The AI ‘neoclouds’ buzz is growing on Wall Street, but risks are bubbling
Investors are raving about “neoclouds” and what their emergence means for the artificial intelligence boom. However, some warning signs are starting to flash. Neoclouds are companies specializing in AI cloud computing. Unlike traditional hyperscalers such as Amazon and Microsoft, these bespoke companies are viewed as attractive alternatives as they install and manage AI hardware and software in a faster and more tailored, cost-effective and potentially more energy-efficient way for clients. CoreWeave , which had its IPO in late March, is the largest publicly traded neocloud. Nebius is an up-and-coming name. A swath of private neocloud companies such as Lambda and Crusoe are attracting venture capital funding and working on high-profile AI contracts. Nvidia has taken an interest in these AI-native cloud startups as well. The chipmaking giant has increased its investments in CoreWeave and neocloud startups while also developing its own cloud business amid increasing hyperscaler competition. But while analysts like the niche neocloud model and their market share potential in the scaling AI cloud market, worries about their longer-term outlook are beginning to crop up. Risks they see are in these companies’ high levels of capex spending, lack of differentiation and overall lofty AI valuations. UBS Global Wealth Management strategist Sundeep Gantori views neoclouds as a high-risk, potentially high-reward investment. “Last year, [neoclouds] were very small, but now it’s something that you cannot ignore in terms of size. Obviously, the revenues from those companies are not on par with the capital spending that they’re doing,” Gantori told CNBC. The problem, however, is these companies are spending so much on capex that it hurts their margins, he said. “They’re spending much more, but it’s probably an investment into the future,” Gantori added. UBS expects AI capex beyond the big four tech giants — Amazon , Microsoft , Meta Platforms and Alphabet — to grow 85% year over year and hit $150 billion in 2025. That latter estimate includes robust spending from China, neoclouds and other hyperscalers and enterprise and sovereign cloud providers. Looking at overall market opportunity, Goldman Sachs takes a longer point of view on the technology. The firm estimates the total addressable market of the GPU-as-a-Service and AI cloud market by 2030 will be roughly $267 billion — with neoclouds comprising 20% share and hyperscalers 80%. Close ties between Nvidia and neoclouds appear risky Nvidia has strategically backed neoclouds CoreWeave, Lambda, Crusoe Energy and Nebius — a move that has popularized the space while boosting the chipmaker’s own business amid increasing hyperscaler competition. Some analysts question the ultimate cost of this relationship. “Nvidia has helped the rapid growth of several new GPU-focused clouds by directly investing in them, providing them large allocations of GPUs and publicity by various means,” HSBC analyst Abhishek Shukla said in a Thursday note. “Having a large number of GPU-focused cloud providers improves the competitive position of Nvidia, as the margin that the cloud providers can charge declines and a bigger share of the value is captured by Nvidia and the end-customer.” Shukla is concerned about a couple of key financial risks neoclouds face, namely that they already generate low returns from their investments and could see declining pricing power as more GPUs become available. More importantly, he believes neoclouds could lose their business if hyperscaler customers eventually source GPUs directly from Nvidia. For example, Shukla said that CoreWeave may only see business from Microsoft as long as there is a shortage of GPUs and as long as Nvidia favors CoreWeave while it allocates new GPUs. “Once this phase is over, Microsoft could purchase GPUs directly from Nvidia … we see significant risk of key customers Microsoft and Open AI disintermediating CoreWeave and dealing directly with Nvidia,” he said. “We assess that more than 50% of debt financing available to CoreWeave is offered by OEMs (includes Nvidia). In our opinion, Nvidia benefits from this arrangement only if CoreWeave is leasing its GPUs to customers and creating more competition in the GPU leasing market, which helps Nvidia,” Shukla said. Neoclouds remain a substantial part of Nvidia’s longer-term growth opportunity. According to some analysts, Nvidia powering neoclouds, along with its own DGX platform for AI development, allows the company to generate revenue from large-scale AI inference workloads and not just its GPUs. “NVDA has shifted its strategic posture with regard to Neoclouds materially the last nine months and that they now view them as their de-facto data center arm,” Loop Capital analyst Ananda Baruah wrote in a recent note, adding that “NVDA’s “invigorated” Neocloud strategy is dovetailing into an inferencing strategy.” Wall Street views on top neoclouds Cloud infrastructure companies CoreWeave and Nebius are the key neocloud players in the market right now. Europe-based Nebius appears to be the better-liked play of the two. The company has four buy ratings from Wall Street analysts covering the name, according to LSEG, and was recently brought into the limelight by a bullish rating from Goldman Sachs analyst Alexander Duval. Duval initiated coverage of Nebius with a buy rating and 12-month price target of $68, which implies 25% upside. “Our rating reflects our more-positive-than-consensus view on the growth of the AI Neocloud market, as the Generative AI market continues to expand rapidly and AI startups/enterprises seek specialized capacity alongside that provided by hyperscalers,” Duval said in a July 14 note to clients. Duval said Nebius stands to benefit from capital allocation from future funding rounds, more customers and capacity expansion driving revenue and asset utilization, as well as deals with large AI labs and companies improving GPU utilization. “We see Nebius as especially well-positioned in light of its full stack software offering, cost advantages and demonstrable expertise at scale,” he said. CoreWeave is the largest publicly traded neocloud by market cap, meanwhile, with a valuation of more than $62 billion. Interest in CoreWeave is still high, but the AI stock’s frothy valuation has garnered speculation on how far the stock can really go. Shares have skyrocketed more than 200% since the company’s public market debut in the spring. However, the stock has tumbled 30% in the past month. CoreWeave’s recent slide came after the company announced plans to acquire data center infrastructure company and longtime partner Core Scientific in an all-stock transition worth roughly $9 billion. The deal could serve as a major test to how Wall Street views parts of the AI glitz and glamour. Worries about the transaction being in stock, and about CoreWeave’s sky-high valuation have been a drag on both stocks. On top of that, the consensus price target of $93.71 implies shares could decline nearly 27%, LSEG data shows. CRWV 1Y mountain CoreWeave stock performance over the past year. HSBC’s Shukla initiated the stock with a reduce rating and $32 price target, saying CoreWeave shares are significantly overvalued. Shukla noted that CoreWeave’s revenue comes primarily from Microsoft and Open AI, who use their own software, which he said “dilutes CRWV’s value proposition.” CoreWeave’s liquidity looks stretched and, moreover, the company faces a high cost of borrowing, he added. “Compared with the likes of Microsoft, AI model providers have lower liquidity and weaker credit profiles, which creates its own set of challenges,” Shukla wrote in a note to clients. “CoreWeave’s main business of leasing GPUs is getting commoditized and offers inferior returns.” Mizuho Securities and Stifel each downgraded CoreWeave to hold earlier this month, while Citigroup on July 8 placed a downside catalyst watch on the stock. CFRA analyst Angelo Zino, who in early June initiated CoreWeave with a hold rating and 12-month target price of $180, shares concerns about the stock’s valuation. However, Zino anticipates upside to CoreWeave’s bookings as the company’s customer base broadens and inference demand grows. He likes CoreWeave’s revenue backlog, which he said offers multi-year visibility, and estimates CoreWeave will boast an operating profit in 2026. “We’re very positive in terms of the near-term outlook on a stock, but we just think that there are better, more affordable ways from a valuation perspective to play the overarching theme of compute,” Zino told CNBC. “Some neoclouds look good for now, but there are definitely storm clouds that you could potentially see in the future.” The neocloud universe includes a slew of private AI infrastructure startups, including Lambda, Covalent and Crusoe Energy. Crusoe, which uses flare gas to build data centers, was tapped by Oracle in March to build a data center campus in Texas for Open AI’s $500 billion Stargate project, with buildings expected to eventually include 400,000 Nvidia Blackwell GPUs across the campus. Crusoe pivoted its business to AI from crypto as CoreWeave did. Lambda, which provides Nvidia GPUs at competitive rates, provides cloud services and software for training and deploying AI models.
