Rising Interest Rates Squeeze AI Data Center Financing
Higher borrowing costs are pressing AI infrastructure deals, with CoreWeave bond yields above 15% and investors fleeing neocloud stocks amid fears of unsustainable debt loads. The shift is clear as market concerns over debt-financed buildout weigh on shares and push yields to distressed levels.
CoreWeave Stock and Bond Stress
CoreWeave’s stock dropped by 15 per cent on the Nasdaq exchange on Wednesday before bouncing back up by nearly 6 per cent on Thursday. The company’s bond yields have risen above 15 per cent in recent days, a level investors consider severely distressed. Meanwhile, the cost of CoreWeave’s credit default swaps, a type of insurance on its bonds, has also soared to distressed levels.
Broader Neocloud Sector Under Pressure
Cerebras Systems Inc. CBRS-Q, Bell’s other anchor tenant, saw its shares slump about 13 per cent since the beginning of the week. Two other neoclouds – Nebius Group N.V. NBIS-Q, and Iren Ltd. IREN-Q – each saw their share prices slump around 14 per cent in the same period. Investor concerns about debt levels associated with the AI build-out are weighing on the shares of several companies in the sector and pushing bond yields related to those infrastructure projects to distressed levels.
Gartner Warns of Future Writedowns
In a September note to investors, Gartner analyst Will Sommer warned that “multitrillion-dollar writedowns of AI infrastructure assets are more likely than not by 2032″ if capital spending continues at its current trajectory. He estimated there will not be enough revenue to go around, forcing more than 80 per cent of neocloud providers to cease independent operations by 2030. “Only once markets consolidate and mature will providers be able to earn sustainable margins,” he wrote.
Flight to Safety and Ongoing Construction
Investors fleeing AI stocks and bonds for government bonds have caused those yields to decline. On Thursday, the benchmark 10-year U.S. Treasury yield fell more than 10 basis points, even after consumer pricing data and wholesale inflation data released on Wednesday and Thursday, respectively, showed that rising prices remain a concern. Construction is currently under way at Bell’s data centre outside Regina, where CoreWeave and Cerebras will be tenants.
Environmental Pressures Complicate AI Data Center Financing
Beyond financial strain, AI data center projects face mounting environmental scrutiny that further complicates debt-financed expansion. By 2026, local community resistance had blocked the construction of AI data centers worth approximately $130 billion due to concerns over power and water consumption, heat generation, noise, and pollution. These facilities consume around 60 kilowatts of electricity per server rack—six times the usage of general-purpose data centers—intensifying strain on regional power grids.
In regions like Frankfurt, Germany, data centers already account for 41% of citywide electricity use, with grid saturation preventing new connections until at least the 2030s. Similar pressures exist in Dublin, where data center electricity consumption doubles that figure. This growing demand has led U.S. utilities to delay retiring 15 coal-fired power plants to meet AI infrastructure needs, prolonging fossil fuel reliance despite broader decarbonization goals.
The environmental footprint extends beyond energy, with AI data centers consuming millions of gallons of water for cooling through evaporation processes. As investors assess long-term viability, these resource demands—combined with rising interest rates and debt concerns—add another layer of risk to large-scale AI infrastructure projects, particularly in water-stressed or grid-constrained regions.
Construction Milestones at Bell’s Regina Data Centre
Construction is currently under way at Bell’s data centre outside Regina, where CoreWeave and Cerebras will be tenants. This facility represents a key component of BCE Inc.’s planned AI data centre in Saskatchewan, designed to support growing demand for AI compute infrastructure. The project underscores ongoing buildout efforts despite rising financing costs and investor caution.
As anchor tenants, CoreWeave and Cerebras are expected to utilize significant portions of the data centre’s capacity for training and running machine learning models. Their involvement highlights the strategic importance of the site within the broader neocloud ecosystem, even as market pressures mount on debt-financed expansion.
The Regina development reflects continued investment in physical AI infrastructure, with construction progressing amid broader sector challenges. While bond yields spike and equity valuations face pressure, the advancement of this facility signals that certain buildout initiatives remain active, supported by long-term commitments from major technology and telecommunications partners.
AI Data Center Power Consumption Trends
AI data centers consume approximately 60 kilowatts of electricity per server rack, significantly higher than the 10 kilowatts per rack used by general-purpose facilities. This sixfold increase in power demand intensifies strain on regional electrical grids, particularly in areas already experiencing high data center concentration.
In Frankfurt, Germany, data centers account for 41% of citywide electricity use, with grid saturation preventing new connections until at least the 2030s. Similarly, in Dublin, data center electricity consumption reaches double that level, further highlighting the growing pressure on urban power infrastructure from AI-driven compute expansion.
AI Data Center Operator Landscape
As of August 2025, The Information tracked 18 planned or existing AI data centers in the United States operated by major technology firms including Amazon Web Services, CoreWeave, Crusoe, Meta, Microsoft/OpenAI, Oracle, Tesla, and xAI. Additional operators such as Digital Realty and Alibaba are also active in the space, with AI data center construction underway in China, India, Europe, Saudi Arabia, and Canada.
The market distinguishes between two primary types of providers: hyperscalers and neoclouds. Hyperscalers refer to large technology companies like Google, Meta, Microsoft, Oracle, and Amazon, while neoclouds represent a newer generation of specialized data center providers. Examples of neoclouds include CoreWeave, Nebius, Nscale, and Lambda, which focus specifically on AI-optimized infrastructure for machine learning workloads.
In January 2025, OpenAI, in partnership with Oracle and Softbank, launched the Stargate project, which by September 2025 comprised six built or proposed AI data centers across the United States. In response, Amazon announced Project Rainier in October 2025—a $11 billion AI data center on 1,200 acres of farmland in Indiana—designed to support machine learning models from Anthropic and projected to use 2.2 gigawatts of electricity upon full buildout.
Sources
- AI data center
- The hidden environmental cost of AI data centers
- Investor doubts over data centre buildout hit AI stocks, bonds
Photo: Anderseidesvik (CC BY-SA 4.0)
