Bloomberg: AI Giants Face Severe Indigestion from Chip and Data Center Overbuild

Source: Bloomberg Opinion

As the astonishing and alarming capabilities of artificial intelligence spark global debate, investors are waking up to a harsh reality: the computing and energy infrastructure that companies like OpenAI and Anthropic PBC depend on is far harder to build than a chatbot prompt.

Data center projects are encountering multiple obstacles, including equipment shortages, labor shortfalls, community opposition, construction halts, and permitting delays. Long wait times for grid connections and lagging production of gas turbines may result in insufficient electricity to power all the energy-hungry server farms.

The Cost of Gigawatt-Scale Computing

A single gigawatt of computing capacity can cost tens of billions of dollars. Fund managers must now assess the financial consequences of delays or failures in these mega-projects, which could hit developers, potential tenants, and semiconductor shipments.

Some experts argue that supply chain bottlenecks may actually help the AI boom by preventing computing capacity from outstripping demand, while simultaneously pushing up technology revenues and data center-related prices.

Winners and Losers in the Supply Chain

Companies that can supply scarce equipment and services are currently reaping huge profits:

• Elon Musk's SpaceX is leasing computing capacity at astronomical prices
• Caterpillar and Bloom Energy have large order backlogs
• Top electricians are earning annual salaries of $750,000

However, the tech industry's frantic construction may still fail to avoid a scenario where there are too many chips but not enough operational server farms. Morgan Stanley analysts warn: "There is not enough electricity to meet all demand, and semiconductor production exceeds the number of sites available." Elon Musk has proposed putting processors into orbit to bypass these obstacles.

Uncertainty in Infrastructure Financing

Despite massive capital inflows into AI infrastructure and the involvement of top Wall Street talent, estimates of future computing capacity vary widely, reflecting uncertainty around supply chain and regulatory bottlenecks. High borrowing costs and increasingly cautious lenders cloud the outlook for project financing.

According to market intelligence firm Currence, half of the large data center projects announced this year have not yet broken ground. BloombergNEF notes that nearly half of the U.S. data center pipeline is being built by first-time developers.

Case Studies: SB Energy and Nscale

SB Energy, backed by SoftBank, disclosed in its IPO filing that its 8.8-gigawatt contracted computing pipeline is primarily aimed at OpenAI. However, only a few projects are under construction, and none are operating data centers. Its $50 billion valuation may deter investors.

Contract terms can protect landlords and lenders from cost overruns, but the level of protection varies. Nscale, a London-based startup, revealed in its IPO filing that its $44.6 billion computing contract with Anthropic provides only "limited supply chain delay relief." If Nscale fails to deliver on time, Anthropic can terminate the agreement without penalty.

Oracle was forced to reassure investors that its lease for the Jupiter data center in New Mexico was on track, after natural gas pipeline permitting issues led to a force majeure notice that raised concerns.

The Power Gap and Market Risks

Overall, U.S. data center developers face a power shortfall of about one-third by 2028. If projects are delayed, stock market optimism about semiconductor shipments may prove unfounded.

Tech giants—the "hyperscalers"—may be reluctant to stockpile chips that cannot be powered and risk becoming obsolete. Amazon typically purchases servers only months before a data center goes live.

BloombergNEF suggests that older chips could be removed to prioritize deployment of Nvidia's latest, more efficient models. However, this would increase capital expenditure and depreciation costs.

Data center capacity may shift to countries with easier access to energy, potentially offering more opportunities to Europe. Big tech companies are building in phases to hedge risks, but they must make long-term financial commitments to secure memory supply.

Nvidia's Exposure and the "Musical Chairs" Risk

Nvidia's market capitalization is approaching $6 trillion, and investors are not worried about demand for its processors. However, the chipmaker has warned that shortages of land, electricity, construction capacity, or capital could hinder data center construction and affect future performance. Nvidia is supporting SB Energy's construction and has provided $105 billion in guarantees for OpenAI's lease obligations.

After years of boom, the AI party may be turning into a high-stakes game of musical chairs—some players could be left holding too many chips with nowhere to sit.

Source: Bloomberg