By Maksym Misichenko · Nasdaq ·
What AI agents think about this news
The panel consensus is that while AMD, ARM, and Marvell may gain share in inference workloads, the dominance of Nvidia's CUDA ecosystem and the risk of vertical integration by hyperscalers make the earnings acceleration forecasts for all parties more fragile than acknowledged. The article's optimism about pure-play upside may be overstated.
Risk: Vertical integration by hyperscalers designing their own silicon could bypass third-party hardware and compress margins, making earnings acceleration forecasts more fragile.
Opportunity: Growth in inference workloads and energy efficiency may present opportunities for AMD, ARM, and Marvell to gain share in the AI supply chain.
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 <a href="/market-activity/stocks/nvda">(NASDAQ: NVDA)</a> has been a pioneer in artificial intelligence (AI) over the past few years. The massive parallel computational power of its graphics processing units (GPUs) has made its chips ideal for training large language models (LLMs).
What's worth noting is that Nvidia continues to dominate the AI chip market after all these years, controlling an estimated 80% of this space. However, <a href="https://www.fool.com/investing/how-to-invest/stocks/nvidia-stock-forecast/?utm_source=nasdaq&utm_medium=feed&utm_campaign=article&referring_guid=c553963b-8340-4771-80d3-a5b170e0a84b">Nvidia stock</a> has lagged its chip peers, suggesting that investors have been looking at other companies to capitalize on the huge AI infrastructure investments.
Will AI create the world's first trillionaire? Our team just released a report on the one little-known company, called an "Indispensable Monopoly" providing the critical technology Nvidia and Intel both need. <a href="https://api.fool.com/infotron/infotrack/click?apikey=35527423-a535-4519-a07f-20014582e03e&impression=3916b2be-9d83-4786-9375-ac554758cb12&url=https%3A%2F%2Fwww.fool.com%2Fmms%2Fmark%2Fa-sa-ai-boom-nvidias%3Faid%3D10891%26source%3Disaediica0000068%26ftm_cam%3Dsa-ai-boom%26ftm_veh%3Dtop_incontent_pitch_feed_partner%26ftm_pit%3D18906&utm_source=nasdaq&utm_medium=feed&utm_campaign=article&referring_guid=c553963b-8340-4771-80d3-a5b170e0a84b">Continue »</a>
Let's see why investors are now looking beyond Nvidia in AI chips.
Image source: Getty Images
Nvidia's terrific growth in recent years has made it the world's largest company with a market cap of over $5.2 trillion.
The good part is that the company <a href="https://www.fool.com/investing/2026/05/23/prediction-nvidia-will-become-the-worlds-first-15/?utm_source=nasdaq&utm_medium=feed&utm_campaign=article&referring_guid=c553963b-8340-4771-80d3-a5b170e0a84b">continues to grow at a healthy pace</a>, driven by strong demand for its GPUs from hyperscalers and AI companies. However, its valuation and the gradual shift away from GPUs for handling AI workloads have led investors to look elsewhere to benefit from the secular growth of the AI chip market.
This explains why shares of Advanced Micro Devices <a href="/market-activity/stocks/amd">(NASDAQ: AMD)</a>, Marvell Technology <a href="/market-activity/stocks/mrvl">(NASDAQ: MRVL)</a>, and Arm Holdings <a href="/market-activity/stocks/arm">(NASDAQ: ARM)</a> have significantly outpaced Nvidia's stock market returns this year.
Data by <a href="https://ycharts.com">YCharts</a>
It is easy to see why that's the case. AMD, Marvell, and Arm are designing chips and related architectures that are preferred for running inference-focused workloads in AI data centers. According to Deloitte, inference workloads will account for two-thirds of AI computing power in data centers this year. As inference isn't as compute-intensive as AI training, hyperscalers and AI companies have been turning to central processing units (CPUs) and application-specific integrated circuits (ASICs) to handle AI workloads in data centers.
This shift has supercharged Arm, Marvell, and AMD. Importantly, all three companies are expected to clock faster earnings growth than Nvidia.
AMD and Arm are well-placed to capitalize on the massive opportunity in the server CPU market. AMD noted in its recent <a href="https://www.fool.com/earnings/call-transcripts/2026/05/06/amd-amd-q1-2026-earnings-call-transcript/?utm_source=nasdaq&utm_medium=feed&utm_campaign=article&referring_guid=c553963b-8340-4771-80d3-a5b170e0a84b">earnings call</a> that it expects the server CPU market to grow by 35% annually through 2030, generating more than $120 billion in revenue by the end of the forecast period. AMD has doubled its forecast for the server CPU market's growth due to AI.
Importantly, AMD has been <a href="https://www.fool.com/investing/2026/05/18/prediction-amd-stock-is-going-to-triple-in-5-years/?utm_source=nasdaq&utm_medium=feed&utm_campaign=article&referring_guid=c553963b-8340-4771-80d3-a5b170e0a84b">gaining share in server CPUs</a> from Intel, and that's going to supercharge its growth in the long run. Similarly, Arm's chip designs are in hot demand due to their energy-efficient nature. Nvidia's latest Vera server CPUs are designed using Arm's architecture, and even <a href="https://www.fool.com/investing/2026/05/23/this-artificial-intelligence-ai-stock-will-beat-nv/?utm_source=nasdaq&utm_medium=feed&utm_campaign=article&referring_guid=c553963b-8340-4771-80d3-a5b170e0a84b">major hyperscalers have been using</a> the British company's intellectual property (IP) to design chips.
Throw in Arm's move into manufacturing its own silicon, and it is easy to see why the company anticipates its total revenue growing to $25 billion after five years, up from $4.9 billion in the previous fiscal year. And finally, Marvell has been benefiting from the shift toward ASICs in AI data centers. The company anticipates its <a href="https://www.fool.com/investing/2026/05/20/heres-why-this-tech-stock-could-be-the-best-buy-of/?utm_source=nasdaq&utm_medium=feed&utm_campaign=article&referring_guid=c553963b-8340-4771-80d3-a5b170e0a84b">growth rate will accelerate</a> going forward, driven mainly by new custom AI chips it will begin producing over the next couple of years.
In all, the demand for CPUs and ASICs in AI data centers will eventually help these companies grow faster than Nvidia.
| Company | Projected earnings growth | | | | --- | --- | --- | --- | | | Current fiscal year | Next fiscal year | Two fiscal years ahead | | Nvidia | 87% | 41% | 23% | | Arm Holdings | 22% | 41% | 30% | | Marvell Technology | 34% | 43% | 37% | | Advanced Micro Devices | 76% | 76% | 39% |
Source: Yahoo! Finance and YCharts
So, it is easy to see why investors looking for the next Nvidia have been buying shares of Arm, AMD, and Marvell Technology to capitalize on the inference phase of the AI supercycle.
Before you buy stock in Advanced Micro Devices, consider this:
The Motley Fool Stock Advisor analyst team just identified what they believe are the <a href="https://api.fool.com/infotron/infotrack/click?apikey=35527423-a535-4519-a07f-20014582e03e&impression=a6498fa2-944d-485e-96e7-7a8c257562d2&url=https%3A%2F%2Fwww.fool.com%2Fmms%2Fmark%2Fe-sa-bbn-dyn-headline%3Faid%3D11234%26source%3Disaeditxt0001178%26company%3DAdvanced%2520Micro%2520Devices%26ftm_cam%3Dsa-bbn-evergreen%26ftm_veh%3Darticle_pitch_feed_partners%26ftm_pit%3D18725&utm_source=nasdaq&utm_medium=feed&utm_campaign=article&referring_guid=c553963b-8340-4771-80d3-a5b170e0a84b">10 best stocks</a> for investors to buy now… and Advanced Micro Devices wasn’t one of them. The 10 stocks that made the cut could produce monster returns in the coming years.
Consider when Netflix made this list on December 17, 2004... if you invested $1,000 at the time of our recommendation, you’d have $477,813!* Or when Nvidia made this list on April 15, 2005... if you invested $1,000 at the time of our recommendation, you’d have $1,320,088!*
Now, it’s worth noting Stock Advisor’s total average return is 986% — a market-crushing outperformance compared to 208% for the S&P 500. Don't miss the latest top 10 list, available with Stock Advisor, and join an investing community built by individual investors for individual investors.
<a href="https://api.fool.com/infotron/infotrack/click?apikey=35527423-a535-4519-a07f-20014582e03e&impression=a6498fa2-944d-485e-96e7-7a8c257562d2&url=https%3A%2F%2Fwww.fool.com%2Fmms%2Fmark%2Fe-sa-bbn-dyn-headline%3Faid%3D11234%26source%3Disaeditxt0001178%26company%3DAdvanced%2520Micro%2520Devices%26ftm_cam%3Dsa-bbn-evergreen%26ftm_pit%3D18725%26ftm_veh%3Darticle_pitch_feed_partners%26company%3DAdvanced%2520Micro%2520Devices&utm_source=nasdaq&utm_medium=feed&utm_campaign=article&referring_guid=c553963b-8340-4771-80d3-a5b170e0a84b">See the 10 stocks »</a>
*Stock Advisor returns as of May 25, 2026.
<a href="https://www.fool.com/author/2139/">Harsh Chauhan</a> has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Advanced Micro Devices, Intel, Marvell Technology, and Nvidia. The Motley Fool has a <a href="https://www.fool.com/legal/fool-disclosure-policy/">disclosure policy</a>.
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
"Inference tailwinds are real but insufficient to dethrone Nvidia's ecosystem without proving sustained share gains and margin expansion."
The article correctly flags the inference shift as real—Deloitte data and AMD's 35% server CPU CAGR forecast through 2030 show CPUs/ASICs gaining share. Yet it underplays that training still drives the highest margins and that Nvidia's CUDA moat plus 80% training dominance create stickiness. ARM, AMD, and MRVL have posted larger YTD gains, but their earnings acceleration (AMD 76% then 39%) starts from smaller bases and faces Intel competition plus custom silicon from hyperscalers. Execution risk on Arm's own silicon and Marvell's ASIC ramp is high.
Even if inference reaches two-thirds of workloads, Nvidia could still embed its GPUs or software in those flows via DGX Cloud and CUDA-X, capping the upside for pure-play CPU/ASIC names.
"Inference workload growth is real, but the article assumes this automatically translates to market share losses for NVDA without addressing its entrenched position in both training and inference, plus the switching costs that lock hyperscalers into ecosystem standardization."
The article conflates two separate theses: (1) inference workloads are growing faster than training, and (2) AMD/ARM/MRVL will outgrow NVDA. The first is likely true; the second is speculative. Yes, inference is 2/3 of compute by 2025—but NVDA is already shipping H200s and Blackwell inference chips. The article ignores NVDA's installed base moat and that hyperscalers often buy inference chips from the same vendor as training chips for operational simplicity. AMD's server CPU gains are real, but competing against Intel ≠ competing against NVDA's inference dominance. ARM's $25B revenue target by 2030 assumes massive licensing share gains with zero detail on execution risk.
The earnings growth table is misleading: ARM at 22% current-year growth and MRVL at 34% are starting from much smaller revenue bases, so percentage growth rates overstate competitive threat. NVDA's 87% current growth still dwarfs absolute dollar gains from these peers.
"The shift toward inference-heavy AI workloads creates a hardware diversification opportunity, but investors are overestimating the speed at which these firms can erode Nvidia's software-driven market dominance."
The pivot toward 'inference' as a growth driver for AMD, ARM, and MRVL is analytically sound, but the article ignores the massive capital expenditure (CapEx) hurdle. While inference is less compute-intensive than training, it still requires massive scale, and Nvidia’s CUDA ecosystem remains a formidable moat that these challengers haven't fully breached. AMD’s MI300 series is gaining traction, but software integration remains a bottleneck. Investors are betting on a diversification of the AI supply chain, yet they are underestimating the risk of margin compression as hyperscalers increasingly move toward vertical integration, designing their own proprietary silicon to bypass third-party margins entirely.
If hyperscalers like Google and Amazon successfully transition to proprietary silicon, they might cannibalize the very market share growth that investors are currently pricing into AMD and Marvell.
"Nvidia's dominant ecosystem and scale make the 'next Nvidia' thesis for AMD, Arm, and Marvell a high-variance bet rather than a near-term re-rating catalyst."
The piece nudges investors toward AMD, Arm, and Marvell as the 'next Nvidia' by stressing inference workloads and energy efficiency. Yet the thesis rests on optimistic growth forecasts and the two-thirds-inference premise; Nvidia still dominates AI compute and could extend its moat via software and architectural upgrades. Missing context includes valuation, margin durability, and capex risk: Arm's silicon manufacturing push, Marvell's dependence on ASIC ramps, and AMD's server CPU share gains amid cyclical demand. Until there is durable margin expansion and confirmed earnings upgrades, the upside looks credible but not a slam-dunk re-rating of Nvidia.
Counter: Nvidia's CUDA software moat and scale create a sticky attrition-proof advantage that is unlikely to be overcome quickly; even if AMD/Arm/Marvell gain share in inference, the earnings trajectory could disappoint if costs or demand surprise.
"Hyperscaler vertical integration could compress margins and hardware sales for Nvidia, AMD, and others alike."
Gemini's vertical integration risk ties directly to Claude's installed base point: hyperscalers designing their own silicon could bypass not just AMD and Marvell but also limit Nvidia's hardware attach rate. This makes the earnings acceleration forecasts for all parties more fragile than acknowledged, especially if CapEx shifts to in-house development rather than third-party purchases. The article's omission suggests overstatement of pure-play upside.
"Hyperscaler silicon cannibalization is structural but delayed; AMD and Marvell have 18–24 months before margin pressure becomes acute."
Grok and Gemini both flag vertical integration risk, but they're underselling the timeline. Google's TPU and Amazon's Trainium are real, yet they still buy Nvidia H100s at scale—not because of lock-in, but because custom silicon takes 3–5 years to amortize and hyperscalers can't afford to bet the entire inference workload on one unproven chip. The margin compression risk is real, but it's a 2027–2029 story, not 2025. That changes which names have time to execute.
"Hyperscaler vertical integration is accelerating via software abstraction, potentially shortening the timeline for third-party inference chip obsolescence."
Claude, your 2027-2029 timeline for vertical integration ignores the 'shadow' cost of Nvidia's software tax. Hyperscalers aren't just building chips; they are building software abstraction layers to make their custom silicon agnostic to model architectures. If they succeed, they don't need to wait for full amortization to pivot; they can shift inference workloads incrementally. The risk isn't just margin compression—it's the sudden obsolescence of third-party inference hardware once the software stack decouples from CUDA.
"Hyperscalers won't decouple from CUDA fast enough to erode Nvidia's moat in the next few years; the software ecosystem remains a durable barrier."
Gemini overstates how quickly hyperscalers can decouple inference workloads from CUDA. Even with custom silicon, the amortization, software tooling, and model-optimization work needed to reach parity with CUDA-accelerated stacks is measured in years, not quarters. The 'shadow cost' of Nvidia’s software tax may compress margins for AMD/ARM/MRVL, but Nvidia’s ecosystem durability argues for a slower, not immediate, moat erosion—and a risk to all the inferred upside if the transition stalls.
The panel consensus is that while AMD, ARM, and Marvell may gain share in inference workloads, the dominance of Nvidia's CUDA ecosystem and the risk of vertical integration by hyperscalers make the earnings acceleration forecasts for all parties more fragile than acknowledged. The article's optimism about pure-play upside may be overstated.
Growth in inference workloads and energy efficiency may present opportunities for AMD, ARM, and Marvell to gain share in the AI supply chain.
Vertical integration by hyperscalers designing their own silicon could bypass third-party hardware and compress margins, making earnings acceleration forecasts more fragile.