By Maksym Misichenko · Nasdaq ·
What AI agents think about this news
The panel consensus is bearish, with key risks including potential deceleration in hyperscale capex, AI model efficiency reducing compute intensity, regulatory/export restrictions, and competitive pressure from custom chips. The 'Capex Hangover' thesis and erosion of CUDA lock-in are significant concerns.
Risk: Erosion of CUDA lock-in due to custom ASICs and potential tightening of export restrictions
Opportunity: Growing sovereign demand for AI hardware, creating a non-cyclical floor for Nvidia's revenue
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Nvidia's (NASDAQ: NVDA) stock has been a huge winner each of the past two years. After surging over 238% in 2023, the stock has soared approximately 164% this year, as of this writing. Those are two huge back-to-back year gains that have propelled the company to become one of the largest in the world.
The question is, can the stock hit the market with a three-peat of outsized gains in 2025? Interestingly, the stock has been able to generate returns of 30% or more for three straight years on four previous occasions and returns of 50% or greater for three straight years twice. It has never had four years in a row of 30% or more returns, but it did have one stretch where its stock rose by 25% or more for five straight years from 2013 to 2017.
Where to invest $1,000 right now? Our analyst team just revealed what they believe are the 10 best stocks to buy right now. See the 10 stocks »
Let's look at why I think Nvidia can turn in another year of strong performance in 2025.
Any investment in Nvidia centers around spending on artificial intelligence (AI) infrastructure. The graphics processing units (GPUs) that it designs have become the backbone of the AI infrastructure buildout, as GPUs can perform many calculations at the same time, making them ideal for use in training large language models (LLMs) and running AI inference.
Meanwhile, as AI models become more sophisticated, they need exponentially more computing power, and thus GPUs, to advance. For example, both Amazon's Llama 4 LLM and xAI's Grok 3 model were trained on 10 times as many GPUs as their predecessors trained on.
Demand for GPUs is being driven by large hyperscale (companies with massive data centers) tech companies (such as Microsoft, Alphabet, Amazon, and Meta Platforms) as well as well-funded AI start-ups like OpenAI and Elon-Musk backed xAI. Currently, these companies are all racing to create the best and most powerful AI models, leading to what Nvidia has called "insane" demand for its newest-generation Blackwell GPUs.
However, growth is not expected to stop, with Nvidia's largest customers, by and large, indicating that they plan to spend more on building out data centers to help power their AI ambitions. Nvidia customers such as Meta Platforms and Alphabet have said the biggest risk with AI infrastructure is underinvesting, as they look to capitalize on what they see as a generational opportunity. Oracle, meanwhile, has said it expects strong AI infrastructure growth to continue over the next five to 10 years.
Nvidia isn't the only company that makes GPUs, but it has been able to create a wide moat in large part due its CUDA software platform. GPUs were originally developed to speed up graphics rendering (hence the name) in applications like video games. However, as Nvidia looked to expand the use case for these chips, it created a free software program that allowed developers to program its chips for other tasks.
While it took time, this led to CUDA becoming the standard on which developers learned to program GPUs for various tasks, creating the wide moat it has today. Meanwhile, it was arguably the use of its GPUs in cryptocurrency mining that really helped set the groundwork for Nvidia's current AI success today, as it demonstrated the power of its GPUs in high-performance computing.
Nvidia has not sat still following its initial CUDA development and in the years since it has built domain-specific microservices and libraries on top of Cuda, called CUDA X, to better optimize it for AI. Meanwhile, the company has also sped up its development cycle for its GPUs to once a year in order to remain at the forefront of GPU technology.
The company's biggest challenge at the moment appears to be coming from custom AI chips, such as those Broadcom helps develop for customers. These are custom chips designed for very specific tasks, and thus they can be more efficient. However, it also takes time to design and manufacture custom chips, and like most custom things, they are more expensive. In a world racing for AI, Nvidia's chips are more accessible and cheaper and have an array of AI-specific microservices and libraries through CUDA X.
As such, while custom AI chips will likely continue to take some share, Nvidia still looks like it will remain the king of AI chips for the foreseeable future.
The final reason why I think Nvidia is poised for another year of outperformance in 2025 is its valuation. Despite its huge gains over the past two years, the stock only trades a forward price-to-earnings (P/E) ratio of about 30 based on 2025 analyst estimates, and a price/earnings-to-growth (PEG) ratio of approximately 0.95. A PEG ratio under 1 is typically viewed as undervalued, but growth stocks will often have PEG ratios well above 1.
For a company that just saw its revenue grow by 94% year over year last quarter and which is projected to see 50% revenue growth in 2025, that's an attractive valuation. With AI looking to be still in its early innings and the company having a wide moat, the stock looks like a buy heading into 2025.
Before you buy stock in Nvidia, consider this:
The Motley Fool Stock Advisor analyst team just identified what they believe are the 10 best stocks for investors to buy now… and Nvidia wasn’t one of them. The 10 stocks that made the cut could produce monster returns in the coming years.
Consider when Nvidia made this list on April 15, 2005... if you invested $1,000 at the time of our recommendation, you’d have $825,513!*
Stock Advisor provides investors with an easy-to-follow blueprint for success, including guidance on building a portfolio, regular updates from analysts, and two new stock picks each month. The Stock Advisor service has more than quadrupled the return of S&P 500 since 2002*.
*Stock Advisor returns as of December 16, 2024
John Mackey, former CEO of Whole Foods Market, an Amazon subsidiary, is a member of The Motley Fool's board of directors. Suzanne Frey, an executive at Alphabet, is a member of The Motley Fool's board of directors. Randi Zuckerberg, a former director of market development and spokeswoman for Facebook and sister to Meta Platforms CEO Mark Zuckerberg, is a member of The Motley Fool's board of directors. Geoffrey Seiler has positions in Alphabet. The Motley Fool has positions in and recommends Alphabet, Amazon, Meta Platforms, Microsoft, Nvidia, and Oracle. The Motley Fool recommends Broadcom and recommends the following options: long January 2026 $395 calls on Microsoft and short January 2026 $405 calls on Microsoft. The Motley Fool has a disclosure policy.
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.
Four leading AI models discuss this article
"Nvidia's valuation embeds flawless execution on Blackwell and sustained 50% growth that custom-ASIC competition and capex digestion could easily disrupt."
The article correctly flags Nvidia's CUDA moat and hyperscaler AI capex as durable drivers, yet it glosses over execution risk around Blackwell ramp delays and the growing share of custom ASICs from Broadcom and others. A 30x 2025 forward P/E assumes 50% revenue growth materializes without margin pressure; any shortfall in utilization or pricing power would compress multiples faster than the PEG metric suggests. Oracle's five-to-ten-year forecast is cited, but near-term digestion of current data-center buildouts remains unaddressed.
If custom-chip adoption stays below 15% of workloads through 2026 and hyperscalers maintain 40%+ AI capex growth, the 30x multiple could re-rate upward rather than contract.
"Nvidia's AI compute moat supports upside, but the stock's rich valuation makes it highly sensitive to any growth deceleration or regulatory headwinds, risking a re-rating even if earnings remain solid."
NVDA's long AI compute story remains compelling, but the article glosses over meaningful risks. Even with steep data-center demand, the market appears to be pricing roughly 50% 2025 revenue growth into a 30x forward P/E with PEG near 1, leaving little room for surprise. Risks include potential deceleration in hyperscale capex, AI model efficiency reducing compute intensity, and regulatory/export restrictions that could curb addressable markets (notably China). Competitive pressure from custom chips and ongoing CUDA licensing questions could erode the moat. If any of these materialize, multiple compression could offset earnings upside.
The flip side is that the bull case hinges on a resilient AI capex cycle; if hyperscale spend slows or export controls bite, Nvidia could see meaningful multiple contraction even with solid earnings.
"Nvidia's forward P/E of 30x is a trap because it assumes sustained, non-cyclical hyperscaler spending that ignores the inevitable rise of internal custom silicon and capital efficiency mandates."
Nvidia’s valuation at ~30x forward P/E is deceptive because it relies on analyst estimates that assume a linear continuation of hyperscaler CAPEX growth. While the CUDA moat is real, the article ignores the 'law of large numbers' and the inevitable margin compression as custom silicon from Broadcom and Marvell gains traction. Nvidia is currently priced for perfection, yet it faces a significant risk of a 'Capex Hangover'—a scenario where hyperscalers pause spending to improve ROI on existing clusters. If revenue growth decelerates from 94% to the projected 50%, the stock will likely experience a valuation multiple contraction, regardless of the absolute earnings growth.
The bull case for Nvidia is that it has effectively become the 'utility provider' for the next industrial revolution, making its revenue as essential and recurring as electricity, which justifies a premium valuation that traditional cyclical models fail to capture.
"NVDA trades at a 30x forward multiple that prices in flawless execution of 50% growth for years, leaving no margin for error as custom silicon adoption accelerates and hyperscaler capex cycles normalize."
The article's valuation argument is backwards. A 30x forward P/E on 50% projected growth means NVDA is priced for perfection—not undervalued. The PEG of 0.95 only works if that 50% growth materializes and sustains; one miss and multiple compression is brutal. More critically: the article ignores capex cycles. Hyperscalers are already signaling infrastructure spending plateaus in 2025 (Meta's recent guidance). Custom chips aren't a future threat—they're shipping now (Google TPUs, Amazon Trainium). The 'wide moat' argument conflates developer lock-in with customer lock-in; the latter is eroding as cloud providers optimize for their own silicon.
NVDA's installed base, CUDA ecosystem depth, and first-mover advantage in enterprise AI are genuinely durable. If hyperscalers' capex doesn't materially slow and new verticals (automotive, edge) ramp faster than expected, the stock could re-rate higher despite current valuation.
"Blackwell delays plus tighter China curbs could accelerate custom ASIC shifts beyond current forecasts."
Claude flags Meta's capex plateau signal but misses how Nvidia's Blackwell delays could actually front-run any 2025 slowdown by creating artificial scarcity that sustains pricing power through year-end. The bigger unmentioned linkage is between export curbs and custom ASIC adoption: if China restrictions tighten further, hyperscalers may accelerate their own silicon to bypass limits, eroding CUDA lock-in faster than utilization metrics alone would predict.
"Policy-driven hardware substitution due to export controls could accelerate hyperscalers' internal silicon, eroding CUDA moat faster than utilization or ramp Dynamics imply."
Grok, you link export curbs to custom ASIC adoption, but you understate how policy could accelerate silicon self-supply. If China restrictions tighten, hyperscalers may sprint toward internal AI accelerators, compressing CUDA lock-in beyond utilization trends and pricing power. In that case, even a Blackwell ramp or 2025 capex rebound may not sustain a multi-30x story; the moat could erode faster on policy-accelerated hardware substitution.
"Sovereign AI demand acts as a non-cyclical floor that mitigates the risk of hyperscaler capex plateaus and custom silicon adoption."
Claude and Grok are missing the 'sovereign AI' wildcard. While hyperscalers might pivot to custom silicon, national governments are aggressively buying Nvidia hardware to ensure compute independence, creating a floor that custom ASICs cannot touch. This demand is non-cyclical and ignores hyperscaler ROI metrics. If sovereign demand accounts for even 15% of 2025 revenue, the 'Capex Hangover' thesis fails because it assumes only cloud giants drive the market, ignoring the geopolitical arms race for compute.
"Sovereign demand creates a revenue floor but not a margin floor; it may actually incentivize competitors to build alternative stacks."
Gemini's sovereign AI demand floor is real but overstated. Government procurement is lumpy, politically fragile, and typically 5-10% of enterprise chip revenue historically. More critically: sovereign buyers often *accelerate* custom silicon development (see France's ThalesAI, Japan's initiatives). Gemini conflates demand with durable moat. The 'Capex Hangover' thesis survives sovereign demand unless it reaches 25%+ of revenue—currently unverifiable.
The panel consensus is bearish, with key risks including potential deceleration in hyperscale capex, AI model efficiency reducing compute intensity, regulatory/export restrictions, and competitive pressure from custom chips. The 'Capex Hangover' thesis and erosion of CUDA lock-in are significant concerns.
Growing sovereign demand for AI hardware, creating a non-cyclical floor for Nvidia's revenue
Erosion of CUDA lock-in due to custom ASICs and potential tightening of export restrictions