OTWopentechwire
Tech Intelligence, Openly Wired
Startups

Nvidia Considers Unprecedented Construction Guarantee for Nscale Data Centre

The chip giant is exploring a residual-value backstop for a West Virginia AI facility, signalling how far the industry will stretch to sustain buildout momentum amid investor scepticism.

LT
Linh T. Pham
Southeast Asia Reporter · Hanoi
Nov 2, 2026
5 min read
Nvidia Considers Unprecedented Construction Guarantee for Nscale Data Centre
Credit: Thomas Fuller / Getty Images

A Deeper Financial Commitment

Nvidia is exploring a guarantee arrangement that would see the chip maker backstop billions of dollars in construction financing for Nscale's planned West Virginia data centre campus, according to multiple reports. The proposed structure - known as a residual value guarantee - would require Nvidia to assume control of the facility if Nscale defaults on construction payments, then locate a buyer and replacement tenant. It represents a marked departure from the chip-lease and purchase support Nvidia has previously extended to cloud providers, pushing the company into exposure on bricks, mortar and electricity infrastructure rather than silicon alone.

The West Virginia site is anchored by a 460-megawatt building for which Anthropic has signed a contract valued at $44.6 billion, according to Nscale's IPO filing. A second building of similar scale is under discussion, with a price tag near $45 billion. Both facilities are expected to deploy Nvidia's forthcoming Vera Rubin chips. For Nvidia, the guarantee would secure a marquee customer for next-generation hardware at a moment when hyperscale buyers are spreading orders across rival chip makers. For Nscale, it would unlock construction capital that has proved difficult to raise on the company's own balance sheet.

The Neocloud Funding Squeeze

Nscale is part of a cohort of AI-focused infrastructure providers - often called neoclouds - that rent compute capacity without the cash flows of an established software or consumer business beneath them. That model has come under mounting pressure. Shares of CoreWeave, Nebius and IREN each fell more than 30 per cent in October alone, reflecting investor anxiety over capital intensity, execution risk and the absence of near-term profitability. At Opentechwire, we've tracked similar valuation compression across the region's GPU-as-a-service start-ups, from Seoul to Singapore, as term sheets tighten and due diligence lengthens.

Nscale is seeking to list at a valuation of up to $35 billion despite having brought only around 25,000 advanced AI chips online as of August - a small fraction of the nearly 500,000 it has committed to deploy. The company's ability to scale hinges on securing low-cost construction debt, a challenge that has grown steeper as lenders scrutinise project timelines, power availability and tenant creditworthiness. The proposed Nvidia guarantee addresses the latter concern, effectively substituting the chip maker's investment-grade credit for Nscale's untested balance sheet.

West Virginia Site Faces Local and Regulatory Headwinds

The West Virginia campus has encountered obstacles beyond capital. Runoff from the construction site flooded neighbouring homes on multiple occasions, drawing local opposition and regulatory attention. After months of negotiation, the project secured state certification in September, clearing a procedural hurdle but leaving community concerns unresolved. The incident underscores a broader pattern: AI data centres are landing in jurisdictions eager for investment and jobs but often unprepared for the environmental and infrastructural demands of multi-hundred-megawatt facilities.

Power availability remains a binding constraint. Earlier this month, Microsoft withdrew from a planned Nscale site in England after grid connection was deferred into the 2030s, a reminder that even anchor tenants will walk when electricity timelines slip. West Virginia offers more immediate grid capacity than many European markets, but the state's coal-legacy transmission infrastructure will require upgrades to handle the load profiles these facilities demand. Whether those upgrades arrive on schedule - and who pays for them - will shape the campus's commercial viability.

Nvidia's $170 Billion Exposure

Nvidia has extended roughly $170 billion in investments and credit support to cloud providers and large customers, according to industry data. That figure includes convertible financing, chip-purchase guarantees and leasing arrangements designed to ease upfront capital requirements for buyers of H100, H200 and next-generation accelerators. In September, Nvidia committed $1 billion in convertible financing to Nscale, a stake that would deepen further under the construction guarantee now being discussed.

The scale of Nvidia's exposure reflects both the company's dominance in AI accelerators and the structural challenge facing its customers: the gap between the capital required to build AI infrastructure and the revenue those facilities generate in their early years. By extending credit and guarantees, Nvidia smooths demand for its chips and locks in design wins against AMD, custom silicon from hyperscalers, and emerging challengers. But it also ties the company's financial performance to the execution risk of borrowers whose operational track records are thin.

Anthropic's Role and Credit Questions

Anthropic's $44.6 billion contract with Nscale is among the largest AI infrastructure commitments disclosed to date. The AI safety and research company has distributed its infrastructure spending across multiple chip architectures and providers, a hedging strategy that reduces dependence on any single supplier but complicates long-term planning for data centre operators. One investor noted that Anthropic's anchor-tenant role likely explains Nvidia's willingness to consider a construction guarantee: losing the contract would leave Nscale without a cornerstone customer and Nvidia without a committed buyer for hundreds of thousands of Vera Rubin chips.

Anthropic filed for an initial public offering this week, but the company faces credit headwinds of its own. Blackstone has cautioned investors that Anthropic may receive a below-investment-grade rating after listing, a designation that would raise borrowing costs and limit access to certain debt markets. A junk rating would also ripple through Anthropic's supply chain: landlords and co-location providers typically price risk based on tenant credit quality, and a downgrade could force Anthropic to seek guarantees from financially stronger partners or prepay capacity. That dynamic may already be influencing the structure of the Nscale contract and Nvidia's discussions around support.

Implications for AI Infrastructure Finance

The residual value guarantee under discussion would represent a new frontier in vendor financing. Chip makers have long offered payment terms, leasing and trade-in programmes to support customer purchases. Extending that support to construction debt - where the collateral is a half-built data centre rather than liquid hardware - introduces duration, execution and real-estate risk that sits outside Nvidia's traditional competence. If the guarantee proceeds, it will set a precedent that other suppliers may feel pressure to match, particularly as neoclouds shop for the most accommodating capital partner.

It also raises questions about how risk is distributed across the AI supply chain. Hyperscalers have historically internalised data centre construction risk, funding campuses from operating cash flow and accessing investment-grade debt markets. Neoclouds, lacking that financial cushion, have turned to private equity, venture debt and now vendor support. The result is a financing structure in which the chip maker, the facility operator and the anchor tenant are all financially interdependent - and all exposed if power timelines slip, utilisation falls short or the AI spending cycle turns.

At Opentechwire, we've observed similar interdependencies in Asia's AI infrastructure build, where GPU suppliers, co-location operators and model developers are increasingly locked into multi-year contracts with limited exit optionality. The model works as long as demand for inference and training capacity grows faster than supply. When that assumption falters, the question of who absorbs losses - and in what order - will move from footnote to front page.

Read next
Startups

Big Tech's AI Infrastructure Bet Faces a Reckoning as Free Cash Flow Turns Negative

Priya Nair · 5 min
Startups

China's AI Chip Makers Face Investor Scepticism Despite Revenue Boom

Wei Zhang · 4 min
Startups

Samsung's Record Quarter Exposes Tension Between AI Profits and Shareholder Returns

Mei-Lin Tan · 5 min
Spot something wrong? Email corrections@opentechwire.com. We log every correction publicly.