Summary
Analysts made a series of notable moves across AI-related equities this week, shifting emphasis among GPUs, memory, connectivity, and industrial software. The changes reflect a common thread: brokerages are re-evaluating where durable demand and scalable economics will accrue as hyperscale customers expand deployments and as component-level bottlenecks evolve. Morgan Stanley returned Nvidia to the top of its semiconductor roster, citing improving business trends and a valuation it describes as an attractive entry point. The bank also recommended buying Samsung on the recent correction, arguing that structural shifts in AI memory use cases will favor layered memory architectures where Samsung has relevant exposure. Marvell drew upgrades after stronger-than-expected guidance tied to custom chips and optical interconnects, Loop Capital initiated Astera Labs as a way to play broader AI infrastructure diversification, and Barclays defended industrial software names against what it views as an exaggerated AI bear case.
Morgan Stanley restores Nvidia as premier chip pick
Morgan Stanley moved to restore Nvidia (NASDAQ:NVDA) as its favored semiconductor stock after observing what it calls a disconnect between Nvidia’s improving business fundamentals and the company’s share-price behavior. The bank’s note notes that Nvidia’s stock has been essentially flat for two quarters even as underlying earnings expectations have risen materially. Analyst Joseph Moore emphasized that investor concerns around the durability of current growth and potential market-share pressure have kept sentiment muted, despite the operational indicators pointing higher.
Moore highlighted valuation as central to the thesis: Nvidia is trading at roughly 18 times what Morgan Stanley expects for 2027 earnings per share, a multiple the analyst described as "a surprisingly good entry point." The firm’s supply-chain checks indicate steady spending commitments from hyperscale customers, including indications that some clients are prepaying to secure capacity through 2028. Those checks underpin Morgan Stanley’s view that the current investment cycle remains intact, countering narratives that the AI hardware build-out is nearing exhaustion.
Competition remains a focal point, but Morgan Stanley’s note frames the near-term competitive environment as one where Nvidia still dominates. The firm cites Nvidia’s roughly 85% share of AI processor revenue, with ASICs at just above 10% and AMD slightly below 5%. While the analyst acknowledges that larger customers may pursue more architecturally flexible strategies over time, he argues that many deployments continue to favor Nvidia’s platform, and that upcoming product updates should help reinforce Nvidia’s leadership roadmap.
Moore also argued that easing supply constraints in AI processors could paradoxically help Nvidia by shifting bottlenecks to other components such as memory, storage, optics, and power. That shift, he posits, could limit the ability of some competitors to convert short lead-time improvements into sustained share gains.
Importantly, Morgan Stanley contrasted its prior stance favoring memory names with the current call. The firm had earlier overweighted memory companies such as SanDisk and Micron on the premise that AI-related earnings leverage would be stronger in the memory segment. That earlier positioning produced substantial gains for memory equities, with Morgan Stanley noting increases between 300% and 900% for those names. Over the same timeframe, Nvidia’s share price was largely unchanged even as the company’s earnings expectations rose about 38% over the previous six months. The reinstatement reflects a recalibration: momentum in Nvidia’s fundamentals, combined with a valuation that Moore finds compelling, justified moving the stock back to the top of Morgan Stanley’s semiconductor list.
Morgan Stanley says Samsung looks attractive on pullback
In a separate note, Morgan Stanley recommended buying Samsung Electronics (KS:005930) after the stock declined more than 13% in the week and underperformed the broader KOSPI index, which was down about 10% over the same period. Analyst Shawn Kim framed the correction as a potential buying opportunity ahead of what he expects will be an evolution in AI memory architecture away from single-technology dominance toward a more heterogeneous, layered system.
Kim described the memory landscape as shifting toward a "complex, hybrid structure" where high-bandwidth memory (HBM) remains central for many workloads but where other memory types are finding specific niches. He singled out Static Random Access Memory (SRAM) as carving out use cases where latency trumps throughput density. The note ties this development to anticipated product announcements from Nvidia, which analysts expect will unveil an inference-focused chip at its GPU Technology Conference. Kim suggested that Nvidia’s rumored Language Processing Unit (LPU) design could rely on substantial on-chip SRAM to optimize sequential inference speed, a design trade-off that would appeal to customers prioritizing latency and willing to pay for performance.
Kim emphasized that the market should not interpret this architectural shift as a zero-sum contest between memory technologies. Instead, he sees a layered approach where SRAM provides hot-path execution and HBM supplies scalable capacity. That hierarchy, he argues, could broaden demand across multiple types of memory rather than concentrating it entirely in HBM.
Another practical implication in Kim’s view is that LPU-based designs might sidestep current supply bottlenecks in HBM and CoWoS packaging, constraints that have affected parts of the AI hardware supply chain. Morgan Stanley reiterated Samsung as its top pick in this context, citing the company’s progress in HBM4 qualification, its SRAM capabilities, the flexibility of its foundry operations, and the potential support from a broader commodity cycle. Kim concluded that, historically, similar corrections have created attractive entry points and that earnings expectations for Samsung still have substantial room to recover.
Marvell’s guidance prompts upgrades
Analysts at Benchmark and Bank of America upgraded Marvell Technology (NASDAQ:MRVL) to Buy following the company’s recent results and forward guidance, which the analysts interpret as evidence of accelerating growth tied to AI demand. Marvell provided fiscal 2028 revenue guidance of nearly $15 billion, a figure well ahead of a Wall Street consensus cited in the note of $12.92 billion. The announcement propelled Marvell’s stock more than 18% higher on the Friday following the update.
The company also raised its fiscal 2027 outlook, now expecting revenue growth exceeding 30% year over year to nearly $11 billion, compared with an earlier projection of roughly $10 billion. Bank of America analysts, led by Vivek Arya, emphasized that Marvell’s update reinforced the company’s exposure to critical AI infrastructure segments, including optical connectivity for data centers and the potential opportunity from an upcoming Microsoft custom chip program. They also noted signs that Marvell is "turning the corner on Amazon XPU transition year," language signaling improved traction in engagements with hyperscalers.
Benchmark’s Cody Acree also moved the stock to Buy and assigned a $130 price target. Acree pointed to multiple growth drivers across Marvell’s data-center and connectivity businesses, arguing the company benefits from broadly accelerating demand trends and improved visibility. He emphasized that the guidance suggests revenue and earnings could materially outpace current consensus over the coming years. Acree further highlighted Marvell’s custom silicon segment as a continuing growth vector, with ongoing engagements with major hyperscale customers such as Amazon and Microsoft expected to expand that business.
Despite the share-price gain following the guidance, Acree described Marvell’s valuation as still attractive relative to the broader AI-focused semiconductor peer group, implying that the company’s prospective growth trajectory is not yet fully reflected in its multiples.
Loop Capital starts coverage of Astera Labs on AI scale-up thesis
Loop Capital began coverage of Astera Labs (NASDAQ:ALAB) with a Buy rating and a $250 price target, identifying the company as a way to gain exposure to AI infrastructure beyond the largest chip vendors. Analyst Ananda Baruah characterized Astera Labs as a "diversified AI silicon ‘pure play' (outside of NVDA)," noting the company’s positioning across GPUs and alternative accelerators used in large-scale data-center deployments.
Baruah emphasized that Astera Labs’ product set targets performance and connectivity bottlenecks across next-generation AI infrastructure. The firm’s solutions, he argued, have applicability across essentially all Gen AI silicon flavors - including GPUs, Tranium, TPU and other XPUs - by addressing interconnect and performance issues at the server and cluster level. As servers and clusters scale in size and complexity, Baruah wrote, the relative importance of robust connectivity and management software increases, enhancing Astera Labs’ potential role within the stack.
Loop Capital also called out Astera Labs’ COSMOS platform - a predictive software and management layer - as a possible point of long-term differentiation. The analyst suggested that if COSMOS achieves broader adoption, it could create "moat-like stickiness" by becoming embedded in customers’ infrastructure management and performance workflows, amplifying the company’s value as AI deployments become larger and more sophisticated.
Barclays: AI fears have unfairly punished industrial software
Barclays offered a counter-narrative to the market’s AI-driven sell-off in industrial software names, arguing the negative reaction has been exaggerated and misconstrues how enterprise software firms create value. Analyst Guy Hardwick said the purported threat from AI - the idea that AI will commoditize and thus sharply reduce the economics of software-as-a-service businesses - neglects the importance of services, operational reliability, and domain expertise that customers continue to pay for.
Hardwick argued the "SaaS AI bear case is overdone," framing AI more as incremental opportunity than existential threat for industrial software providers. He noted that customers pay for more than code generation: they pay for deep domain knowledge, consistent service levels, deterministic-plus-probabilistic architectures, and operational reliability. In this view, automated code generation may commoditize portions of the development process, but it does not obviate the value of the service and expertise that deliver production-grade outcomes.
Barclays estimated that coding costs likely represent a relatively small share of revenue within mature SaaS companies - roughly 4% to 8% by the bank’s calculation - implying that automated code generation would have a limited direct financial impact on revenue. Nevertheless, valuations across industrial software names have materially declined, with Barclays noting that multiples for covered names have fallen about 50% over the past six months and that valuations now sit at COVID-era levels relative to the S&P 500. At the same time, the sector is offering stronger free-cash-flow yields than many industrial tech hardware peers, according to the note.
Hardwick also flagged the economics of operating AI systems as a factor that could protect established software providers. Even if generative AI lowers development costs, running AI at scale requires capital and recurring operating expenditure - inference costs associated with GPUs, energy, and infrastructure - which create a "hard economic price floor" for productive software deployments. That dynamic, he argued, should benefit companies capable of managing those costs effectively.
The Barclays analyst identified several potential catalysts that could prompt a sector rotation back into industrial software. Those include investor fatigue with AI hardware trades, a reversal in semiconductor outperformance versus software names, and elevated short interest across the sector. Each of these would potentially shift relative investor appetite toward software equities that Hardwick believes remain undervalued on a fundamental basis.
Takeaways for markets and sectors
The string of analyst moves highlights several recurring themes relevant to investors focused on AI-related cycles. First, the drag from short-term share-price performance can create entry opportunities when underlying indicators point to sustained demand and favorable structural positioning. Morgan Stanley’s reinstatement of Nvidia and its buy recommendation on Samsung are both rooted in that dynamic: improving fundamentals and unique positioning relative to expected shifts in the memory and compute stacks.
Second, the breadth of the AI investment opportunity continues to expand beyond raw processors. Upgrades to Marvell and the initiation on Astera Labs emphasize the importance of connectivity, custom silicon, and system-level interconnects as sources of durable demand. Hyperscale customers’ engagements with vendors for custom chips and interconnects are shaping revenue trajectories for suppliers that can capture those design wins.
Third, the debates over who captures value from AI - core chip vendors, memory suppliers, connectivity specialists, or software operators - remain unresolved in detail but are narrowing in practice as customers scale deployments. Brokers are responding by repositioning recommendations to reflect where they see leverage and defensible economics emerging as hyperscale capex rolls out.
Key points
- Morgan Stanley reinstated Nvidia as its top semiconductor pick, citing improving fundamentals, steady hyperscaler spending, and a valuation near 18x expected 2027 EPS that the firm views as attractive; Nvidia retains roughly 85% of AI processor revenue share while ASICs and AMD occupy smaller portions of the market.
- Morgan Stanley also flagged Samsung as a buy after a pullback of more than 13% in the week, arguing that shifts toward layered AI memory architectures - combining HBM for capacity and SRAM for latency-sensitive hot paths - play to Samsung’s strengths across HBM4 qualification, SRAM capabilities, and flexible foundry operations.
- Market participants beyond chipmakers are gaining attention: Benchmark and Bank of America upgraded Marvell following stronger-than-expected guidance tied to custom chips and optical connectivity, while Loop Capital initiated Astera Labs with a Buy as a diversified AI infrastructure exposure; Barclays countered negative sentiment on industrial software, arguing AI fears have been overstated.
Risks and uncertainties
- Demand durability concerns - Several analysts note investor anxiety about the sustainability of current AI spending trends; if hyperscale customers materially curtail commitments, equities exposed to AI infrastructure could face pricing pressure. This risk primarily affects semiconductors, memory, and interconnect suppliers.
- Competitive and architectural shifts - While some analysts view upcoming product updates and architectural choices as reinforcing incumbent positions, competition from alternative accelerators and custom ASIC approaches could alter share dynamics over time. This uncertainty impacts GPU makers, ASIC vendors, and companies supplying complementary components such as memory and optics.
- Supply-chain and component bottlenecks - Changes in where bottlenecks sit (for example, from AI processors to memory, packaging, or optics) could influence which suppliers benefit from the ongoing investment cycle. Companies reliant on constrained components or specialized packaging face execution risk tied to supply-chain dynamics.
Conclusion
This week’s analyst activity underlines how brokerages are refining their views as AI deployments scale and as the supply chain adjusts. Reinstating Nvidia at the top of a semiconductor list, recommending Samsung on weakness, upgrading Marvell, initiating coverage of Astera Labs, and defending industrial software valuations together sketch a market narrative that value capture from AI is distributed across chips, memory, interconnects, and software services. Each call rests on specific assumptions reflected in company guidance, supply-chain checks, and product road maps; those assumptions, if borne out, point to differentiated winners across the AI stack. Conversely, shifts in demand durability, competitive responses, or supply constraints remain key uncertainties that market participants will continue to monitor closely.