Advertisements
The dynamics of the semiconductor and artificial intelligence industries are continuously evolving, and at the forefront of this evolution is Nvidia, a company that has established itself as a major player in the world of AI and graphics processing units (GPUs). Recently, a report from Morgan Stanley's team of North American analysts has reaffirmed Nvidia's position as their top stock pick, emphasizing that recent sell-offs triggered by the so-called DeepSeek event represent a rare buying opportunity for investorsThe analysts, headed by Joseph Moore, set a price target of $152 for Nvidia's stock, anticipating an increase of approximately 21.8% based on recent closing prices.
This positive outlook from Morgan Stanley comes in light of the market's reaction, which saw Nvidia's stock rise by more than 3% as of Thursday's close, signaling robust investor confidence in the company's business prospects
Advertisements
Despite a general deterioration in investor sentiment regarding potential long-term risks associated with certain technological developments, Moore insists that Nvidia's short-term business remains strongThe increasing visibility around the supply of Blackwell chips and clear consumer demand underscore the analysts’ belief that Nvidia remains a favored investment among semiconductors.
While acknowledging the hurdles posed by the DeepSeek incident—particularly regarding export controls and long-term investment challenges—Morgan Stanley's analysts view this event as merely one of many significant advancements in the artificial intelligence landscapeNvidia's CEO has previously indicated that the company has enhanced processor-level AI performance by a staggering one million times over the past decade and aims to achieve similar breakthroughs in the next ten years.
Although the DeepSeek situation has led to some short-term market fluctuations, the analysts assert that such obstacles are not sufficient to undermine Nvidia’s robust performance
Advertisements
They anticipate that some market adjustments are indeed to be expectedThe analysts also maintained a highly optimistic view of Nvidia's balanced growth prospects leading up to 2025, highlighting four main supporting factors.
Firstly, there is growing confidence in the prospects of Hopper and Blackwell chipsDespite currently being in a transitional phase, confidence in these chips is steadily increasing, and analysts expect positive feedback from the market by the end of the current quarterFor the Hopper chip, while demand has shown signs of weakening recently, industry surveys indicate that recovery is likelyThe risks associated with export controls may play a role here, as regions affected by these controls are adopting preventive measures that could ultimately drive demand upwards.
For the Blackwell chip, analysts are closely monitoring Nvidia's updates regarding its "unprecedented complexity," believing that the challenges faced are being effectively addressed, including the final shape of the GB200 chip
Advertisements
While some limitations persist, there is a strengthening confidence in the demand for Blackwell componentsEarlier temporary measures, such as the rushed release of the GB300 to mitigate adjustments to the GB200, are no longer deemed necessaryDemand across various forms of Blackwell technology appears strong, with an initial focus potentially leaner toward non-rack scale solutions, indicating sustained interest from analysts.
The analysts are optimistic that Nvidia will deliver performances in line with expectations during their financial guidance in AprilHowever, it's expected that the management’s outlook may pivot significantly, reflecting a solid demand for Blackwell components and enhanced market robustness overall.
The second factor relates to the continuing momentum of building large training clustersDespite pressures on investor sentiment toward these expansive projects, analysts note signs indicating that large cluster construction remains underway
Commentary on capital expenditures from Nvidia's major clients reaffirms this investment trajectory, highlighting that supply-demand mismatches are ongoing in the near termNvidia’s cloud customers endorse a clear revenue algorithm: the more GPUs they acquire, the higher their earnings soarEven clients who have not yet turned a profit remain committed to adopting cutting-edge technologyMany architects of the largest AGI clusters continuously reaffirm their commitment to expanding their training clusters, with no indication that the DeepSeek challenge has hindered this momentum.
Executives from Oracle, Microsoft, Meta, and Google recently commented positively on AI spending during their earnings calls, a trend that Morgan Stanley suggests is a testament to the robustness of capital expenditure growth in these fieldsTheir hardware research team estimates that the top ten cloud vendors will see a 24% year-over-year growth in capital expenditures by 2025, a significant revision from the previous 21% expectation.
The third aspect focuses on the promising growth in the inference market, where Nvidia’s position remains firmly established
While the inference market exhibits a deflationary trend in comparison to the training market, analysts argue that there is a strong preference for high-performance solutions within this domainEven Nvidia's lower-priced inference solutions are gradually giving way to the comprehensive training systems based on Hopper and the new Blackwell architectureThe analysts are confident that Nvidia, being the largest beneficiary of long-term inference workloads, continues to strengthen its foothold within this segment.
In recent years, analysts postulated that cheaper inference-specific products, such as the T4 and L40, would gain greater popularityHowever, as workloads increasingly grow in complexity, high-end GPUs remain viewed as the primary source for inference computing needsThe extension of computation test times has elevated the computational requirements for inference tasks, favoring high-performance solutions, and Nvidia is well-positioned to meet this demand.
Using ML performance benchmarks, the B200 provides a token throughput on the LLama2 70b system that is 2.5 times higher than that of the H200 per chip, with the expectation of even greater multi-chip performance due to the transition to Blackwell's fifth-generation NV-link, which boasts double the bandwidth of the previous generation
This translates into at least a twofold performance enhancement per dollar spent, a projection that applies to products that have only recently begun shippingAnalysts believe this sets a remarkably high comparative standard for ASICs, and anticipate the upcoming release of Nvidia's Rubin chip, which may further elevate the benchmarks that ASICs will strive to meet.
The fourth and final argument centers on a potential shift in stock market preferences, with analysts predicting a reversal favoring Nvidia's GPUs in the latter half of 2025. Though the markets presently lean towards ASICs over GPUs, Morgan Stanley expects a resurgence in GPU revenues as the second half of the year progressesWhile both markets hold significant long-term promise, analysts predict a greater shift in market sentiment towards Nvidia in the latter half of the year.
They note that the recent success of ASICs following the emergence of DeepSeek can be partially attributed to a heightened focus on service costs, viewing ASICs as optimal low-cost inference platforms for many Nvidia customers
The analysts endorse this viewpoint to some extent, as optimized products are indeed better suited for specific workloadsHowever, from their observations, the analysts maintain that the real shift in cost curves in 2025 can be attributed to Blackwell rather than ASICs.
Moreover, some speculate that ASIC projects exhibit a longer-term investment appeal relative to Nvidia’s offeringsThe analysts concede this perception holds some truth, particularly because there are commitments to enhance ASIC capabilities in less penetrated areasNonetheless, should demand in training slow down, capital commitments tied to training and inference may swayBroadcom has indicated that its prospective market size of $60 to $90 billion is largely training-related.
As analysts argue that Nvidia's existing advantages in training will serve the company well if training market growth decelerates, they underscore that upon completing training runs, the GPUs traditionally allocated for inference become available
Some speculate that, without initial purchases, entry barriers could become lower, presenting a long-term riskHowever, in the short run, Nvidia’s GPUs may replace inference ASIC projects due to the company's flexibility, enabling higher degrees of unification across cloud, on-premises, inference, and training workloads.
Nvidia is also enhancing its software stack and providing these updates to customers for free, a move that has already proven to significantly boost performance in the pastEven if overall AI capital expenditures wane, the analysts are not surprised by Nvidia’s market share gains—especially when compared to first, second, or third-generation ASIC productsIn sum, the landscape surrounding Nvidia is complex, filled with both challenges and immense opportunities, pushing the company into a solid position as it navigates this transformative era of technology.