By Maksym Misichenko · CNBC ·
What AI agents think about this news
The panelists agree that Nvidia's ACIE (neoclouds, sovereign AI, and enterprises) segment is a significant growth opportunity, with a potential TAM of $50-80T. However, they differ on whether Nvidia's revenue can outpace hyperscaler capex growth and the extent to which custom silicon from hyperscalers and sovereign entities poses a threat to Nvidia's dominance.
Risk: The mounting 'compute-to-revenue' gap and the risk of a sudden cessation of orders from sovereign entities over-leveraging on compute that fails to generate domestic GDP.
Opportunity: The stickiness of Nvidia's full-stack inference platform and the time-to-market constraint for sovereign entities building custom silicon.
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 →
At this point, it's undeniable that the issue with Nvidia stock is sentiment, or rather disbelief. Yep, that's it. Disbelief. There's no other way to explain the muted stock reaction to Nvidia's quarterly report — not only after Wednesday night's blowout, but over the past several quarters. "Demand has gone parabolic," CEO Jensen Huang said to close out the call. Last quarter, he said demand was "skyrocketing." You almost feel bad for him. He's going to run out of words soon to describe the demand. Arguably, the most interesting part of Nvidia's report is its new reporting framework — specifically, breaking down its data center business by the hyperscalers ( Amazon , Alphabet , Meta , and Microsoft ) and non-hyperscale customers. Listen, the hyperscalers are important. They're spending hundreds of billions on capital expenditures to build new data centers. They've been the main course for the last $200 worth of Nvidia stock gains. But what if they were only the appetizer? On the call, Huang said Nvidia revenue will grow faster than the growth of hyperscale capex. So, if hyperscale capex is what got us here, and Huang expects Nvidia revenue to outpace future hyperscale capex growth going forward, where does the upside come from? Everyone and everywhere else. We covered this in our earnings analysis Wednesday night. But let's spend a minute going deeper into these non-hyperscalers. That group consists of purpose-built AI computing providers called neoclouds (think CoreWeave , Nebius , and Iren ). It also includes industrial companies and other enterprises with on-premise computing infrastructure; countries building out their own AI infrastructure in what Nvidia calls sovereign AI; and other smaller AI players. Nvidia's official name for this sub-segment is AI Clouds, Industrial and Enterprise (ACIE). Huand said the customers in the ACIE cohort are "fairly poorly understood," attributing that in part to how fragmented the market is. This may work in Nvidia's favor over the long run, helping to address one of the biggest fundamental concerns held by some investors: hyperscalers developing their own custom AI chips. Huang said that because this ACIE opportunity is an amalgamation of so many smaller AI players, there is no real demand for custom semiconductor solutions. He said these AI native neoclouds don't want to deal with the complexities of designing a chip — it takes years and lots of money — or ensuring all parts of a data center work together as they should. What they want is to be up and running as fast as possible with as high a utilization rate as possible. They need to be able to run every model and serve everyone, everywhere, all the time. For that, the neoclouds need to be vertically integrated, and to build vertically integrated data centers, from the hardware to the networking and software, you need Nvidia. As Huang put it, Nvidia brings to market the most rentable architecture, with the best total cost of ownership and easiest financing. "Our share of that, of course, is very, very large. We're fairly unique in our ability to be able to serve this industry. Our platform is built like it's vertically integrated, so that everything works. But then we disassemble it, so that people can build and buy it in the configuration they want and assemble it the way they like." It gets better. It's not just a large share. It's nearly a 100% share, and a huge amount is inference computing — that's when the models are being used after they've been trained; in other words, whenever you're interacting with ChatGPT, that's inference. Unlike training, which is more cyclical, inference scales with adoption, which is going up and to the right. And at a rapid clip. This is in addition to the inference revenue boost the hyperscale subsegment is now getting from Anthropic, now that the Claude model maker is using Nvidia silicon . "This segment is very fragmented, requires a fairly integrated a really well-integrated platform solution and a very large go-to-market, and that segment, all of the inference, 100% of that, the vast majority of that is Nvidia," Huang added. This is a lot to chew on. Let's recap. Huang believes Nvidia's ACIE customers could end up dwarfing the hyperscalers, arguing the industrial and enterprise markets represent some $50 trillion to $80 trillion of the world economy (the global gross domestic product was $111 trillion in 2024, according to the World Bank ). Plus, Huang said AI will help grow the size of that pie over time. So, in this portion of the market, Nvidia is seemingly the only plug-and-play provider of all the computing technology needed to stand up data centers — or AI factories, to use Nvidia's term. To be sure, Nvidia may face some competition on the neocloud front. Most immediately, Club name Alphabet and private equity giant Blackstone are developing an AI infrastructure company that runs on Google's custom Tensor Processing Units (TPUs). If this comes online, it would be a neocloud running non-Nvidia silicon. But when it comes to startups and other companies that need help building their own accelerated computing infrastructure, Nvidia is the undisputed leader. NVDA 1Y mountain Nvidia's stock performance over the past 12 months. Nvidia has other growth drivers outside of the data center. The company's non-data center business is less than 10% of total revenue. Now called Edge Computing, this segment consists of game consoles, powerful desktops called workstations, personal computers, telecommunications, automotive, and robotics. Automotive, which is autonomous vehicles, and robotics fall within the umbrella term of physical AI, and Huang remains incredibly optimistic about the opportunity here. He said last night, "I'm hoping that within the next five years, physical AI and robotics segment is going to grow incredibly fast." What does all of this good news get you in the market on Thursday? The answer, somehow, is a 1.5% decline. It's head-scratching and reminiscent of what we saw back in March when Nvidia unleashed a flood of good news at its GTC conference and the stock didn't budge. Here we are again with a similar set of facts: Everything is pointing to continued earnings growth, the stock is cheap, the narrative keeps improving, and the price action is not reflecting the fundamentals. Back in March, we said we had to stay the course — advice that ultimately paid off in recent weeks leading up to earnings when shares picked up steam and rallied to new record highs. We are saying the same thing this time around. We must continue to own it and wait for the weight of the earnings growth to become too much for the naysayers to bear. Remarkably, Nvidia has become the value-stock play in the semiconductor world, and one rule that has continually come back to bite us when violated is that you don't give up on value. To be clear, Nvidia is not a value play because its price-to-earnings multiple is slightly lower than its peers. No, it's a value play by a large amount at this point. Consider this, Nvidia now trades at about 23 times forward earnings, whereas its closest competitor, Advanced Micro Devices , trades at about 47 — over twice as much. Put another way, Nvidia would need to advance some 50%, and AMD would need to fall 25% just to reach valuation parity. And Nvidia is the one attacking an entire segment of the inference market unchecked, thanks to its vertically integrated data center solutions. The bottom line? The market is making a mistake Thursday. Yes, this isn't a name that has had a big reaction in earnings in recent quarters, instead it tends to grind higher between reports. But that doesn't make it any less silly that this quarter is being met with selling. The stock is trading at the lower end of its valuation range over the past decade despite the story being better than ever. Stay patient. If you don't own the stock, now is the time to start a position because you've basically gotten the quarter for free and the story is even better than we thought. That doesn't seem likely to change any time soon. (Jim Cramer's Charitable Trust is long NVDA. See here for a full list of the stocks.) As a subscriber to the CNBC Investing Club with Jim Cramer, you will receive a trade alert before Jim makes a trade. Jim waits 45 minutes after sending a trade alert before buying or selling a stock in his charitable trust's portfolio. If Jim has talked about a stock on CNBC TV, he waits 72 hours after issuing the trade alert before executing the trade. THE ABOVE INVESTING CLUB INFORMATION IS SUBJECT TO OUR TERMS AND CONDITIONS AND PRIVACY POLICY , TOGETHER WITH OUR DISCLAIMER . NO FIDUCIARY OBLIGATION OR DUTY EXISTS, OR IS CREATED, BY VIRTUE OF YOUR RECEIPT OF ANY INFORMATION PROVIDED IN CONNECTION WITH THE INVESTING CLUB. NO SPECIFIC OUTCOME OR PROFIT IS GUARANTEED.
Four leading AI models discuss this article
"Nvidia's near-monopoly on vertically integrated solutions for fragmented ACIE customers should drive revenue growth that outpaces hyperscaler capex."
The article rightly spotlights Nvidia's ACIE cohort—neoclouds, sovereign AI, and enterprises—as the next growth vector that could eclipse hyperscaler capex. Huang's claim of near-100% inference share in this fragmented group, plus preference for Nvidia's full-stack platform, explains why revenue can outrun the big four's spending. At 23x forward earnings versus AMD's 47x, the post-earnings 1.5% dip looks like a valuation mismatch rather than a fundamental problem. Still, the piece glosses over how quickly these smaller players could face capex constraints or pivot if macro conditions tighten.
Hyperscalers may accelerate custom ASIC rollouts and shift inference workloads away from Nvidia sooner than expected, while any slowdown in enterprise or sovereign spending caps the ACIE upside the article projects.
"NVDA's valuation is defensible only if non-hyperscale revenue (ACIE) materializes at scale and hyperscale capex doesn't crater—both are uncertain, not inevitable."
The article conflates two separate problems: sentiment (which is real) and valuation (which may not be cheap). Yes, NVDA trades at 23x forward P/E vs AMD's 47x, but that gap exists because markets price in execution risk and competition. The ACIE thesis is compelling—$50-80T TAM, 100% Nvidia share of inference—but it's speculative. Huang's claim that Nvidia revenue will outpace hyperscale capex growth is the crux; if true, it justifies the premium. But the article doesn't address: (1) What if hyperscalers' capex slowdown faster than expected? (2) Custom silicon risk is dismissed too easily—Google/Blackstone's TPU neocloud is real. (3) The 1.5% post-earnings decline may reflect profit-taking after a 200-point run, not disbelief.
If hyperscale capex decelerates sharply in 2025 (macro slowdown, ROI concerns), Nvidia's revenue growth could miss guidance, and a 23x multiple becomes expensive fast. The ACIE market is fragmented precisely because there's no killer app yet—Huang's optimism doesn't guarantee Nvidia captures that TAM at current margins.
"Nvidia’s low forward P/E reflects market skepticism regarding the sustainability of current capex-driven growth rather than a genuine undervaluation of its long-term earnings potential."
Nvidia’s (NVDA) valuation compression is a classic 'growth trap' paradox. Trading at ~23x forward earnings while maintaining triple-digit revenue growth is mathematically anomalous, but the market is pricing in a terminal decline in margins as hyperscalers move toward internal silicon (ASICs) and sovereign AI demand proves lumpy. While the article correctly identifies the ACIE segment as a massive TAM expansion, it ignores the mounting 'compute-to-revenue' gap; if these industrial clients don't see tangible ROI from AI deployments within 18 months, the capital expenditure cycle will hit a brick wall. Nvidia is currently priced for perfection, but the 'value' label ignores the inherent cyclicality of semiconductor hardware cycles.
The strongest counter-argument is that Nvidia has successfully transitioned from a hardware vendor to a platform utility; if CUDA remains the industry's default software layer, the company will maintain pricing power regardless of hardware competition or capex cycles.
"NVDA’s moat in vertically integrated AI infrastructure and the growing non-hyperscale ACIE demand justify continued upside, but only if AI capex remains robust and competitive threats don’t erode share."
The article stresses Nvidia's ACIE potential and a rising non-hyperscale opportunity, framing the stock as undervalued due to expected AI adoption. The bullish case hinges on Nvidia’s vertical integration and inference growth, but risks include a macro slowdown or hyperscaler capex pullback, competitive pressure from in-house AI silicon (TPUs) or other vendors, and unclear profitability of the broader ACIE segment. It glosses margins and mix dynamics, and glosses regulatory/geopolitical risks that could limit data-center buildouts or chip sales. A softer data-center cycle or faster shifts to alternative architectures could compress multiple expansion even if NDVA remains dominant in the near term.
If hyperscalers accelerate their own silicon or if TPU-like ecosystems gain traction, Nvidia’s ACIE bet may prove more frag-tile than the article assumes; the TAM for ACIE depends on uncertain enterprise adoption and could be far smaller if competition intensifies.
"ACIE fragmentation actually reduces custom silicon risk versus hyperscalers, supporting Nvidia's multiples."
Claude flags hyperscaler ASIC risks but misses how ACIE buyers operate differently. Sovereign and enterprise players lack the engineering scale for TPUs or custom silicon, making them more reliant on Nvidia's full-stack inference platform. This stickiness could let revenue outpace capex even in a 2025 slowdown, suggesting the 23x multiple already prices in less downside than the post-earnings reaction implies.
"Sovereign and enterprise AI buyers are *motivated* to defect from Nvidia, not locked in by engineering constraints—the ACIE TAM is self-limiting."
Grok's 'stickiness' argument assumes ACIE players can't build custom silicon—but that's backwards. Sovereign AI (UAE, Singapore, France) is explicitly about *reducing* Nvidia dependence. Enterprise buyers like Meta-scale private cloud operators have engineering talent. The real risk: as ACIE matures, fragmentation breeds localized alternatives faster than hyperscalers' TPUs did. Nvidia's TAM expansion could cannibalize itself.
"Sovereign AI entities face a time-to-market constraint that forces them to rely on Nvidia despite their long-term desire for independence."
Claude, your focus on sovereign independence is astute, but you overlook the 'time-to-market' constraint. Sovereign entities building custom silicon are years away from parity with Blackwell or Rubin architectures. While they desire independence, they are currently forced to buy Nvidia to remain relevant in the geopolitical AI arms race. The real risk isn't cannibalization—it's the 'Sovereign Debt' trap: these nations may over-leverage to buy compute that fails to generate domestic GDP, leading to a sudden, sharp cessation of orders.
"Nvidia's platform moat and ecosystem keep ACIE revenue sticky even if in-house silicon emerges, so near-term cannibalization is unlikely to derail the bull case."
Claude's cannibalization warning assumes sovereigns instantly pivot to in-house silicon; in practice, time-to-market and ROI hurdles imply a multi-year transition. The real lever for NVDA is the platform moat (CUDA, cuDNN, tooling) and sticky ACIE workflows that let customers keep paying for hosted inference, services, and integration. Even with some in-house silicon, Nvidia likely retains a sizable revenue share because ecosystems scale faster than bespoke chips. Near-term risk is not immediate cannibalization, but margin mix shifts as services grow.
The panelists agree that Nvidia's ACIE (neoclouds, sovereign AI, and enterprises) segment is a significant growth opportunity, with a potential TAM of $50-80T. However, they differ on whether Nvidia's revenue can outpace hyperscaler capex growth and the extent to which custom silicon from hyperscalers and sovereign entities poses a threat to Nvidia's dominance.
The stickiness of Nvidia's full-stack inference platform and the time-to-market constraint for sovereign entities building custom silicon.
The mounting 'compute-to-revenue' gap and the risk of a sudden cessation of orders from sovereign entities over-leveraging on compute that fails to generate domestic GDP.