AI The boom in infrastructure investment is facing a severe mismatch between supply and demand: chip production capacity is expanding faster than data center construction, and power shortages, supply chain bottlenecks, and approval delays have led to delays in numerous projects. Morgan Stanley estimates that by 2028, the power gap in US data centers will account for about one-third of the electricity required by chip demand. AI The return on capital expenditure and the prospects for chip sales are under test.

According to analysis by Bloomberg columnist Chris Bryant, despite the tech industry spending money on computing infrastructure at an unprecedented pace, the construction of data centers is severely lagging behind due to multiple obstacles such as power shortages, supply chain bottlenecks, and delays in regulatory approvals. Analysts at Morgan Stanley warned in a recent report: "There is not enough electricity to support everyone's plans, and the number of semiconductors being produced has already exceeded the capacity of available facilities." This judgment corroborates similar statements made by Elon Musk before.

The direct market impact of this mismatch cannot be ignored: The power shortage faced by US data center developers is expected to reach about one-third of the electricity required for chip sales during the same period as predicted by Morgan Stanley by 2028. This means that the stock market's optimistic expectations regarding semiconductor shipments are at risk of being disproven, and it may have a cascading effect on the financial performance of chip manufacturers, data center developers, and their tenants.

Half of the large-scale projects have not yet begun construction.

The actual progress of data center construction lags far behind market expectations. According to data from the market intelligence agency Currence, half of the large-scale data center projects announced this year have not yet begun construction.

Construction obstacles cover multiple aspects: shortages of equipment and labor, community opposition, construction bans, delays in obtaining permits, as well as long grid connection times and insufficient capacity of gas turbines among other energy supply issues. Data from BloombergNEF further indicates that, in terms of capacity, nearly half of the data center pipelines under construction in the United States are led by developers new to this field. The lack of industry experience makes it difficult for outsiders to have confidence that these problems will be resolved quickly.

At the same time, rising borrowing costs and more cautious lending institutions have also increased the uncertainty of project financing. Currently, there are significant differences in the predictions made by various institutions regarding the final computing power that can be achieved over the next few years, reflecting the general skepticism in the market about when supply chain and regulatory bottlenecks will be resolved.

Contract protection has loopholes, posing different risks to developers and tenants

For data center landlords and lenders, contract terms can provide a certain level of protection against cost overruns and restrict tenants' rights to exit in the event of delays. However, as Bloomberg colleague Paul Davies pointed out, such protection mechanisms vary from contract to contract, and tenant qualifications as well as the quality of guarantors are equally important.

A document from London-based startup Nscale Ltd reveals potential risks: a $44.6 billion computing contract signed with Anthropic "only provides limited relief" in the event of supply chain delays. If Nscale fails to deliver on time, Anthropic, a subsidiary of Dario Amodei, has the right to terminate the contract "without any liability."

Oracle Corp also faces similar pressures. The company had to assure investors that the lease for its data center in Jupiter, New Mexico, was progressing smoothly, despite delays due to natural gas pipeline licensing issues. The force majeure notice issued subsequently by Oracle did not boost market confidence. It is reported that even though rent payments could be deferred, Oracle still has to fulfill certain payment obligations to the project financiers.

SoftBank Group SB Energy: $50 billion valuation questioned

The case of SB Energy under SoftBank Group highlights the current concerns regarding the bubble in infrastructure investments of AI. The company disclosed in its IPO prospectus that it has 8.8 gigawatts of contracted computing project pipeline, most of which are targeted at OpenAI, but currently only a very small portion is in the construction phase, and there are no data centers that have been put into operation yet. This situation has led investors to question its valuation of 50 billion dollars.

It is worth noting that NVIDIA is getting involved with a guarantee of $105 billion – providing endorsement for the leasing obligations of tenant OpenAI to support the advancement of the construction of SB Energy. This move itself indicates that NVIDIA is highly vigilant about the potential impact of delays in data center construction on its processor sales prospects. NVIDIA also explicitly mentioned in its latest financial report that shortages in land, electricity, construction, or capital could constrain its future financial performance.

Super-large cloud providers face the dilemma of "hoarding goods"

For "ultra-large-scale" cloud computing giants such as Amazon and Microsoft, the mismatch between supply and demand of chips and data centers is evolving into a real financial risk. These companies typically purchase servers and networking equipment "months" before the data centers are put into use, and therefore are reluctant to stockpile chips that cannot be powered on in advance and that face the risk of becoming technologically obsolete.

Currently, the scale of assets held by AI that have not yet been put into use is approaching $350 billion. To bridge the production capacity gap before the launch of the new site, BloombergNEF suggests replacing the older chips in the existing data centers with NVIDIA's latest models, which are more efficient in computing. However, this will further increase the already substantial capital expenditures of these ultra-large cloud providers and add to the pressure of depreciation.

Another potential solution is to transfer data center capacity to countries with more abundant renewable energy, which could provide Europe with more opportunities to participate in the construction of AI infrastructure.

NVIDIA's market value approaches $6 trillion, but risks are already evident in its financial reports

Although NVIDIA's market value is approaching $6 trillion, investors currently do not seem to be overly concerned about the ability of the companies under Jensen Huang to find buyers for their high-profit processors. However, NVIDIA has explicitly identified data center construction bottlenecks as a potential risk factor in its latest financial report, and this statement itself is worth the market's attention.

Currently, supply chain bottlenecks have indeed brought substantial returns to some participants in the short term: Elon Musk's SpaceX is renting out computing power at high prices, Caterpillar and Bloom Energy Corp's on-site power generation equipment orders are piling up, and the salaries of top electrical engineers can even reach up to $750,000. However, whether this localized prosperity can continue depends on whether the overall infrastructure construction can keep up with the pace of chip shipments.

As Chris Bryant said, after several years of revelry, the AI feast may be evolving into a high-risk "chair grab" game – in the end, there might be people who hold too many chips but have nowhere to put them.

Risk and Disclaimer Notice