By Maksym Misichenko · Nasdaq ·
What AI agents think about this news
Despite competition from hyperscalers' custom chips, Nvidia's CUDA ecosystem, software moat, and dominant scale at TSMC make it well-positioned to maintain market leadership in AI chipsets. The key risk is potential capacity constraints at TSMC, while the key opportunity lies in Nvidia's ability to monetize via software and tooling even as in-house chips grow.
Risk: Potential capacity constraints at TSMC
Opportunity: Monetizing via software and tooling
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 →
Hyperscalers like Amazon and Alphabet have been seeing healthy demand for their custom AI processors.
These companies are leasing access to their in-house chips to third parties, and they have already landed lucrative contracts.
Their progress isn't good news for Nvidia, which has been the dominant player in AI chips over the past three and a half years.
Nvidia (NASDAQ: NVDA) has been one of the biggest beneficiaries of the artificial intelligence (AI) chip boom. Its graphics processing units (GPUs) are parallel processors, designed to break down certain types of massively complex calculations into a host of smaller parts, and then perform all of those small calculations simultaneously, rather than taking each task in sequence. And it turns out, the process of training large language models (LLMs) depends heavily on just the sort of tasks where GPUs excel.
As a result, over the past few years, demand for Nvidia's industry-leading GPUs has skyrocketed, driving stunning growth in the company's revenue and earnings.
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. Continue »
Major hyperscalers and AI companies, such as Amazon (NASDAQ: AMZN), Microsoft, Meta Platforms, and Alphabet's (NASDAQ: GOOG) (NASDAQ: GOOGL) Google, have long relied on Nvidia's hardware to train powerful AI models.
What's worth noting is that Nvidia's rivals haven't been able to make much of a dent in its AI chip dominance. It controls an estimated 81% of the AI data center chip market, according to IDC. The good news for Nvidia stock investors is that the company's red-hot growth could continue -- the company is forecasting total sales of $1 trillion for its Blackwell and Vera Rubin architectures across 2026 and 2027.
However, there is ample evidence that Nvidia's position in AI chips is gradually weakening.
Training LLMs requires a lot of computing power, which is why Amazon, Meta, Microsoft, Alphabet, and others have been purchasing millions of Nvidia GPUs. However, these customers have also been designing their own chips to run AI workloads cost-effectively in their data centers. The high costs and supply constraints associated with Nvidia's popular graphics cards explain why these customers have been working on their own chips in-house for a long time.
Google, for instance, launched the first generation of its Tensor Processing Unit (TPU) in 2015, while Amazon's in-house Trainium custom chip was launched in December 2020. Both companies have improved their chips over the years. In fact, they are now selling these chips to third parties.
Amazon, for instance, recently revealed that its chip business recorded 40% sequential growth in the first quarter of 2026. The annual revenue run rate of Amazon's semiconductor business is now more than $20 billion. What's more, the "Magnificent Seven" company notes that the segment's revenue run rate is improving by triple-digit percentages year over year.
Another key point is that the segment's annual run rate would be closer to $50 billion if it included its "sales" of chips to itself for use in AWS data centers. What's more, the demand for Amazon's Trainium chips is so strong that access to them is fully booked. Its custom AI processors are being deployed by Anthropic, OpenAI, Uber, and even Meta Platforms, which uses Amazon's in-house Graviton central processing unit (CPU) to support agentic AI applications.
As it turns out, Amazon has a whopping $225 billion in purchase commitments for its Trainium AI chips, clearly suggesting that its semiconductor business is poised for terrific growth.
Meanwhile, Google has also been making waves in the AI chip market. The tech giant has sizable deals in place with Meta Platforms and Anthropic for the deployment of its TPUs. CEO Sundar Pichai sees the TPU business as one of its key growth drivers, and the company is now selling its chips to more customers.
On Alphabet's latest earnings call, Pichai remarked:
As TPU demand grows from AI labs, capital markets firms, and high-performance computing applications, we will begin to deliver TPUs to a select group of customers in their own data centers in the hardware configuration to expand our addressable market opportunity.
This addressable opportunity could be massive in the long run. Though Google hasn't publicly revealed the size of its TPU business yet, investment firm D.A. Davidson estimates that it could be worth a whopping $900 billion in the long run, assuming the company decides to seriously sell its chips to third parties.
It now appears that Google is indeed becoming serious about its TPU business, and that's likely to create more problems for Nvidia's AI chip empire.
Nvidia isn't going to sit and watch while its customers turn into competitors. The reason Amazon and Google's custom processors have been gaining tremendous traction is that they are application-specific integrated circuits -- chips that are optimized to handle a relatively narrow range of workloads, in contrast to Nvidia's more flexible GPUs, which are suitable for a broad range of tasks. Custom chips can thus perform AI inference tasks more efficiently, reducing the total operating cost of data centers.
Nvidia is countering the threat from the likes of Amazon and Google by making improvements to its own hardware that significantly reduce the cost of AI inference with its GPUs. Also, Nvidia has decided to offer its Vera server CPU as a stand-alone product for the first time, rather than only offering it as part of the Vera Rubin platform. The company has made this move as it is seeing strong interest in its Vera server CPUs. In fact, the company believes that its server CPU business could become a multibillion-dollar play.
Nvidia's efforts to push the envelope in product development should help it ward off the rising competition. Also, investors shouldn't forget that the AI chip market continues to expand rapidly. Bank of America estimates that the global semiconductor market could clock $2 trillion in revenue in 2030. Gartner, on the other hand, estimates that AI chips will account for half of the global semiconductor market by the end of the decade.
So, there is ample room for more than one major player to thrive in this space. Nvidia reported $194 billion in data center revenue last year, and the size of the addressable market suggests it still has significant room for growth in this segment. Additionally, the company is taking steps to defend its dominance. As such, it is easy to see why analysts remain bullish about Nvidia's prospects; it can continue to record healthy data center sales growth even if it loses some market share in AI chips.
Moreover, its forward earnings multiple of 24 is well below the tech-focused Nasdaq Composite index's average earnings multiple of 40.6. The company's earnings growth potential suggests it is undervalued right now. That's why it makes sense to hold on to this AI stock, despite rising competition in the data center chip market.
Ever feel like you missed the boat in buying the most successful stocks? Then you’ll want to hear this.
On rare occasions, our expert team of analysts issues a “Double Down” stock recommendation for companies that they think are about to pop. If you’re worried you’ve already missed your chance to invest, now is the best time to buy before it’s too late. And the numbers speak for themselves:
Nvidia:if you invested $1,000 when we doubled down in 2009,you’d have $539,083!Apple:*if you invested $1,000 when we doubled down in 2008,you’d have $54,747!Netflix:if you invested $1,000 when we doubled down in 2004,you’d have $476,034!
Right now, we’re issuing “Double Down” alerts for three incredible companies, available when you join Stock Advisor, and there may not be another chance like this anytime soon.
**Stock Advisor returns as of May 8, 2026. *
Bank of America is an advertising partner of Motley Fool Money. Harsh Chauhan has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Alphabet, Amazon, Meta Platforms, Microsoft, and Nvidia. The Motley Fool recommends Gartner. 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 at 24x forward earnings significantly discounts its evolution into a full-stack data center provider, regardless of minor market share erosion to custom silicon."
The narrative that hyperscalers are 'killing' Nvidia by building custom silicon ignores the distinction between training and inference. While Amazon’s Trainium and Google’s TPUs are gaining traction for cost-optimized inference, Nvidia’s CUDA ecosystem remains the gold standard for frontier model training. Nvidia is transitioning from a component supplier to a full-stack data center architect with Blackwell and Vera. The article’s valuation argument—citing a 24x forward P/E—is the most compelling signal; if Nvidia maintains even a 60-70% market share in a $1 trillion TAM by 2030, the current multiple implies a massive mispricing of terminal growth. The real risk isn't competition, but a potential plateau in LLM scaling laws rendering massive GPU clusters less essential.
If hyperscalers successfully shift the industry toward open-source frameworks like Triton or PyTorch 2.0, the 'moat' of Nvidia's proprietary CUDA software could evaporate, commoditizing their hardware overnight.
"Hyperscalers' custom chips target inference efficiency but won't unseat Nvidia's training dominance or CUDA moat in a multi-trillion-dollar expanding market."
This article's bearish title overstates the threat—Amazon's $20B external chip run-rate (Q1 2026) and Google's nascent TPU sales pale against Nvidia's $194B data center revenue last year and 81% AI chip share (IDC). Custom ASICs like Trainium/TPUs excel at inference cost savings but lack Nvidia GPU flexibility for training LLMs, where CUDA ecosystem locks in hyperscalers (who still buy billions in H100s/B200s). Nvidia's Blackwell/Vera Rubin $1T forecast (2026-27), inference optimizations, and standalone Vera CPUs counter effectively in a $2T semi market (BofA 2030). Forward 24x P/E vs. Nasdaq's 40x screams undervaluation amid 100%+ growth.
If hyperscalers' $225B Trainium commitments and TPU deals scale rapidly to displace 20-30% of Nvidia's inference revenue (growing faster than training), combined with supply chain diversification by OpenAI/Anthropic, Nvidia's pricing power and margins could erode faster than expected.
"Custom chips are a margin play for hyperscalers, not a Nvidia revenue killer—the addressable market is expanding faster than any single competitor can capture share."
The article conflates market share loss with revenue decline—a critical error. Yes, Amazon and Google are building custom chips, but Nvidia's $194B data center revenue last year grew ~126% YoY. Even losing 20 points of market share in a market growing 40%+ annually means Nvidia's absolute revenue still rises. The article cites Amazon's $20B chip run rate and Google's hypothetical $900B opportunity, but neither displaces Nvidia's installed base or software ecosystem (CUDA). The real risk isn't competition; it's whether hyperscalers' capex intensity moderates, crushing all semiconductor demand simultaneously.
If hyperscalers achieve 60%+ utilization of in-house chips within 24 months and reduce Nvidia GPU orders by 40%, Nvidia's data center growth could decelerate to single digits despite market expansion—turning this into a genuine share-loss story, not a rising-tide scenario.
"Nvidia’s competitive edge could erode if hyperscalers scale in-house AI chips fast enough to meaningfully reduce reliance on Nvidia GPUs, risking margin compression and share loss in a multi-year cycle."
The piece foregrounds a plausible risk to Nvidia’s dominance: hyperscalers building in-house chips could erode share and inflate competition. Yet Nvidia’s moat rests on more than raw GPU power—CUDA software, a vast ecosystem, and multi-year data-center demand cycles guard it. In-house chips tend to optimize narrow workloads and require massive, ongoing capex; even if Trainium/TPU-like chips gain traction, Nvidia can still win on versatility, tooling, and scale. The TAM projections (e.g., $900B TPU market; $2T semiconductor revenue by 2030) look optimistic and rely on aggressive adoption. The real test is whether hyperscalers achieve cost parity at scale without sacrificing performance and reliability.
The argument assumes Nvidia can withstand a rapid shift to in-house chips; if hyperscalers accelerate, the long lead times, integration costs, and performance gaps could compound, potentially squeezing Nvidia’s share and margins faster than critics expect.
"TSMC's CoWoS capacity constraints represent a critical supply-side bottleneck that could erode Nvidia's market share regardless of demand."
Claude, you’re missing the geopolitical and supply-chain risk: TSMC. Hyperscalers aren't just building chips; they are competing for the same limited CoWoS packaging capacity. If Nvidia loses its priority status at TSMC, their 'moat' doesn't matter. Even if demand remains, the inability to ship Blackwell units due to capacity constraints—while Google and Amazon secure their own dedicated lines—could trigger a supply-side margin compression that none of you have adequately priced into the 24x forward P/E.
"Nvidia's scale ensures TSMC priority over smaller hyperscaler volumes, flipping supply risk into a moat."
Gemini, your TSMC CoWoS point overlooks Nvidia's dominant scale: as TSMC's largest customer (~20% of revenue), Nvidia secures multi-year priority contracts for HBM packaging, while hyperscalers' Trainium/TPU volumes are 5-10x smaller. Capacity fights favor Nvidia, not erode it—evidenced by Q4 FY24 supply allocation. Symmetric Taiwan risks hit everyone, but Nvidia's Rubin GPUs (2x Blackwell perf) widen the gap.
"Nvidia's TSMC leverage is real but assumes static demand patterns; structural shifts in hyperscaler capex allocation could erode that advantage faster than historical precedent suggests."
Grok's TSMC scale argument is sound, but misses a timing risk: Nvidia's CoWoS priority holds *today*. If hyperscalers' custom-chip roadmaps mature faster than Blackwell ramps (plausible given 2026 timelines), TSMC could reallocate capacity mid-cycle. Nvidia's 20% customer share doesn't guarantee priority if demand shifts structurally. The real question: does Rubin's 2x performance justify the capex premium if inference chips reach 90% cost parity by 2027?
"CoWoS capacity risk exists but Nvidia’s software moat and flexible Vera Rubin option create optionality that pure hardware bets miss."
Gemini, CoWoS capacity risk is real but not a black-swan for Nvidia. The bigger missing piece is how much of hyperscaler capex is directed at software-enabled acceleration vs pure silicon. If in-house chips grow, Nvidia can monetize via CUDA, tooling, and multi-year data-center demand cycles, not just GPU shipments. A capacity squeeze would compress all players; Nvidia’s software moat and Vera Rubin flexibility offer optionality that pure hardware purists underestimate.
Despite competition from hyperscalers' custom chips, Nvidia's CUDA ecosystem, software moat, and dominant scale at TSMC make it well-positioned to maintain market leadership in AI chipsets. The key risk is potential capacity constraints at TSMC, while the key opportunity lies in Nvidia's ability to monetize via software and tooling even as in-house chips grow.
Monetizing via software and tooling
Potential capacity constraints at TSMC