Author: Junxi Recently, the global AI infrastructure sector has undergone drastic changes. First, Meta announced plans to lease out some of its idle AI computing power, triggering severe volatility in the capital markets. Subsequently, SoftBank Group officially established a new company, SB Neo, making a massive foray into the U.S. computing power leasing market. Meanwhile, the veteran private equity firm Blackstone Group suddenly halted its plan to invest over $100 billion in the world's largest data center project, while Microsoft abandoned a $3 billion cloud computing power leasing agreement with Oracle due to security concerns.
On one hand, tech giants are entering the market to "sell computing power"; on the other hand, hundred-billion-dollar infrastructure projects are hitting the emergency brakes. This series of seemingly contradictory moves has prompted the market to exclaim: Is AI computing power already in oversupply? Is the investment bubble in computing infrastructure about to burst?
An in-depth analysis from the perspective of the semiconductor industry reveals that what we are seeing is not simply "computing power oversupply," but a profound reconstruction of the development logic of the AI industry. The era of "barbaric growth" over the past two years—characterized by spending at all costs and racing to claim territory—is drawing to a close, and competition in AI infrastructure is fully entering a new phase where "efficiency is king."
01. Are Giants Opting to Monetize Computing Power? In July 2026, Meta announced plans to launch a cloud infrastructure business, selling AI computing power and model access to external customers. Following the news, Meta's stock price surged nearly 9% in a single day, with its market capitalization skyrocketing by approximately $127 billion. However, the AI computing power supply chain faced collective pressure—shares of computing power leasing newcomers like CoreWeave and Nebius plummeted by over 13%, and memory chip giants Micron, SK Hynix, and Samsung Electronics all recorded significant declines.
The market's initial reaction was panic: if even Meta cannot absorb its own GPUs, it indicates that computing power is in oversupply.
However, this linear thinking overlooks the particularity of AI computing power assets and the true intentions of the giants. Meta's leasing behavior is essentially an upgrade in asset operational efficiency, rather than a signal that demand has peaked.
In 2026, Meta's capital expenditure guidance is as high as $125 billion to $145 billion, the vast majority of which is directed toward data centers and GPU procurement. To date, Meta has cumulatively committed $183 billion to AI infrastructure. As a company that relies on advertising for 98% of its revenue, annual investments of hundreds of billions of dollars are accumulated into massive computing clusters, but the open-source Llama model itself does not generate direct revenue. Monetizing previous-generation computing power or temporarily idle resources externally can not only directly amortize depreciation and O&M costs but also represents a crucial step in transforming GPU clusters from "pure cost centers" into "revenue-generating assets." Morgan Stanley estimates that if Meta leases out 250MW of computing power for a year, it could generate approximately $10 billion in revenue.
This is not Meta's original idea. Previously, Elon Musk's xAI had already successfully leased out the computing power of its Colossus supercomputing cluster on an astonishing scale. According to multiple media reports, Anthropic leased the entire capacity of Colossus 1—approximately 220,000 Nvidia GPUs—paying up to $1.25 billion per month in rent, with the contract running until May 2029, totaling approximately $40 billion; Google is also paying $920 million per month to lease bridging computing power to make up for the shortfall in its own data center construction progress. These two deals alone generate over $2.1 billion in monthly cash flow for xAI, a SpaceX subsidiary. According to institutional estimates, based on this monthly rent level, the implied return on investment indicates the full capital expenditure can be recouped in roughly two years.
SoftBank Group's entry further confirms the attractiveness of this track. On July 2, SoftBank announced the establishment of SB Neo, planning to launch cloud services based on Nvidia's latest GPUs for U.S. enterprises in fiscal 2027. The goal is to build 10GW-scale AI data center infrastructure, with the initial deployment site being an 800MW facility in Ohio. To fund this expansion, SoftBank is securing a $10 billion loan, pledging its OpenAI shares as collateral.
From Meta to xAI and then to SoftBank, giants are transforming into "computing power landlords" not because they no longer need computing power, but because they must find new paths for investment returns against the backdrop of high capital expenditures on computing power. As TF Securities pointed out: "Meta doing AI cloud does not equal admitting comprehensive GPU oversupply. This is not the end of the AI capital expenditure trade, but the evolution of the business model from purely cash-burning infrastructure to chargeable platform assets."
Neocloud (new computing cloud) market revenue topped $25 billion in 2025, representing a year-on-year growth of over 200%. Gartner predicts that by 2030, neocloud providers will capture 20% of the AI cloud market share. However, McKinsey simultaneously warns that this business model faces commoditization risks—when GPU supply gradually loosens, a model that relies purely on GPU availability for competition will face margin compression. The entry of hyperscalers like Meta undoubtedly exacerbates this competitive pressure.
02. Data Center Construction Hits Bottlenecks While the computing power leasing market is booming, physical data center construction keeps hitting the brick wall of reality.
In early July, QTS, a data center operator under Blackstone Group, officially halted the Digital Gateway project located in Virginia. Spanning 2,100 acres, the project was originally planned to involve an investment of over $100 billion to build 37 data center buildings, which would have become the world's largest data center campus upon completion. However, after enduring five years of resistance from local residents, facing a state court ruling that the zoning approval was invalid, and the multiple pressures of partners withdrawing first, Blackstone ultimately chose to cut its losses and exit. Days prior, Blackstone had already sold three mature data center assets in Virginia for $3.5 billion, sending a clear signal of strategic contraction.
Coincidentally, computing infrastructure company Crusoe announced in June the "suspension" of a massive 1.8GW data center project in Wyoming. Reportedly, this was due to pressure from its major client, Google, which raised "serious concerns" about the project. The power consumption of this project would be sufficient to supply a medium-sized city.
The stillbirth of these super projects exposes the multiple practical dilemmas underlying the rapid surge in AI computing infrastructure.
First is the physical bottleneck of power supply. Data centers are veritable "power hogs." According to the Electric Power Research Institute, data centers currently account for 5% of U.S. electricity demand, which could triple by 2035. In Virginia—the most data-center-dense region globally—this proportion already exceeds 25%. The existing power grid simply cannot match the scale and growth rate of AI infrastructure demand. JPMorgan Chase's analysis shows that among data center projects planned for completion in 2027, over 60% have not yet broken ground, with power supply bottlenecks being one of the core reasons. In the first quarter of 2025 alone, the total value of delayed data center projects nationwide reached approximately $130 billion.
Second is public resistance and tightening policies. A Gallup poll shows that 70% of Americans oppose building AI data centers near their homes. Practical issues such as high energy consumption, noise, water resource consumption, and driving up the cost of living have frequently hindered the grand AI narrative at the community level. In the first quarter of 2026, opponents nationwide hindered or delayed at least 75 data center projects. The number of active grassroots opposition groups against data centers nationwide surged from 396 at the end of 2025 to 833 in March 2026, covering 49 states. Statistics show that in 2025, the number of canceled data center projects quadrupled to 25, with $18 billion worth of projects halted and $46 billion worth delayed.
Furthermore, compliance and security requirements have also become constraining factors. Microsoft abandoned the $3 billion cloud computing power leasing agreement with Oracle precisely because Oracle lacks the federal security certifications required to manage U.S. government data and is unwilling to undertake large-scale engineering renovations for this purpose. This incident indicates that against the backdrop of increasingly abundant computing power supply, security compliance is becoming a hard threshold for computing power transactions.
Lack of power, water, and permits is replacing the "chip shortage" as the biggest hard constraint on computing infrastructure. Blackstone's exit and Crusoe's suspension mark that capital's investment attitude toward AI infrastructure is returning to rationality from frenzy. These bottlenecks will not kill computing power demand, but they will delay the realization pace of demand—locked-in orders will not disappear, but the landing cycle for new projects will be significantly lengthened.
03. Computing Power Supply and Demand Usher in a New Landscape What does the rise of computing power leasing and the delay of infrastructure projects mean for the semiconductor supply chain?
First, it must be clarified that high-end AI computing power is not in oversupply. Industry research points out that the current computing power market is characterized by a "structural mismatch"—although some low-end general-purpose computing power lacking application scenarios is idle, the gap in high-end intelligent computing power supporting large model training remains as high as 40%, remaining in a state of supply falling short of demand.
This judgment has been fully corroborated in the financial reports of semiconductor giants. Nvidia's full-year revenue for FY2026 reached a record $215.9 billion, a year-on-year increase of 65%, of which data center revenue was as high as $193.7 billion, accounting for nearly 90% of total revenue. The latest quarterly financial report is even more impressive, with data center revenue surging 92% year-on-year. TSMC (Taiwan Semiconductor Manufacturing Company Limited) CEO C.C. Wei explicitly stated in June that global AI chip demand remains strong, and although the company is striving to expand production, supply will still fail to meet demand in the coming years. TSMC's May revenue surged 30% year-on-year, and its capital expenditure for 2026 is expected to be $52 billion to $56 billion, with internal preference leaning toward the upper end. According to the latest reports, AI chip manufacturers like Nvidia are still facing shortages, and TSMC's advanced process and Advanced Packaging capacity utilization rate remains tight.
In the memory sector, competition for HBM remains white-hot. Relying on its leading position in the HBM market, SK Hynix's market capitalization has surpassed Samsung Electronics, making it the highest-valued company in South Korea. Both Samsung and SK Hynix have advanced the mass production of the next-generation HBM4 to early 2026 to cope with the continuously soaring AI demand.
However, the popularization of the computing power leasing model is indeed reshaping the procurement logic of the supply chain. When giants like Meta and xAI open up computing power to external customers, they are actually improving the computing power utilization rate of the entire society. Small and medium-sized AI enterprises no longer need to purchase expensive hardware but are turning to leasing. A report by Apollo points out that GPU prices have risen to around 8 times their level from early 2025, making the leasing model more attractive to small and medium-sized enterprises. This resource-sharing model, to a certain extent, delays the absolute growth rate of total computing power demand, prompting cloud vendors to pay more attention to cost-effectiveness and energy efficiency when procuring hardware.
This is also the deep reason why AI giants are rushing to invest in self-developed chips. On June 24, OpenAI, in conjunction with Broadcom, officially launched the Jalapeño chip, which is specifically optimized for large model inference—this is OpenAI's first self-developed chip, taking only 9 months from design to production. Meanwhile, Anthropic is in talks with Samsung for custom AI chip development; Meta's fourth-generation self-developed chip "Iris" is planned to go into production in September, with the goal of doubling computing power. Faced with high GPU costs, AI giants are reducing unit inference costs and getting rid of over-reliance on Nvidia through customized application-specific chips. This trend constitutes a major benefit for chip design companies like Broadcom, but poses a potential threat to Nvidia's long-term share in the inference market.
From a more macro perspective of the supply chain, the rise of the computing power leasing model is giving birth to a new supply and demand balance mechanism. In the past, the supply chain of AI computing power was linear: chip designers shipped to cloud vendors, and cloud vendors used it themselves or resold it to end customers. Today, giants are both the largest buyers of chips and the lessors of computing power. This dual identity of "self-use + external leasing" has significantly improved the allocation efficiency of computing power resources. For semiconductor equipment manufacturers, this means that the procurement behavior of downstream customers will be more rational—no longer panic hoarding, but refined procurement based on actual utilization rates and investment returns. In the short term, this may lead to a slowdown in the pace of some orders; but in the long run, a healthier demand structure will instead be conducive to the sustainable growth of the supply chain.