Perhaps no stock was more profoundly affected by the news from DeepSeek than Nvidia (NVDA 1.33%). In a sense, DeepSeek validated its dominance by announcing its H800 accelerators trained its model for less than $6 million. That cost is a tiny fraction of the hundreds of millions of dollars OpenAI has spent on ChatGPT's development cost.
However, the news calls into question the demand for Nvidia accelerators. Lower-cost AI will likely increase demand for the technology, which may help Nvidia.
Still, if entities can build AI models on less expensive technologies, they will presumably have less need to pay premium prices for its top-of-the-line AI chips.
Knowing that, investors may wonder how best to invest in Nvidia stock going forward. In this article, three Motley Fool contributors will share their opinions on Nvidia and what investors should do next.
NASDAQ: NVDA
Key Data Points
Long-term investors would be wise to tune out the DeepSeek noise
Jake Lerch: When it comes to my opinion on Nvidia, I'm unmoved by the recent DeepSeek AI announcement.
First of all, I'm taking the news with a healthy spoonful of skepticism. Granted, some AI experts have praised the open-source code that has been released, noting it is an impressive and significant breakthrough.
However, questions remain about which -- and how many -- Nvidia GPUs were used to train this new model. But bear in mind that the U.S. government has placed export restrictions on Nvidia GPUs, specifically to prevent Chinese interests from gaining access to them.
In other words, don't hold your breath waiting for full transparency on how DeepSeek was built -- and for what purpose.
In any event, Nvidia's position as the leading horse in the race to design the best AI chips remains intact. Moreover, the next step in the AI revolution isn't about who will build the next great large language model (LLM).
Instead, investors should remain focused on the bigger picture: AI-powered tools and applications are poised to deliver real-world productivity gains across the economy. Logistics, healthcare, finance, and every other sector will likely generate more sales and more profits as the AI revolution rolls on.
Moreover, this breakthrough, such as it is, will not alleviate the need for better and faster chips. After all, if the technological revolution of the last four decades has taught us anything, it's that no one wants to be stuck with outdated technology. People, companies, and governments are always eager to pay up for newer and faster computational power.
Remember: The world wasn't satisfied with dial-up internet, and we won't be satisfied with Nvidia A100s, either. The company will continue to innovate with improved chips like its latest, the Blackwell. And it is my belief that those chips will deliver innovations that will outshine DeepSeek.
Nvidia's stock plunged at the mere thought of reduced AI spending
Justin Pope: It's been quite a stressful week for Nvidia shareholders. The stock plunged by double digits after DeepSeek, a Chinese AI company, recently sent Wall Street scrambling. DeepSeek claims its AI models deliver similar performance to OpenAI's while being open-source (freely available to developers) and costing significantly less to train and operate.
More specifically, DeepSeek claims it trained its V3 (chat) AI model using a cluster of 2,048 Nvidia H800 GPUs, which cost approximately $5.5 million. This would be a remarkable achievement considering the tens of millions of dollars AI companies, like OpenAI and Google (owned by Alphabet), have spent on Nvidia's superior H100 chips. The company credits software engineering for the accomplishment, leaning toward efficiency rather than hardware to deliver the performance it needs.
The fear is that Nvidia's GPU sales could cool off as other companies replicate DeepSeek's techniques. Nvidia depends on a small handful of AI hyperscalers for a considerable chunk of its revenue.
Multiple industry experts are highly skeptical of the $5.5 million training costs. While Nvidia has claimed that DeepSeek operated within the parameters of chip export restrictions, some question whether the figure excludes additional costs or whether the model was genuinely developed from scratch. That said, DeepSeek's breakthroughs in efficiency seem more plausible. Because its model is open-source, anyone can see how the company did it, and can also run tests. Plus, DeepSeek is severely undercutting OpenAI's pricing. Its R1 (reasoning) model's API outputs cost just $2.19 per 1 million tokens versus $60 for ChatGPT's o1.
The bottom line for Nvidia is that it's probably too early to react. Technology companies like Microsoft still can't support their AI demand, so it's unlikely that these massive AI investments will drop off overnight. Business should remain strong for Nvidia. However, over time, a push for cost efficiency could dent GPU demand if efficiency through software engineering can help lengthen the lifespan of existing AI chips.
That arguably makes Nvidia a riskier stock moving forward than a few weeks ago. It's still one of the best AI stocks one can buy, but to minimize risk, consider dollar-cost averaging and diversifying your portfolio.
Decelerate your enthusiasm in this AI accelerator stock
Will Healy: The challenge to investing in Nvidia after the DeepSeek news comes down to one attribute: competitive advantage.
Nvidia's financials and stock have been on fire since the spring of 2023 when OpenAI revealed that Nvidia AI accelerators powered GPT-4. None of Nvidia's competitors had offered an alternative at that time. Nvidia's accelerators and corresponding CUDA software remain far ahead of competitive offerings.
Indeed, demand for the most advanced accelerators is unlikely to disappear, and it is possible Nvidia's competitive advantage will remain intact once analysts dig into the details of DeepSeek's breakthrough. Additionally, DeepSeek did not leave the Nvidia universe to train its models, which bodes well for the company.
However, high demand has taken the prices of its state-of-the-art H200 accelerators above $30,000. Should demand at the top end fall, it could force Nvidia to cut the price of this technology.
This is concerning since the data center segment, which designs the accelerators, was 87% of company revenue in the first nine months of fiscal 2025 (ended Oct. 27, 2024). Also, the $91 billion in revenue during that time frame rose 135% year over year.
Nonetheless, if this latest development reduces its pricing power, that could mean significantly lower revenue per unit, which would presumably slow or possibly reverse Nvidia's revenue growth.
Moreover, valuations show Nvidia stock may have a long way to fall. Many investors may overlook this issue since a 47 P/E ratio is arguably low for a fast-growing stock. Unfortunately, the price-to-book ratio of 46 is far above AMD's book value multiple of 3. Investors would likely question that book value multiple in an environment of slowing revenue growth.
Furthermore, Nvidia has fallen by more than 50% twice since 2018, a reminder that Nvidia's momentum can turn quickly to the negative.
It is premature to assume DeepSeek's breakthrough will bring another such decline in the stock price, and even if it does, Nvidia is unlikely to lose its overall competitive advantage. Still, with considerable downside possible, now may not be a good time to hold a disproportionately large position in Nvidia stock.
