The AI Infrastructure Boom Hits a Wall: Too Many Chips, Not Enough Power
TREE NEWS reports: The artificial intelligence infrastructure boom is colliding with a harsh physical reality: there may not be enough electricity to power the semiconductors being produced at record pace. Morgan Stanley analysts warn that the gap between planned data center capacity and available power could reach one-third of the electricity needed to support projected chip sales by 2028. In other words, the market’s optimistic semiconductor shipment forecasts are at risk of being invalidated by something as mundane as the power grid.
Half of the large data center projects announced this year have not yet broken ground. The obstacles are manifold: equipment and labor shortages, community opposition, construction moratoriums, permitting delays, long grid interconnection queues, and insufficient gas turbine capacity. BloombergNEF data shows that nearly half of the U.S. data center pipeline by capacity is led by first-time developers, raising doubts about whether these bottlenecks can be resolved quickly. Rising borrowing costs and increasingly cautious lenders further exacerbate financing uncertainty.
Contractual Protections Are Full of Holes
For data center landlords and lenders, contract terms offer some protection against cost overruns and constrain tenant exit rights in case of delays. However, these protections vary widely. London-based startup Nscale Ltd.’s IPO filing reveals a $44.6 billion compute contract with Anthropic that provides “only limited remedies” in the event of supply chain delays. If Nscale fails to deliver on time, Anthropic has the right to terminate the contract “without liability.”
Oracle Corp. faces similar pressure. The company had to assure investors that its lease for the Jupiter project data center in New Mexico is progressing normally, despite the project being hampered by natural gas pipeline permitting issues. Oracle’s subsequent force majeure notice did little to boost market confidence. Even if rent can be deferred, Oracle must still meet certain payment obligations to project financiers.
The SoftBank SB Energy Case: A $50 Billion Valuation Under Scrutiny
The case of SoftBank Group’s SB Energy epitomizes concerns about an AI infrastructure investment bubble. In its IPO prospectus, the company disclosed an 8.8 gigawatt contracted compute project pipeline, mostly for OpenAI, but only a tiny fraction is under construction, and no data center is operational yet. This has led investors to question its $50 billion valuation.
Notably, Nvidia is intervening with a $105 billion guarantee to backstop tenant OpenAI’s lease obligations and support SB Energy’s construction. This move itself signals that Nvidia is highly alert to how data center construction delays could impact its processor sales outlook. In its latest earnings report, Nvidia explicitly warned that shortages of land, power, buildings, or capital could constrain its future financial performance.
Hyperscalers Face a ‘Stockpiling’ Dilemma
For hyperscale cloud giants like Amazon and Microsoft, the mismatch between chips and data centers is becoming a tangible financial risk. These companies typically purchase servers and networking equipment only months before a data center goes live, so they are reluctant to stockpile chips that cannot be powered and face technological obsolescence. Currently, AI companies hold nearly $350 billion in assets not yet in use.
To bridge the capacity gap before new sites come online, BloombergNEF suggests replacing older chips in existing data centers with Nvidia’s latest, more computationally efficient models. However, this would further inflate hyperscalers’ already massive capital expenditures and increase depreciation pressure. Another potential path is relocating data center capacity to countries with more abundant renewable energy, which could offer Europe more opportunities to participate in AI infrastructure buildout.
Nvidia’s $6 Trillion Market Cap vs. the Risk in Its Earnings
Despite Nvidia’s market capitalization approaching $6 trillion, investors appear unconcerned about the company’s ability to find buyers for its high-margin processors. However, Nvidia’s latest earnings report explicitly lists data center construction bottlenecks as a potential risk factor — a statement that itself deserves market attention.
In the short term, supply chain bottlenecks are delivering handsome returns to some participants: SpaceX is renting out compute capacity at high prices, Caterpillar and Bloom Energy Corp. have order backlogs for on-site power generation equipment, and top electrical engineers command salaries as high as $750,000. But whether this localized prosperity can persist depends on whether overall infrastructure construction can keep pace with chip shipments.
After years of exuberance, the AI feast may be turning into a high-stakes game of musical chairs — and when the music stops, some may be left holding too many chips with nowhere to put them.
Key Takeaways for Investors
- Semiconductor shipment forecasts are at risk: The power gap could invalidate optimistic chip sales projections, pressuring semiconductor stocks.
- Data center developers and their tenants face financial strain: Delays and contractual loopholes could lead to revenue shortfalls and legal disputes.
- Hyperscalers may see capex and depreciation pressure rise: Stockpiling chips or replacing them prematurely could weigh on earnings.
- Nvidia’s growth story faces a physical constraint: Even with a near-$6 trillion market cap, the company acknowledges infrastructure bottlenecks as a risk.
- Watch for opportunities in power generation and renewable energy: Companies like Caterpillar and Bloom Energy are benefiting from on-site power demand.
