Summary
DeepSeek's cost-efficiency claims initially alarmed Nvidia Corporation investors, but comprehensive analysis shows the company's dominance in AI chips will likely be challenging to disrupt.
Despite long-term risks, Nvidia's robust ecosystem, innovation, and financial strength position it well to maintain leadership in the AI chip market.
If management's guidance for the first quarter of FY 2026 beats analysts' estimates, much of the negativity surrounding the stock due to DeepSeek will likely dissipate.
I maintain my NVDA buy rating for aggressive growth investors.
BING-JHEN HONG
Nvidia Corporation's stock has stalled over the last several months due to valuation concerns, making it sensitive to the market selling off the stock if the company's high revenue growth and profitability come under an actual or perceived threat. An example of a perceived threat to revenue growth was when DeepSeek's (DEEPSEEK) news of training an AI model at lower costs than OpenAI's model hit the market, opening up the possibility that demand could fall for Nvidia's powerful AI chip. The market's immediate reaction was to sell off the stock. It dropped around 17% on January 27. While it has recovered some since then, the DeepSeek news continues to weigh down the stock.
Data by YCharts
Before the DeepSeek news hit the market, I upgraded NVDA stock to a Buy rating for only aggressive growth investors on December 10, 2024. Although the stock's high valuation and risks are genuine, I reasoned that the market may undervalue the company's estimated +50% revenue and earnings growth rate in its fiscal year (“FY”) 2026. However, this buy rating is only appropriate for investors willing to accept the stock's significant potential downside if the company misses quarterly results expectations or the long-term technological threat of customers reducing their reliance on Nvidia's increasingly powerful AI chips becomes more immediate.
This article will discuss the DeepSeek threat and why it's not a concern for Nvidia over the near and medium term. It will examine its third-quarter earnings call and what to expect when the company reports fourth-quarter results. The article will also review Nvidia's long-term risks and valuation, and why it is a buy for aggressive growth investors willing to speculate on the AI infrastructure market.
The DeepSeek Threat
It is difficult to question Nvidia's short- and medium-term dominance because the AI chip market is supply constrained. However, once DeepSeek came onto the scene, claiming that it could train AI models at substantially lower costs, it exposed a latent fear among some Nvidia bulls that technological advances in AI models could make increasing the raw computing power in chips less necessary over the long term. Customers could shift to chips that are powerful enough but consume less energy, which may neuter some of Nvidia's head start in creating increasingly more powerful AI chips.
Before I go further into this discussion, it's essential to understand that the conversation around DeepSeek can be confusing for a non-technical person. I had to revise this discussion about DeepSeek several times because I discovered that some articles about the topic have some misconceptions. The first thing to understand is that there are three steps in training an AI model:
Pretraining (before the model use)
Post-training (after the models use)
Test-time scaling (during the inferencing use of the model).
I will explain what pretraining an AI model entails later in this article, but it is likely the most expensive step. Post-training is like fine-tuning the AI model for specific topics or use cases, and is less costly because training datasets are smaller and training takes less time than pretraining. Test-time scaling takes place during inferencing, the time the AI model is generating a response to a prompt. Test-time scaling is the lowest cost out of all three training steps.
A few initial articles discussing how DeepSeek lowered costs focused on pretraining costs. For instance, Bain and Company recently published an article that stated:
What sets DeepSeek apart is the prospect of radical cost efficiency. The company claims to have trained its model for just $6 million using 2,000 Nvidia H800 graphics processing units ("GPUs") vs. the $80 million to $100 million cost of GPT-4 and the 16,000 H100 GPUs required for Meta's LLaMA 3. While the comparisons are far from apples to apples, the possibilities are valuable to understand.
This statement alarmed some people because if DeepSeek's claims are valid, and it is possible to create models the equivalent of GPT-4 or LLaMa 3 at far less cost with much fewer GPUs, that could, in the long term, create far less demand for NVIDIA's highest-end GPUs. It also raises the possibility that companies could train and inference their AI models on more price-performant (good enough performance at a lower cost) AI chips, such as Amazon's (AMZN) Trainum chips. Amazon's Chief Executive Officer Andy Jassy said on the company's recent fourth quarter 2024 earnings call:
Trainium 2 just launched at our AWS Reinvent Conference in December and EC2 instances with these chips are typically 30% to 40% more price performant than other current GPU-powered instances available. That's very compelling at scale. Several technically capable companies like Adobe (ADBE), Databricks, poolside, and Qualcomm (QCOM) have seen impressive results in early testing of Trainium 2. It's also why you're seeing Anthropic build their future frontier models on Trainium 2. We're collaborating with Anthropic to build Project Rainier, a cluster of Trainium 2 ultra servers containing hundreds of thousands of Trainium 2 chips. This cluster is going to be 5 times the number of exaflops as the cluster that Anthropic used to train their current leading set of cloud models.
Nvidia will likely not see any adverse impact from competition in a supply constrained AI chip market. However, over the long term, as supply and demand balance out, companies like Marvell (MRVL), Broadcom (AVGO), and others building increasingly more price-performant chips may make it more challenging for Nvidia to grow at the rates investors require to justify its valuation — or at least that is one bearish theory.
Don't underestimate Nvidia
The need for Nvidia to create increasingly more powerful chips may not diminish for quite a while, if ever. While DeepSeek may have developed an innovative AI model, the company's cost-efficiency claims for pretraining an AI model may be far less than those cited in the Bain news article. Semianalysis published a news article on January 31, 2025, that stated:
DeepSeek's price and efficiencies caused the frenzy this week, with the main headline being the "$6M" dollar figure training cost of DeepSeek V3. This is wrong. This akin to pointing to a specific part of a bill of materials for a product and attributing it as the entire cost. The pretraining cost is a very narrow portion of the total cost...The $6M cost in the paper is attributed to just the GPU cost of the pretraining run, which is only a portion of the total cost of the model. Excluded are important pieces of the puzzle like R&D and TCO of the hardware itself. For reference, Claude 3.5 Sonnet cost $10s of millions to train, and if that was the total cost Anthropic needed, then they would not raise billions from Google and tens of billions from Amazon. It's because they have to experiment, come up with new architectures, gather and clean data, pay employees, and much more.
Pretraining a large language model (“LLM”) involves feeding the model massive datasets consisting of publicly accessible text online, such as books, articles, web pages, code, and video transcripts, to learn language patterns. The costs for pretraining consist of money spent on hardware (GPUs, memory, and networking equipment), data acquisition and cleaning, AI and Data engineer employee expenses, Research and Development (R&D), and operational costs such as server maintenance and utility costs. What the above statement implies is that DeepSeek only included the costs of buying or renting the GPU and possibly the utility costs of using the GPU. It excluded R&D expenses, data acquisition and cleaning costs, labor, and ongoing expenses related to server maintenance, API usage, and infrastructure scaling (adding more computing, storage, and networking resources). In short, the above commentary means the overall costs for training an AI model involve more than just GPU costs, and the $6 million figure may be deceptive.
OpenAI released a full version of its latest AI model, GPT-o1, on December 05, 2024. DeepSeek released its latest R-1 model on January 20, 2025. I haven't found a communication from OpenAI outlining how much it costs to develop GPT-o1, so it is difficult to compare developmental costs between DeepSeek R-1 and OpenAI GPT-o1. However, I wouldn't be surprised if the total cost to create the DeepSeek-R1 model and GpT-o1 are similar.
On the day the DeepSeek news roiled the markets, Seeking Alpha published an article that stated,
“'DeepSeek is an excellent AI advancement and a perfect example of Test Time Scaling,' an Nvidia spokesperson told Seeking Alpha in response to the stock market selloff.”
Test Time Scaling is a new AI paradigm that adds extra computing power at the point the AI model generates text and images or makes a prediction. This additional computing power gives the AI more time to “think” or reason, theoretically leading to better answers. More “thinking time” may require more powerful AI chips, meaning that Test Time Scaling could lead to higher demand for Nvidia's chips over price-performant chips. Theoretically, the AI may take longer to respond if price-performant chips cannot provide the same inferencing power as NVIDIA's chips. Depending on an organization's need for its AI to respond quickly, it still may want to use Nvidia's chips over Amazon's Trainium chips, for instance.
Going deeper down the rabbit hole, both R-1 and GPT-o1 use Time Test Scaling when inferencing. The comparison between the two AI models when responding to questions may be the primary reason people are interested in DeepSeek, not the pretraining costs. Seeking Alpha discourages linking to specific YouTube videos, so I will refrain from doing so here. However, if you want to see a few comparisons of the two AI models answering questions, Google “GPT-01 versus DeepSeek comparison,” and several videos will pop up.
One big difference between the two AI models is that DeepSeek's models are free and open source. In contrast, GPT-o1 is only available to premium users of ChatGPT for $20 a month for limited service and $200 for unlimited service through ChatGPT pro plan. Be aware that the o1 model through the ChatGPT pro plan is slightly more advanced than the AI model on the $20 limited service. This free versus open-source battle illustrates how open-source models could potentially hurt companies using an AI-as-a-service business model.
In response to questions, R-1 appears to have a more extensive reasoning process for generating an answer. The $20/month plan GPTo1 service typically has a shorter reasoning process behind obtaining a final answer than R-1. In the videos that I have watched, GPT-o1 ($20/month version) responds (“thinks”) faster, but occasionally gets difficult questions wrong. The $200 version of GPT-o1 answered the questions correctly that the $20 of GPT-o1 got wrong. The free R-1 and the $200 GPT-o1 appear to be equivalent services. However, there are reasons enterprises would choose GPT-o1 over R-1. KrebsonSecurity recently published an article that stated:
New mobile apps from the Chinese artificial intelligence ((AI)) company DeepSeek have remained among the top three "free" downloads for Apple and Google devices since their debut on January 25, 2025. But experts caution that many of DeepSeek's design choices — such as using hard-coded encryption keys, and sending unencrypted user and device data to Chinese companies — introduce a number of glaring security and privacy risks... DeepSeek's rapid rise caught the attention of the mobile security firm NowSecure, a Chicago-based company that helps clients screen mobile apps for security and privacy threats. In a teardown of the DeepSeek app published today, NowSecure urged organizations to remove the DeepSeek iOS mobile app from their environments, citing security concerns.
I don't think DeepSeek will ever be a serious threat to OpenAI, Microsoft (MSFT), or Alphabet (GOOGL, GOOG) because it's unlikely a Western company will trust a Chinese company's technology with sensitive information. However, DeepSeek has opened up the path for open-source companies from Western nations, possibly introducing an app that could threaten the AI-as-a-Service business model.
What about Nvidia? If you believe the company's management team, Test Time Scaling will only increase demand for Nvidia's more powerful AI chips in the inference phase over more price-performant chips. Additionally, AI models are becoming increasingly more complex. Chief Executive Officer Jensen Huang said on the company's third-quarter earnings call:
On the other hand, the models are getting larger, they're multimodality. Just the number of dimensions that inference is innovating is incredible. And this innovation rate is what makes Nvidia's architecture so great because our ecosystem is fantastic. Everybody knows that if they innovate on top of CUDA on top of Nvidia's architecture, they can innovate more quickly and they know that everything should work. And if something were to happen, it's probably likely their code and not ours. And so that ability to innovate in every single direction at the same time, having a large installed base so that whatever you create could land on a Nvidia computer and be deployed broadly all around the world in every single data center all the way out to the edge into robotic systems, that capability is really quite phenomenal.
CEO Huang alludes to several advantages in the statement above that help keep it at the top of the AI chip market, including the CUDA platform, the ecosystem, developer tools, software, customer support, reliability, and a massive performance advantage over nearest competitors. A Seeking Alpha article quoting Evercore ISI analyst Mark Lipacis stated:
Nvidia is still the platform of choice for hyperscalers' customers, with it being 5 to 10 years ahead of "anything else in the market," Lipacis added. AMD (AMD) and Amazon Web Services are seen as a "distant" second and third, Lipacis explained. And while application specific integrated circuits [price-performant chips] will have a role, external workloads (such as cloud facing AWS, Google Cloud and Azure), and enterprise on-premise will likely remain dominated by Nvidia, Lipacis said.
The company also has several other defenses against a future where companies may shift away from more powerful chips toward more price-performant chips. For instance, Nvidia is a free cash flow (“FCF”) machine today, and it can invest heavily in developing price-performant chips if it sees that's the direction in which the market is evolving.
Nvidia may have started as a hardware company that manufactured GPUs. However, it is evolving into a company that provides end-to-end AI solutions. It gives customers software tools to build chatbots, AI virtual assistants, and virtual agents and now sells cloud services alongside the company's high-performance AI chips. So, it's more than just a chip provider—it's a full-fledged AI company.
It also emphasizes the Total Cost of Ownership (“TCO”) over its entire AI infrastructure solution rather than focusing solely on making the chip, which provides a powerful defense against competitors only selling price-performant chips. NVIDIA includes the whole hardware and software ecosystem, support, operating expenses, and the ability to deploy AI solutions rapidly into its TCO calculations. Management wants to be able to go to a customer and say, “Our AI chips may cost more upfront, but our whole AI solution saves money over the long term.”
So, although DeepSeek and momentum towards driving AI costs lower threaten Nvidia in the long term, it has ways of fighting back, and whether this risk will manifest adversely in the company's business remains highly uncertain.
Company fundamentals
NVIDIA's Third-quarter FY 2025 Investor Presentation.
Outstanding Hopper sales during the third quarter helped the Data Center segment revenue increase 17% sequentially and 112% year-on-year to $30.8 billion. Hopper is the name of a specific Nvidia GPU architecture family, and H200 is the name of its most advanced model within the Hopper family. Chief Financial Officer (“CFO”) Colette Kress said on the company's third quarter FY 2025 earnings call (emphasis added):
Sequentially, Nvidia H200 sales increased significantly to double-digit billions, the fastest product ramp in our company's history. The H200 delivers up to 2 times faster inference performance and up to 50% improved TCO. Cloud service providers [CSPs] were approximately half of our data center sales with revenue increasing more than 2 times year-on-year. CSPs deployed Nvidia H200 infrastructure and high-speed networking with installations scaling to tens of thousands of GPUs to grow their business and serve rapidly rising demand for AI training and inference workloads. Nvidia H200-powered cloud instances are now available from AWS, CoreWeave, and Microsoft Azure with Google Cloud and OCI [Oracle (ORCL) Cloud Infrastructure] coming soon. Alongside significant growth from our large CSPs, Nvidia GPU regional cloud revenue jumped 2 times year-on-year as North America, EMEA, and Asia Pacific regions ramped Nvidia cloud instances and sovereign cloud buildout.
Outside CSPs, it more than doubled its revenue from consumer internet companies buying its Hopper technology to power training next-generation AI models, multimodal and agentic AI, deep learning recommender engines, generative AI, and content creation. CFO Kress also said on the earnings call (emphasis added):
“Nvidia's Ampere [the architectural family before Hopper] and Hopper infrastructures are fueling inference revenue growth for customers. NVIDIA is the largest inference platform in the world.”
Management likely pointed out that Nvidia has the largest inference platform because that's where it is most vulnerable to losing market share in the long term to competitors with price-performant chips.
CFO Kress said on the earnings call that it shipped its first Blackwell samples to customers during the third quarter. Blackwell is the company's latest architectural family and is in high demand because it's the most powerful chip on the market today. Blackwell can dramatically reduce the training time of large AI models that once took weeks or even months on hardware from two or three years ago. In the rapidly evolving AI industry, the faster developers can bring their innovations to market, the better their chance of succeeding.
The CFO said on the earnings call:
Blackwell is now in the hands of all of our major partners and they are working to bring up their Data Centers. We are integrating Blackwell systems into the diverse Data Center configurations of our customers. Blackwell demand is staggering and we are racing to scale supply to meet the incredible demand customers are placing on us.
Investors can expect Blackwell to drive significant Data Center revenue growth over the next year.
Gaming revenue increased 14% sequentially and 15% year over year to $3.3 billion. Automotive revenue increased 30% sequentially and 72% year-over-year to $449 million. Professional Visualization revenue increased 7% sequentially and 17% year over year to $486 million.
Nvidia's third-quarter FY 2025 total revenue increased 17% sequentially and 94% year-over-year to $35.1 billion, above the company's guidance of $32.5 billion and beating analysts' estimates by $1.95 billion.
NVIDIA's Third-quarter FY 2025 Investor Presentation.
Nvidia's GAAP (Generally Accepted Accounting Principles) gross margin expanded 60 basis points (“bps”) year over year to 74.6% due to a higher mix of higher-margin Data Center revenue. Non-GAAP gross margins remained flat year over year at 75%. However, GAAP and non-GAAP gross margins declined sequentially by 50 bps and 70 bps, respectively. A mix shift from mature H100 systems to the newer, more complex H200 systems was primarily responsible for lower sequential gross margins. When ramping up new hardware, gross margins typically decline until the company refines its manufacturing processes and supply chain.
The company's third quarter FY 2025 GAAP operating income increased 110% over the previous year's comparable quarter to $21.869 billion. GAAP operating margin expanded by 484 bps year over year to 62.33%. Its non-GAAP operating income increased 101% year over year to $23.276 billion. The non-GAAP operating margin was 66.34%, which was flat year over year.
Nvidia's GAAP and non-GAAP net income were $19.309 billion and $20.010 billion, respectively. It grew GAAP diluted earnings-per-share (“EPS”) by 109% to $0.78, beating analysts' estimates by $0.08. Non-GAAP diluted EPS grew by 103% year over year to $0.81, beating analysts' estimates by $0.06.
Data by YCharts
The company ended the third quarter with cash and marketable securities of $38.49 billion and long-term debt of $8.46 billion. Nvidia's trailing 12-month (“TTM”) cash from operations (“CFO”) was 52.05%, which means that for every $1 in sales, the company generates $0.52 in CFO—extremely impressive.
Data by YCharts
Nvidia's third-quarter FY 2025 CFO increased 22% sequentially and 140% year over year to $17.63 billion. TTM CFO was $58.96 billion.
Data by YCharts
The company's third quarter FY capital expenditures (“CapEx”) increased 192% year over year to $813 million, and TTM CapEx increased 82.25% year over year to $2.413 billion. Although Nvidia doesn't manufacture the chips it designs, it has high CapEx needs because it has evolved into an AI infrastructure company. CFO Colette Kress said on the earnings call,
“Our investments include building data centers for development of our hardware and software stacks and to support new introductions.”
As a result, Nvidia has more CapEx than a typical fabless semiconductor manufacturer.
Data by YCharts
Nvidia's third-quarter FY 2025 FCF grew 138% year over year to $16.814 billion. TTM FCF grew 223% year over year to $56.55 billion.
Data by YCharts
Management used some of its cash flow to repurchase $11.0 billion in shares and pay $245 million in cash dividends.
NVIDIA's Third-quarter FY 2025 Investor Presentation.
Management forecasts total revenue of $37.5 billion in the fourth quarter of FY 2025. If it meets this guidance, fourth-quarter revenue will rise 69.7% year over year, and full-year FY 2025 revenue will increase 111% over FY 2024 to $128.66 billion. Nvidia CFO Kress said the following about revenue guidance:
[Revenue guidance] incorporates continued demand for Hopper architecture and the initial ramp of our Blackwell products, while demand is greatly exceed supply, we are on track to exceed our previous Blackwell revenue estimate of several billion dollars as our visibility into supply continues to increase. On gaming, although sell-through was strong in Q3, we expect fourth-quarter revenue to decline sequentially due to supply constraints.
Management forecasts fourth quarter FY 2025 GAAP gross margins of 73.0% and non-GAAP gross margins of 73.5%. This guidance is well below the company's peak GAAP gross margins of 78.4% and non-GAAP gross margins of 78.9% in the first quarter of FY 2025. On the earnings call, CFO Kress stated the reason for the gross margin decline:
As Blackwell ramps, we expect gross margins to moderate to the low-70s. When fully ramped, we expect Blackwell margins to be in the mid-70s.
Later in the earnings call, analysts asked several questions about the Blackwell ramp and how long it would be before gross margins rebounded. Colette Kress responded by saying:
Could we reach the mid-70s in the second half of next year [Calendar year 2026]? And yes, I think it is reasonable assumption or a goal for us to do, but we'll just have to see how that mix of ramp goes. But yes, it is definitely possible.
Analysts are highly interested in when gross margins will rebound because this can impact FCF and EPS, the numbers people use to value the stock. A prolonged decline in gross margins could adversely affect investors' valuation of the stock.
When the company reports its fourth-quarter results on February 26, note what management reports for the first quarter FY 2026 guidance. Currently, consensus analyst revenue estimates are for $41.94 billion. If the company's guidance exceeds those revenue estimates, the market will likely take it as a sign that the DeepSeek news was a non-event. If management's first-quarter guidance comes under analysts' estimates, concerns about DeepSeek could resurface.
Valuation
Nvidia's price-to-sales (P/S) ratio is 28.92. With the significant decline in the company's market cap over the last several months, it trades closer to its three-year median, and some may consider it fairly valued. Yet others may believe the market still overvalues the stock based on where it trades above its five-, seven-, and ten-year median.
Data by YCharts
The company's price-to-earnings (P/E) ratio is 52.44, which is well above the information technology (“IT”) sector's median P/E of 32.63, which may lead some to consider it overvalued. However, some may think the market undervalues it, based on the stock's P/E trading at 38.18% below its five-year average P/E.
Nvidia's one-year forward P/E-to-growth is 0.59 (One-year forward P/E of 29.90 divided by analysts' estimated EPS growth of 50.43%). Typically, growth investors will allow a stock to reach a PEG ratio of 2.0 before considering it overvalued. According to that convention, the market may vastly undervalue Nvidia's EPS growth over the next year. If its PEG ratio were 1.0, the stock price would be $223.91, a 71% rise above its February 12 closing stock price of $131.14. If its PEG ratio were 2.0, the stock price would be $447.81.
Seeking Alpha
Let's analyze Nvidia using a reverse discount cash flow (“DCF”). This DCF uses a terminal growth rate of 4% because the company should continue steadily growing its cash flow well above the market average after the ten-year analysis period. I use a discount rate of 9%, which is the opportunity cost of investing in Nvidia, reflecting a less-than-moderate risk level. This reverse DCF uses an unlevered FCF for the following analysis.
If Nvidia can maintain a 50% FCF margin for the next ten years, the company must grow revenue by 17.9% annually to justify the February 12, 2025, closing price. Is this feasible? Its revenue has grown by around 31% over the last ten years, and in a previous article on the company, I explained why I believe the company can achieve at least 20% annual revenue growth over the next ten years. Analysts estimate Nvidia's revenue will grow at a 22.25% CAGR over the next ten years. Hence, a 17.9% yearly growth rate is within the realm of possibility.
However, a nearly 50% FCF margin is likely Nvidia's peak. In a market with rising competition, it is unlikely to sustain a 50% FCF margin over the next ten years. Broadcom (AVGO), a peer, achieved a peak FCF margin of 49.23% over the last ten years but has a median FCF margin of 40.28% in the previous ten years. Let's assume NVIDIA's FCF averages 40% over the next ten years; at an 18% annual growth rate, the estimated intrinsic value would be $106.19. Assuming the company grows revenue at analysts' estimated growth rate of around 22% annually over the next ten years, the estimated intrinsic value would be $143.62. A revenue growth rate of 20.8% over the next ten years at an average FCF margin of 40% would justify the February 12, 2025, stock price of $131.14.
The above reverse DCF uses many assumptions, and no one should believe that the company will hit any of the numbers above exactly. I used the above reverse DCF to loosely determine what the current stock price implies about the market's expectations for the company's FCF margin and revenue growth rates.
Data by YCharts
Nvidia's price-to-FCF is 57.63, well below its three-, five-, and seven-year median. It only exceeds its ten-year median. Some may consider the stock undervalued because of where its price-to-FCF trades compared to its medians. However, Seeking Alpha's quant grades the stock's valuation as a D-.
Risks
Nvidia had at least three customers, who accounted for more than 10% of revenue in the three- and nine-month period ending on October 27. Additionally, it has an indirect customer who purchased through customer C and resellers, which makes forecasting demand from that customer much more difficult. Nvidia's reliance on a small number of customers means the company has a customer concentration risk. Nvidia could underperform investors' revenue and profitability expectations if any of its top customers have financial issues or cancel or reduce orders.
NVIDIA Third Quarter FY 2025 10-Q
Additionally, the AI market is rapidly evolving. If Nvidia misses a new industry trend or fails to adapt quickly to industry changes, it could lose market share in a competitive market. Although I don't think DeepSeek's technology presents a real threat to Nvidia's business, it made people aware that if a company or technology does appear that can legitimately reduce the need for increasingly more powerful chips to train and run AI models, it could potentially threaten Nvidia's position as top AI chip provider by making price-performant chips more desirable.
Nvidia stock remains a buy
As I said in a previous article, most buy-and-hold investors would be better off waiting for the stock's FCF yield to rise above 2% before buying. Despite the stock's high returns over the last several years, I have been leery about recommending it for long-term investors since it may not have been a good value to invest in long-term since 2022. At an FCF yield of 1.73%, risk-averse investors should avoid it, considering a potential selloff could be steep if the company misses revenue, EPS, and FCF expectations.
Data by YCharts
However, this company has a high potential upside for at least the next year or two. Analysts expect revenue to grow 51.69% year over year in FY 2026, which is impressive for a company with $113.27 billion in TTM revenue. That revenue growth comes on the back of Blackwell's ramping over the next year and recent earnings reports from cloud companies indicating they are increasing CapEx in 2025, with much of that spending likely winding up with NVIDIA; aggressive growth investors interested in the continuing adoption of AI and willing to accept the risk of the stock's potential downside should consider buying. I rate Nvidia a buy for aggressive growth investors only.
