Leading financial institutions have released their most significant analyst views on the artificial intelligence (AI) sector this week. Morgan Stanley has reinstated NVIDIA (NASDAQ:NVDA) as its top semiconductor stock pick. The firm believes the stock's recent stagnation fails to reflect improvements in its underlying business fundamentals. An analyst noted in a Monday report that NVIDIA's share price has remained flat over the past two quarters while its business has continued to strengthen, attributing this divergence to concerns over the sustainability of current growth. The analyst suggested the current valuation offers an attractive entry point, highlighting that NVIDIA trades at approximately 18 times estimated 2027 earnings per share, which he described as a surprisingly favorable opportunity. This marks a shift from the firm's earlier stance, which favored memory companies like SanDisk and Micron Technology anticipating stronger AI-related profit leverage in that segment. That strategy proved highly successful, with memory stocks surging 300% to 900% since the view was published. Over the same period, NVIDIA's stock has been largely flat, despite earnings estimates rising about 38% over the past six months. The analyst indicated that investor concerns about demand sustainability and potential market share pressures have weighed on sentiment. However, he believes both issues are beginning to improve, with upcoming product refreshes expected to reinforce NVIDIA's leading roadmap and help address competitive worries. Regarding demand, the firm's supply chain checks indicate continued strong spending commitments from hyperscale customers. The analyst wrote that there are currently no signs the current investment cycle is ending, noting that some customers are prepaying to secure capacity through 2028. He added that easing supply constraints for AI processors could ultimately benefit NVIDIA if other components like memory, storage, optics, and power become primary bottlenecks. The analyst also stated that investor focus on short-term supply metrics may obscure stronger underlying demand trends. In his view, any improvement in GPU delivery lead times could support market share gains similar to those seen in previous cycles. NVIDIA currently holds about 85% of AI processor revenue share, with ASICs slightly above 10% and AMD slightly below 5%. While acknowledging competition may gradually intensify as large customers pursue more architecturally flexible strategies, the analyst said many deployments still favor NVIDIA. He added that major ASIC and AMD customers are projected to increase their NVIDIA-related spending by over 80% in 2026.
In a separate report, Morgan Stanley suggested that Samsung Electronics (KS:005930) stock appears increasingly attractive after recent declines, viewing this pullback as a potential entry opportunity ahead of changes in AI memory architecture. The stock fell over 13% this week, underperforming the broader KOSPI index, which declined about 10%. An analyst noted the adjustment comes as the AI memory ecosystem begins shifting toward more complex hybrid structures as chips become more specialized. While High Bandwidth Memory continues to dominate many AI workloads, the analyst said new use cases are emerging for other memory types. The analyst indicated that Static Random-Access Memory is carving out a niche for workloads where latency is more critical than throughput density. The analyst anticipates NVIDIA will unveil a new inference-specific chip at its upcoming GPU Technology Conference. According to the analyst, this chip might use a Language Processing Unit design built around substantial on-chip SRAM. He described this architecture as designed for the sequential speed of inference, potentially attracting customers willing to pay for speed. The analyst believes the industry should not view this evolution as direct competition between memory technologies. Instead, the market is moving toward a tiered system where different memory types play complementary roles. He added that both technologies will remain layered, using SRAM for hot path execution and HBM for scalable memory capacity. The analyst also stated that LPU-based designs could avoid current supply chain bottlenecks in HBM and CoWoS packaging, which constrain parts of the AI hardware ecosystem. Despite uncertainty about how the AI memory market will ultimately evolve, Morgan Stanley reiterated Samsung Electronics as its top pick. The analyst cited the company's progress in HBM4 certification, SRAM capabilities, flexibility in fab operations, and potential support from the broader commodity cycle. The analyst wrote that historically, such adjustments have provided good buying opportunities, adding that earnings expectations still have significant room for recovery.
Benchmark and Bank of America analysts upgraded Marvell Technology (NASDAQ:MRVL) to Buy ratings after the chipmaker's latest results and guidance reinforced expectations for accelerated growth linked to AI demand. Marvell projected fiscal year 2028 revenue approaching $15 billion, highlighting strong demand for its custom chips and interconnect solutions for AI data centers. This forecast significantly exceeded Wall Street expectations of $12.92 billion, driving the stock up over 18% on Friday. The company also raised its fiscal year 2027 outlook, now expecting revenue to grow over 30% year-over-year to nearly $11 billion, up from a prior forecast of approximately $10 billion. Bank of America analysts stated the company's latest earnings call strengthened confidence in its exposure to key AI infrastructure segments. According to the analyst team, the update highlighted Marvell's solid leverage to AI optical connectivity, potential success with an upcoming Microsoft custom chip project, and signs the company is navigating the annual Amazon XPU transition effectively. A Benchmark analyst also upgraded the stock to Buy with a $130 price target, citing multiple growth drivers in Marvell's data center and connectivity businesses. The analyst said the company benefits from broad-based accelerating demand trends and improved forward visibility, with its guidance suggesting revenue and earnings could be substantially above current market consensus in the coming years. He added that Marvell's custom silicon segment is expected to continue expanding, supported by ongoing collaborations with hyperscale customers including Amazon and Microsoft. Despite the recent stock price increase, the analyst stated Marvell's valuation remains particularly attractive compared to broader AI semiconductor peers.
This week, Astera Labs (NASDAQ:ALAB) received a bullish initiation from Loop Capital, with the investment bank assigning the semiconductor company a Buy rating and a $250 price target, citing its central role in the AI super-cycle. A Loop Capital analyst stated the company represents one of the clearest ways for investors to gain AI infrastructure exposure beyond the largest chipmakers. The analyst wrote that the firm initiates coverage with a Buy rating and $250 target price because it views ALAB as the company that best represents a diversified AI silicon pure-play, excluding NVIDIA. The analyst believes Astera Labs is well-positioned within the broad AI processor ecosystem, including GPUs and alternative accelerators for large-scale data center deployments. He stated the company's products are designed to support the performance and connectivity requirements of next-generation AI infrastructure. The analyst wrote that Astera Labs has opportunities across essentially all generative AI silicon types through technology aimed at solving performance bottlenecks in AI servers and clusters. The analyst believes the growing scale and complexity of AI systems will make the company's solutions increasingly critical over time. He stated that as servers and clusters become larger and more complex, ALAB becomes more critical and valuable. The analyst also noted the expanding range of AI accelerators coming to market beyond those developed by NVIDIA, suggesting this trend could expand Astera Labs' potential addressable market. Long-term, the analyst believes the company will build a stronger competitive position as its technology becomes embedded in the AI infrastructure stack. He highlighted Astera Labs' predictive software and management platform, COSMOS, which he said could become an industry standard and a true performance amplifier. He added that widespread adoption could create moat-like stickiness with customers as AI systems continue to scale.
Industrial software stocks have been caught in a broader market sell-off driven by fears of AI disruption in the software sector, but Barclays stated this reaction is overdone and misinterprets how enterprise software businesses create value. A bank analyst argued that the mainstream narrative—that AI will disrupt the economics of Software-as-a-Service companies—overlooks the importance of the services, operational reliability, and industry expertise customers actually pay for. He stated the SaaS AI bear thesis is overdone, adding that AI should represent an incremental opportunity rather than a threat for industrial software companies. He added that the AI-driven sell-off has unfairly punished industrial software. The analyst said the view that AI will commoditize enterprise software is based on a flawed assumption that customers primarily pay for code. He wrote in a research report that while AI commoditizes code generation, it highlights the economic value of service in SaaS. If code is a commodity, then the value lies in deep domain expertise, reliable service, and hybrid architectures. The analyst added that code is an input, not the product or output, noting that successful SaaS companies moved beyond selling code years ago and now compete primarily on service and sector expertise. The market reaction implies coding constitutes a large portion of enterprise software value, which the analyst believes is inaccurate. In mature SaaS companies, coding likely represents only a small fraction of total expenditure. According to Barclays estimates, coding costs might account for only about 4% to 8% of revenue, suggesting automated code generation would have a limited direct financial impact. Nevertheless, valuations for industrial software companies have declined significantly, with the bank noting that covered companies' valuation multiples have fallen approximately 50% over the past six months. The analyst also pointed out that the economics of running AI systems are another factor that could benefit established software providers. While generative AI may lower development costs, operating AI software at scale could be expensive due to inference costs related to GPUs, energy, and infrastructure. The analyst stated that even if software can be generated cheaply or for free, it cannot be operated cheaply at scale, arguing this creates a hard economic price floor favoring companies that can manage these costs effectively. He added that industrial software valuations have now retreated to pandemic-era levels and are at historic lows relative to the S&P 500, despite the sector offering free cash flow yields higher than many industrial tech hardware peers. The analyst believes several potential catalysts could drive renewed inflows into the sector, including investor fatigue with AI hardware trades, a potential reversal in semiconductor performance relative to software, and the sector's high short interest.
Comments