End of the ghost story? Zuckerberg denies Meta has excess computing capacity

富途寰球私享匯

joined discussion ·

Sector Deep Dive | Computing Power Leasing Special: Overseas Markets Under Pressure and Diverging, Domestic Supply-Demand Remains Tight

AI Compute Leasing Spotlight: Shifting Global Landscape and Domestic Supply-Demand Opportunities

💡 Core insight

1. The global computing power leasing industry is in a strong expansion cycle, with overseas NeoCloud leaders deeply integrated into NVIDIA's ecosystem, $Meta Platforms (META.US)$Their entry may temporarily suppress sector stock prices, but the medium- to long-term growth thesis remains intact; investors can consider accumulating positions on pullbacks.。

2. Among the two pure-play U.S. listed companies, $NEBIUS (NBIS.US)$the business model combining self-built data centers and in-house server development is healthier, with margins expected to continue rising, outperforming $CoreWeave (CRWV.US)$ the high-leverage expansion strategy.

3. High-end supply in China's AI compute leasing market is severely insufficient, prompting cloud providers to accelerate their shift from self-building to leasing,Combined advantages in GPU sourcing, client base, financing capabilities, and AI data center (AIDC) delivery capacity continue to reinforce the competitive moats of leading players.。

4. Key A-share names to watch $Range Intelligent Computing Technology Group (300442.SZ)$ (Stable and low valuation, with expectations for computing power leasing not yet fully priced in) and Xiechuang Data (high growth, high elasticity).The sector is currently facing dual headwinds from Meta-related events and the mid-year earnings season. After the recent pullback, consider establishing small positions, prioritizing domestic names.。

🔍 I. Structure of the Computing Power Leasing Industry Chain (Overseas Example)

Computing power leasing is essentially the "power plant" of the AI era, positioned in the midstream of the AI computing power industry chain and currently in a strong expansion phase. The industry chain has a clear three-tier structure: upstream partners include chip suppliers like NVIDIA and hardware vendors providing servers, networking equipment, and other supporting infrastructure to deliver foundational computing resources; the midstream consists of $CoreWeave (CRWV.US)$ 、 $NEBIUS (NBIS.US)$ the midstream is represented by NeoCloud companies such as CoreWeave, which build full-stack AI cloud platforms through large-scale hardware procurement; $Microsoft (MSFT.US)$ 、 $Meta Platforms (META.US)$ and the downstream delivers computing power services to leading AI labs and tech firms such as OpenAI, Anthropic, and others.

Chart 1: Computing Power Leasing Industry Chain | Source: Compiled by Futu Securities

II. Key Players and Evolving Landscape

The overseas computing power leasing market shows a trend toward concentration among top players, primarily comprising cloud providers and NeoCloud leaders, with their roles undergoing dynamic evolution.The four major cloud providers $Alphabet-A (GOOGL.US)$ 、 $Microsoft (MSFT.US)$ 、 $Amazon (AMZN.US)$ 、 $Meta Platforms (META.US)$ It is a core demander of computing power, with combined capital expenditure guidance for 2026 exceeding USD 70 billion. Meanwhile, Meta is establishing its 'Meta Compute' business, planning to lease out surplus AI computing capacity externally, thereby transitioning from a pure demand-side player to a supplier—a development introducing new variables into the market landscape.

Among NeoCloud leaders, $CoreWeave (CRWV.US)$ and $NEBIUS (NBIS.US)$ is deeply integrated into NVIDIA’s ecosystem and holds significant long-term, large-volume contracts with major clients. CoreWeave has secured a collaboration order from Meta, USD 21 billionand NEBIUS has clinched a Meta infrastructure agreement worth up to USD 27 billion, both having already established notable first-mover advantages among major clients.

The overseas AI computing rental industry commonly adopts a capital-intensive operational model characterized by 'pure AI cloud platforms + long-term contracts with large clients + high-leverage expansion': companies pre-build or lease data centers and procure core hardware—such as NVIDIA GPUs—at scale, then sign long-term, large-value contracts with top-tier clients to lock in revenue and recoup upfront investments. The advantage of this model lies in its extremely high revenue visibility, but it also entails heavy reliance on a concentrated client base and substantial depreciation and interest burdens. Taking leading player CoreWeave as an example, its net interest expense for Q1 2026 reached USD 536 million, , net losswith total interest expenses amounting to USD 740 million, highlighting the cash flow challenges inherent in high-leverage expansion.

🔍 III. Key Stocks (U.S. Equities): CRWV vs. NBIS

There are two pure-play U.S.-listed companies in the computing power leasing space: $CoreWeave (CRWV.US)$and $NEBIUS (NBIS.US)$ . The two differ significantly in their operating models. Since its IPO in 2025, CoreWeave has been defined by aggressive expansion: approximately 70% of its infrastructure is not self-owned, relying heavily on third-party colocation providers for facilities, transformers, and liquid-cooling infrastructure, incurring substantial 'interconnection and colocation premiums' that ultimately result in higher costs than owning heavy assets outright.While this model enabled faster initial expansion and more contracts signed early on, it has also planted seeds of concern on the income statement.

Nebius, by contrast, has taken a more prudent path: it owns and operates its own data centers in Europe and the United States and has fully transitioned to in-house designed servers, effectively eliminating interconnection and colocation fees.NBIS’s model appears healthier over the medium to long term, with controllable depreciation expenses and steadily rising profit margins expected in the future.Industry research indicates that computing power leasing prices will continue to rise in the second half of the year, with an expected increase ofapproximately 20%by year-end. If supply-demand imbalances worsen further,prices could even rise by as much as 40%.; After Meta completes its GPU deployment, prices are expected to gradually stabilize.NBIS’s overall profit margin is expected to rise from the current range of 35%–40% to around 45% by year-end., with token leasing revenue for inference services continuing to increase as a share of total revenue. This business segment carries a gross margin of approximately 75% and is growing the fastest.

Neocloud's revenue primarily comes from two sources.The first is bare-metal leasing,which typically accounts for over 75% of revenue and operates under take-or-pay contracts—clients must pay in full regardless of actual usage. This model ensures stable income but faces intense commoditized competition and significant pricing pressure.The second is token-based leasing,i.e., GPU-as-a-service, which includes not only raw computing power but also value-added services such as software premiums, solution architecture, and cluster optimization. NEBIUS acquired Clarifai’s core team and intellectual property in May, strengthening its software capabilities,potentially raising the gross margin on token leasing from 75% to over 85%.。Combined with market rumors of Google Cloud’s computing capacity shortage and NBIS’s previously announced leasing deal with SpaceX, the company still has opportunities to expand its roster of top-tier clients.

Meta’s entry into the compute leasing market is expected to drive down GPU leasing prices, with an optimistic scenario pointing to an impact starting in late Q3 to early Q4. As it takes time to decouple GPUs from Meta’s primary cloud infrastructure for external customer use,Meta will likely continue to weigh on U.S. listed compute-leasing companies’ stock prices in the near term, but after recent corrections, their medium- to long-term growth potential remains strong.

Chart 2: NEBIUS vs. CoreWeave Comparison | Source: Company Financial Reports, Compiled by Futu Securities

Chart 3: NEBIUS EBITDA Margin Reversal and Valuation Multiple Changes | Source: Morgan Stanley, Compiled by Futu Securities

🔍 IV. China's Computing Power Leasing Market: Supply-Demand Dynamics and Investment Logic

4.1 Demand Side

Domestic demand for computing power is accelerating simultaneously on both the training and inference fronts.Downstream large-model training cycles are shortening, creating urgent demand for high-end training chips; meanwhile, AI applications on the inference side are becoming ubiquitous across China, with domestic token-based applicationsposting a weekly growth rate of 80%,driving similarly rapid growth in demand for inference chips.$Z.AI (02513.HK)$ and $MINIMAX-W (00100.HK)$ A major refinancing round conducted in July directly targeted computing power procurement and infrastructure development, sending a clear signal from the demand side.

Cloud providers are also undergoing a structural shift in their computing power procurement models: by 2026, leading cloud vendors such as Alibaba and ByteDancewill accelerate their transition from self-built computing infrastructure to leasing models.Two core drivers underpin this shift: first, regulatory compliance—Alibaba, as a U.S.-listed company, seeks to avoid exposure to U.S. Bureau of Industry and Security (BIS) review risks and faces constraints in acquiring high-end chips itself; second, capital efficiency—cloud providers prefer asset-light operations, opting to let third parties bear the capital expenditure of hardware procurement while focusing on serving public cloud customers and leveraging their platform advantages.

4.2 Supply Side

Currently, high-end computing power supply is severely insufficient, with virtually no idle computing clusters exceeding a scale of 16–32 servers being released onto the market.A cluster size of 16–32 servers represents the threshold for training large models with practical utility. The industry as a whole faces a significant supply-demand imbalance,exhibiting pronounced seller's market characteristics.。

4.3 Classification of Chinese Computing Power Leasing Companies

Domestic computing power leasing companies can generally be divided into three categories.

The first category isAIDC/IDC heavy-asset operators, represented by companies such as Runze Technology and Data Harbor. Their advantages lie in land, power, data center resources, and customer stickiness, while their profit realization pace is influenced by rack utilization rates and delivery cycles.

The second category isHeavy-asset GPU lessors, represented by companies such as Xiechuang Data, Litong Electronics, and Hongjing Technology. These firms are characterized by high earnings elasticity and rapid profit realization, but they rely heavily on GPU supply sources and financing capabilities.

The third category isscheduling/platform-type companies, such as Parallel Tech and UCloud, which face relatively manageable capital expenditure pressure but lack proprietary IDCs and high-end GPU resources, and directly compete with MaaS offerings from large players like Alibaba and Zhipu AI,posing a risk of being outcompeted by superior rivals。

4.4 Industry Barriers

Access to GPU supply channels, core clients, and financing capabilities have historically been the three key barriers in the computing power leasing industry.In the new environment of 2026, a fourth dimension—AI data center (AIDC) delivery capability—has significantly increased in importance. Delivering large-scale AI clusters now involves more than just procuring chips; it requires integrated coordination across liquid cooling, power distribution, low-latency networking, and data center deployment.Leading companies with end-to-end delivery capabilities are continuously strengthening their competitive moats.。

Following the tightening of BIS regulations and stricter restrictions on overseas transshipment, participants relying on gray or borderline channels are being rapidly phased out,market share is accelerating toward industry leaders。In a seller's market, resource barriers will further amplify rather than dilute, which is also why market sentiment toward leading companies in 2026 is clearly more favorable than toward smaller and mid-sized players.

🔍 V. Key Investment Targets (A-Share Market)

5.1 $Range Intelligent Computing Technology Group (300442.SZ)$

$Range Intelligent Computing Technology Group (300442.SZ)$ It is a stable choice for medium- to long-term allocation and the undisputed domestic leader in AIDC. The company generates revenue from two main segments: traditional IDC services (accounting for 56% of revenue in 2025) and AIDC services (44%, including AIDC operations and GPU-as-a-Service), serving top-tier clients such as ByteDance, Tencent, and Alibaba. The company has deployed nine AI infrastructure clusters globally, with a total planned computing capacity of approximately 6 GW and an operational capacity of around 750 MW, backed by industry-leading3.2 GW of approved energy quotas, most of which are located in first-tier cities, highlighting their scarcity. Leveraging its competitive edge in large-scale liquid-cooled data center operations and access to scarce, low-cost power resources, the companyhas maintained an average gross margin of 49% from 2018 to 2025, 15 percentage points above the industry average。

The market holds clear optimistic expectations for Runze’s interim report: the first half is expected to delivernearly RMB 2 billionPure computing power leasing revenue,Net profit growth exceeding 30%，Full-year non-GAAP net profit growth of 30%–50%。The current market prices Runze more as a 'high-demand infrastructure' asset, with limited valuation contribution from computing power leasing expectations. Its TTM P/E ratio is approximately 20x, significantly lower than the 70x+ valuations of pure-play computing power leasing peers, suggesting notable room for upward re-rating.

Chart 4: Runze Technology’s IDC-to-AIDC Business Transformation and Capacity Expansion Plan | Source: Company Annual Report, HSBC, Compiled by Futu Securities

5.2 $Sharetronic Data Technology (300857.SZ)$

$Sharetronic Data Technology (300857.SZ)$ It is China’s largest company in the computing power leasing sector by both revenue and profit scale, enjoying significant economies of scale and earnings visibility. Q1 2026 net profit beat expectations. Looking ahead to Q2, as new clusters are progressively delivered, computing leasing profitability is expected to improve further on a sequential basis. Since 2026, domestic high-end GPU leasing priceshave cumulatively risen by approximately 40%, but most Q1 contracts were legacy low-price orders signed in 2025, so pricing elasticity has not yet been fully reflected;Newly delivered clusters in Q2 will be contracted at current market rates, allowing the profit upside from higher unit prices to materialize predominantly in the second quarter.。

The company’s storage business serves as a stable profit base, including consumer- and enterprise-grade SSDs. It jointly won, together with SanDisk, Alibaba Cloud’s RMB multi-billion enterprise SSD framework procurement tender, targeting a share of Alibaba Cloud’s enterprise storage procurement volume in 2026 of20%Tencent and ByteDance, as core major clients in computing power leasing, are also procuring enterprise-grade SSDs and DDR memory modules from the company to complement their AI computing clusters, establishing an integrated 'computing plus storage' bundled delivery model. Over the medium to long term, leveraging AI infrastructure optimization and scheduling capabilities developed with leading clients, the Token Factory business model is expected to further boost profit margins.Taking Zhipu's estimated computing power procurement scale of approximately RMB 20 billion in 2026 as an example, Xiechuang Data has already secured orders worth around RMB 4 billion, reflecting strong client stickiness among top-tier customers.Overall,Xiechuang Data is a relatively high-valuation, higher-beta investment targetsuitable for investors with a higher risk appetite.

Chart 5: Xiechuang Data’s Profitability Improvement and Business Mix | Source: Company Annual Report, Guotai Junan Securities, Compiled by Futu Securities

💡 Summary

Currently, U.S.- and China-listed computing power leasing companies trade at approximately20x+ P/E multiples for 2027,and given the trends of rising fixed asset investments, execution of long-term contracts, and increasing computing power pricing in 2027,both 2027 EPS and valuation multiples have room for further upside.The Meta incident, combined with the mid-year earnings season, has caused a pullback in the computing power leasing sector.After the pullback, small positions could be considered for entry, with domestic-listed stocks given priority.。

⚠️ Risk Warning

Commercialization of AI applications has fallen short of expectations.

Token usage and paid conversion rates have underperformed expectations.

[Investment Advisory Information]

Yang Yi, Licensed Representative, CE No.: BUR210

[Disclaimer]

This report is prepared by Futu Securities International (Hong Kong) Limited (“Futu Securities”). Without prior written consent from Futu Securities, this report and the information contained herein may not be (i) reproduced, copied, or stored in any form, or (ii) directly or indirectly distributed or transmitted to any other person for any purpose. The information in this report is derived from sources that Futu Securities believes to be accurate and reliable as of the date of publication. However, this report is not intended to contain all information necessary for investor decision-making and may be subject to delays, disruptions, or interceptions in transmission. Futu Securities makes no express or implied representation or warranty as to the adequacy, accuracy, completeness, reliability, or fairness of such information or opinions. Accordingly, Futu Securities and its affiliates (collectively, the “Futu Group”) shall not be liable for any losses of any kind (including but not limited to direct, indirect, or consequential losses) arising from any actions taken by third parties relying on the content of this report. The views, recommendations, suggestions, and opinions expressed in this report do not necessarily reflect those of Futu Securities or its affiliates and are subject to change without notice. Futu Securities undertakes no obligation to update any information or opinions contained herein. This report is provided for general informational purposes only and is intended solely for general reading by clients of Futu Securities. It does not take into account the specific investment objectives, financial situation, or particular needs of any individual recipient. Nothing in this report constitutes or should be construed as an offer, recommendation, or solicitation by any member of the Futu Group to buy or sell any securities, investments, or other financial instruments, nor should it be interpreted as an offer or invitation to purchase or sell securities. Any decision to purchase securities mentioned in this report should be based on publicly available information, including relevant prospectuses. The products referenced in this report may not be suitable for all investors, and readers should fully consider relevant factors and seek professional advice before making any investment decisions. This report is provided to recipients on the assumption that they are capable of independently evaluating investment risks and exercising independent judgment in making investment decisions. In certain jurisdictions or countries, the distribution, issuance, or use of this report may violate local laws, regulations, rules, or other registration or licensing requirements. This report is not intended for distribution to, or use by, any person or entity in such jurisdictions or countries. Hong Kong investors with questions regarding Futu Securities research reports should contact Futu Securities directly. The Central Entity (CE) number of the author’s SFC license is disclosed next to the author’s name on the cover page of this report. The analyst primarily responsible for preparing this report confirms that: (i) the views expressed in this report accurately reflect his/her personal views about the subject companies; and (ii) his/her compensation, past, present, or future, directly or indirectly, is not linked in any way to the specific recommendations or views expressed in this report. The analyst also confirms that neither he/she nor any related party has traded in the securities of the subject companies or their related securities within 30 days prior to the publication of this report or within three business days after its publication. Neither the analyst nor any related party serves as a senior officer of the subject companies or holds any financial interest in them. Futu Securities does not hold a financial interest of 1% or more of the market capitalization of the subject company and has had no investment banking relationship with the company in the past 12 months. Employees of Futu Securities are not employees of the subject company.

212K Views