AI chips are being manufactured in ever-increasing quantities, but data centers are running out of power: Semiconductor shipment expectations are facing challenges.
The AI infrastructure investment boom is facing a severe supply-demand mismatch crisis.
According to an analysis by Bloomberg columnist Chris Bryant, while the tech industry is pouring money into AI computing infrastructure at an unprecedented pace, data center construction is severely hampered by multiple obstacles, including power shortages, supply chain bottlenecks, and regulatory approval delays. Morgan Stanley analysts warned in a recent report that "there is not enough power to support everyone's plans, and the amount of semiconductors being produced already exceeds the available space." This assessment corroborates similar statements made by Musk previously.
The direct market impact of this mismatch cannot be ignored: the power shortage faced by US data center developers is projected to reach about one-third of the electricity needed for chip sales during the same period, as predicted by Morgan Stanley. This means that the stock market's optimistic expectations for semiconductor shipments risk being proven false, and could have a cascading impact on the financial performance of chip manufacturers, data center developers, and their tenants.
Half of the large-scale projects have not yet broken ground.
The actual progress of data center construction is far behind market expectations. According to data from market intelligence firm Currence, half of the large data center projects announced this year have not yet started construction.
Construction obstacles encompass multiple levels: equipment and labor shortages, community opposition, construction bans, permit delays, and energy supply issues such as lengthy grid connection cycles and insufficient gas turbine capacity. BloombergNEF data further shows that, by capacity, nearly half of the data center pipelines under construction in the United States are led by developers venturing into the field for the first time, and this lack of industry experience undermines confidence in the ability to resolve these issues quickly.
At the same time, rising borrowing costs and more cautious lending practices have exacerbated uncertainty in project financing. Currently, there is significant disagreement among institutions regarding their predictions for the computing power that will ultimately be available in the coming years, reflecting widespread market concerns about when supply chain and regulatory bottlenecks will be resolved.
Contract protection has loopholes, and developers and tenants face different risks.
For data center landlords and lenders, contract terms can provide some protection against cost overruns and bind tenants to their right to opt out in case of delays. However, according to Bloomberg colleague Paul Davies, such protections vary from contract to contract, and tenant qualifications are just as crucial as the quality of the guarantor.
The IPO filing of London-based startup Nscale Ltd. revealed a potential risk: its $44.6 billion computing contract with Anthropic offers "limited relief" in the event of supply chain delays. Anthropic, owned by Dario Amodei, has the right to terminate the contract "without liability" if Nscale fails to deliver on time.
Oracle Corp. is facing similar pressure. The company had to assure investors that its data center lease in the Jupiter project in New Mexico was progressing normally, despite the project being stalled due to gas pipeline permitting issues. Oracle's subsequent force majeure notice failed to boost market confidence. Reportedly, even with deferred rent payments, Oracle still has partial payment obligations to the project's financiers.
SoftBank-backed SB Energy: $50 billion valuation faces scrutiny
The case of SB Energy, a subsidiary of SoftBank Group, vividly illustrates the current concerns about a potential bubble in AI infrastructure investment. The company disclosed in its IPO prospectus that it has 8.8 gigawatts of contracted computing pipeline, most of which is geared towards OpenAI, but only a very small portion is currently under construction, and no data centers are yet operational. This situation has led investors to question its $50 billion valuation.
It's worth noting that Nvidia is stepping in with a $105 billion guarantee—endorsing tenant OpenAI's lease obligations to support SB Energy's construction progress. This move itself demonstrates Nvidia's high level of vigilance regarding the potential impact of data center construction delays on its processor sales prospects. Nvidia also explicitly stated in its latest financial report that shortages of land, electricity, buildings, or capital could constrain its future financial performance.
Hyperscale cloud vendors face "stockpiling" dilemma
For hyperscale cloud computing giants like Amazon and Microsoft, the supply-demand mismatch between chips and data centers is evolving into a real financial risk. These companies typically purchase servers and network equipment "months" before their data centers go into operation, and are therefore unwilling to stockpile chips that cannot be powered and face the risk of becoming obsolete in advance.
Currently, AI companies hold nearly $350 billion in assets that are not yet in use. To bridge the capacity gap before new sites go live, BloombergNEF suggests replacing older chips in existing data centers with Nvidia's latest, more efficient models. However , this will further increase the already massive capital expenditures of hyperscale cloud vendors and add to depreciation pressures.
Another potential solution is to shift data center capacity to countries with more abundant renewable energy sources, which could provide Europe with more opportunities to participate in AI infrastructure development.
Nvidia's market capitalization is approaching $6 trillion, but risks have already emerged in its financial reports.
Despite Nvidia's market capitalization approaching $6 trillion, investors currently don't seem overly concerned about Jensen Huang's company's ability to find buyers for its high-margin processors. However, Nvidia's latest earnings report explicitly listed data center construction bottlenecks as a potential risk factor, a statement that itself warrants market attention.
Currently, supply chain bottlenecks have indeed brought substantial returns to some participants in the short term: SpaceX, owned by Musk, is leasing computing power at high prices, Caterpillar and Bloom Energy Corp. are facing a backlog of orders for on-site power generation equipment, and top electrical engineers are earning as much as $750,000. However, whether this localized prosperity can be sustained depends on whether overall infrastructure construction can keep pace with chip shipments.
As Chris Bryant put it, after years of revelry, the AI feast may be turning into a high-stakes game of musical chairs—ultimately, some may have too many chips and nowhere to put them.
Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.