second Scorpio X hyperscale cloud customer is still in the pre-production stage, and other new customers are expected to complete development certification in the first half of 2027 and contribute revenue in the second half; Leo memory products are expected to ramp at midyear, near-package optics (NPO)-related products are expected to begin contributing revenue in the second half, and custom solutions are mainly aimed at ramping in 2028. Morgan Stanley's quantitative model has in fact already lowered sequential revenue growth in the first two quarters of 2027 to about 4.1% and 3.6%, but still forecasts full-year growth of 67.5%. These forecast dynamics also reveal the report's true stancethat growth can shift from a steep ramp to sustained expansion, and the key is whether new customers, new platforms, and new products can catch the first round of deployment; the migration from copper interconnect to optical interconnect therefore involves both replacement of existing products and incremental opportunities from optical drivers, transimpedance amplifiers (TIAs), and higher system value. Why AI agent workloads need Astera Labs: making expensive compute wait less and produce more! Morgan Stanley sharply raises earnings forecasts Astera Labs' main business can be summarized as data interconnect, switching, and memory connectivity between AI servers and racks: Aries handles PCIe/CXL high-speed signal conditioning, Scorpio handles switching networks, Taurus serves high-speed Ethernet connectivity, Leo provides memory expansion, and COSMOS handles diagnostics, telemetry, and system management. Its economic value is reflected in reducing waiting caused by data transmission, improving accelerator utilization, and shortening the time from design to deployment for complex systems. Orion further integrates these capabilities into a rack-level reference architecture that can select different GPUs and CPUs, enabling the company to extend from single-chip sales to boards, software, and design licensing. The nearly endless incremental demand brought by the global AI agent frenzy comes from longer contexts, more frequent inference, and more complex multi-accelerator collaboration. Long contexts and concurrent requests increase key-value cache (KV Cache) capacity requirements, while multi-card inference increases communication and synchronization pressure; if data supply cannot keep up, expensive GPUs cannot be fully converted into token output. Astera Labs reduces communication hops through switching networks with greater connectivity scale, carries part of the KV cache through connected memory, and establishes a hierarchy between limited HBM and larger-capacity memory. Orion promotes this capability to new cloud service providers and the enterprise inference market; management gave the example that a fully configured single rack with 80 GPUs can include about $100,000 of Astera Labs product, software, and licensing value. Its investment logic is therefore linked both to continued large-scale expansion of AI compute demand and to improving returns on existing compute investment. In terms of customers and partners, Amazon.com, Inc.'s AWS cloud computing business is a long-term major customer publicly confirmed by the company; NVIDIA Corporation, AMD, Arm, and Intel Corporation are important platform ecosystem partners; and system partners include Gigabyte, Foxconn's Ingrasys, Inventec, Quanta's QCT, and Wiwynn. Cooperation with AI chip leader NVIDIA Corporation (NVIDIA) has covered PCIe connectivity across multiple generations of platforms and extended into the NVLink Fusion ecosystem; these relationships support ALAB in serving different processor architectures. In PCIe, Astera Labs can be described as the de facto connectivity chip supplier: Aries retimers have already been deployed in volume in NVIDIA Corporation Hopper, Blackwell, and related server platforms, and Scorpio P-series switch chips also participate in MGX platform integration and have completed PCIe 6 interoperability validation with NVIDIA Corporation GPU compute clusters. In NVLink Fusion, Astera Labs is an officially announced partner of NVIDIA Corporation, developing custom interconnect solutions supporting NVLink connectivity with NVIDIA Corporation and hyperscale cloud customers to help customers connect self-developed accelerators into rack-level computing architectures; the company also disclosed that related custom solutions have already achieved customer Design Wins. On earnings expectations, Morgan Stanley sharply raised Astera Labs' 20262028 revenue forecasts to approximately $2.030 billion, $3.401 billion, and $4.412 billion, corresponding to year-over-year growth of 138.1%, 67.5%, and 29.7%; adjusted earnings per share forecasts are $4.37, $7.00, and $8.64, respectively, raised by about 9.5%, 20.1%, and 24.7%. In Morgan Stanley's latest quantitative model, adjusted gross margin falls from 73.2% in 2026 to 70.5% in 2028, while adjusted operating margin remains around 42.5%43.4%, reflecting revenue expansion diluting R&D, sales, and administrative expenses. Operating cash flow forecasts also rise from about $598 million in 2026 to $1.815 billion in 2028. These adjustments also highlight that what Morgan Stanley's research report endorses is the total profit and cash generation capability brought by product portfolio expansion, rather than requiring gross margin to remain at early high levels. On valuation, Morgan Stanley's $389 base-case target price uses 16.5x 2028 enterprise value/sales (EV/Sales), supported by an approximately 47% revenue compound growth rate from 2026 to 2028 that justifies a premium to peers. Earnings forecasts were significantly raised, yet the target price was only raised 1.8%, which to some extent indicates that the Morgan Stanley analyst team is more positive on the business outlook while still constraining valuation expansion. On bull-case target prices, Morgan Stanley's most optimistic scenario target price is $566, corresponding to stronger new product ramp and maintenance of existing share; the bear-case target price is only $252, corresponding to loss of Aries share and Scorpio and new business contributions below expectations, using 20.6x and 12.4x 2028 revenue valuation, respectively. Based on the same closing price benchmark, the bull and bear target prices imply about 153.5% and 12.9% potential upside. Wall Street after earnings: bullish on Astera Labs' growth prospects, but under heavy AI financing cost pressure, inclined to pay less for growth Rising financing pressure for AI compute infrastructure will affect valuations across the AI compute infrastructure supply chain through the cost of capital for customer infrastructure projects (especially highly leveraged, high-load new cloud players such as CoreWeave), deployment pace, and investors' required rate of return, ultimately leading to continued contraction in valuation multiples for AI compute and technology-themed stocks. This is why, after the earnings release, Wall Street analysts' main disagreement on Astera Labs centers on how much valuation premium they are willing to pay for high growth. Some analysts remain bullish on the company's growth prospectsthat is, bullish on the growth space brought by Scorpio ramp and memory, optical interconnect, and rack-level solutionsbut under heavy AI financing cost pressure they place more emphasis on how quickly these opportunities convert into profit and cash flow, and therefore demand to buy this growth at a lower valuation premium. Based on the November 4 closing price of $223.27 and the disclosed static share count of approximately 173.49 million shares, RBC maintained "Outperform" but cut its target price from $500 to $400, still implying 79.2% potential upside; Stifel maintained "Buy" and cut its target price from $460 to $370, implying 65.7% potential upside; Jefferies also maintained "Buy" and cut its target price from $450 to $350, implying 56.8% potential upside. Jefferies analyst Blayne Curtis believes that the Scorpio X ramp drove another strong quarter for the company and therefore remains positive on its prospects. These latest adjustments show that institutions remain positive on operating growth but are more restrained in valuation pricing. Northland upgraded the rating from "Market Perform" to "Outperform" while adjusting its target price from $350 to $315, implying 41.1% potential upside; the institution believes that Scorpio, which has become the largest product family, will continue to carry the main growth task in 2027. Evercore ISI maintained "Outperform" and cut its target price from $452 to $311, implying 39.3% potential upside and a market value of about $54.0 billion. Needham's adjustment especially explains the divergence between the results and the stock price: the institution raised earnings forecasts but, due to contraction in AI sector valuation multiples, cut its target price from $425 to $300 while maintaining "Buy," implying 34.4% potential upside. These adjustments also mean that analysts at these top Wall Street institutions recognize revenue and profit growth, but that does not mean they are willing to continue granting the previous high valuation multiples. More cautious institutions focus more on how much sustained profit growth can ultimately be converted into. UBS Group AG (UBS) maintained a "Neutral" rating and cut its target price from $300 to $290, implying 29.9% potential upside; although it raised 20262028 forecasts, it remains focused on the impact of AWS's future architecture choices on ALAB's switch chip share and value per accelerator. Susquehanna maintained "Neutral" and cut its target price from $310 to $258, implying 15.6% potential upside, believing that Orion can expand long-term growth space but that R&D and operating investment need to be converted into revenue and profit accretion. TD Cowen maintained "Hold" and cut its target price from $300 to $225, implying about 0.8% potential upside. These views collectively point to the key pricing factors for the next stage: new Scorpio customers entering mass production, Leo and optical interconnect generating revenue, and Orion expanding system value and contributing profit will determine whether earnings upgrades can further stock price recovery.