After endorsing CoreWeave, now investing in Firmus—Nvidia becomes the "AI Central Bank"
After publicly supporting CoreWeave but seeing its stock price plummet by 11.9%, Nvidia now plans to inject hundreds of millions of dollars into Australian AI cloud service provider Firmus Grid, whose IPO attempt failed. Nvidia is intervening in the AI infrastructure financing crisis in a manner increasingly reminiscent of a central bank, driven by a straightforward logic: every dollar invested by Jensen Huang enables customers to buy one more Nvidia chip. Currently, neocloud serves as a critical line of defense for Nvidia against tech giants developing their own chips.
Nvidia is intervening in the AI infrastructure financing crisis in a manner increasingly reminiscent of a central bank.
After verbally backing its key holding CoreWeave failed to calm the market, the chip giant is reportedly planning to inject hundreds of millions of dollars into Firmus Grid, a troubled Australian AI cloud service provider.
According to a report by The Information on Tuesday, Nvidia is negotiating with Firmus on an investment worth hundreds of millions of dollars. After Firmus’s $30 billion IPO plan fell through at the beginning of the month, it is seeking at least $2 billion in private financing. Nvidia has participated in Firmus’s last three funding rounds and is one of the company's major shareholders.
This potential follow-on investment would be Nvidia’s latest signal to the market—at a time when AI stocks are facing massive sell-offs, it is willing to support the entire ecosystem with real cash rather than just words. With the financing environment for AI data center operators deteriorating sharply and public market investor confidence remaining low, whether Nvidia’s strategic capital injection can truly reverse the downward trend remains a core concern for the market.
Firmus IPO Fails, Nvidia Steps In
Firmus’s journey to public listing ended in failure, becoming one of the triggers for the latest AI stock sell-off.
The seven-year-old Australian company initially planned to list on the Australian Securities Exchange at a $30 billion valuation, aiming to raise $5 billion, but ultimately withdrew its application citing “market volatility.” Reports indicate that investors generally believed its valuation was too high and speculative, with little alignment between Firmus’s current revenue scale and its planned data center capacity.
Firmus’s main investors include Blackstone, Coatue Management, and Jane Street. In terms of financing structure, the company has raised over $13 billion to date, about $10 billion of which is debt. Nvidia has already deeply tied itself to Firmus under its credit support program—this June, Nvidia agreed to provide credit backing for Firmus to procure at least 170,000 Rubin and Blackwell GPU server racks, making it easier for the company to secure debt financing. These server racks, valued at about $15 billion, will be deployed in a facility in Batam Island, Indonesia, built by Singapore data center developer DayOne.
After the failed IPO, Firmus is now seeking at least $2 billion in private financing. According to The Information, citing people familiar with the matter, Nvidia had already planned to invest significant capital in the IPO, and the size of Nvidia’s investment in this private round may still change, with the possibility that the financing may fall through.
Nvidia’s Logic: Every Dollar Invested Is Paying for Its Own Business
Nvidia continues to inject capital into new cloud service providers (neocloud), underpinned by a clear business rationale.
Google, Microsoft, Amazon, and other hyperscale cloud providers are ramping up development of their own chips to reduce dependence on Nvidia, while neoclouds are an important pillar of Nvidia GPU demand. The logic behind Nvidia’s additional investment in Firmus is straightforward: for every dollar Jensen Huang puts in, he helps customers buy another chip.
According to The Information’s analysis, Nvidia has already provided more than $170 billion in credit guarantees and capital support to neoclouds and other large clients. As of July this year, Nvidia’s holdings of unlisted securities—mainly private company equity—reached $51.2 billion, more than doubling in six months.
To sustain this ecosystem, Nvidia is also adjusting its support policies. According to high-level insiders, Nvidia initially provided credit support to neoclouds in exchange for a share of their future cloud revenue, but some larger neoclouds did not accept this term. Nvidia has now shifted to offering credit agreements with shorter durations, in exchange for a smaller share of revenue. In addition, Nvidia has set up a dedicated internal team, co-led by Jensen Huang’s son-in-law Nico Caprez, to help neoclouds find data center locations with available power capacity.
The Warning from CoreWeave: Words Are No Match for the Market
Firmus’s predicament is not unique; the entire neocloud industry is facing the twin pressures of surging financing costs and collapsing investor confidence, and Nvidia’s verbal support has proved ineffective.
Last week, Jensen Huang publicly defended CoreWeave at an event in Stockholm, claiming its “cost of capital is actually coming down.” The market was not persuaded—CoreWeave’s share price plunged 11.9% the next day. CoreWeave’s borrowing costs have moved in the opposite direction from Huang’s statements: its project finance spread, backed by chips and customer contracts, has widened from 225 basis points above the benchmark rate in March to 550 basis points in August; for unsecured bonds, its 2032 bonds now yield about 14%, up from around 8.5% only four months ago.
The experience of Australian data center operator Nextdc offers another warning: on Wednesday, the company’s share price crashed 41% in a single day after project delays due to public opposition and lowering of profit forecasts. Just last month, Nextdc issued A$1.1 billion in convertible bonds at a low coupon of 1.75%, but its share price has now fallen to about one-third of the conversion price, rendering these securities deep out of the money.
In the face of persistent market skepticism, Jensen Huang previously told investors “this too shall pass,” urging them to “buy the dip.” Now, Nvidia is being asked to back up this promise with action. For a company dubbed “the AI central bank” by the market, investing in Firmus may be just the latest step in its role as the “lender of last resort,” but whether it can truly restore market confidence remains to be seen.
Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.
You may also like
Positive response to Changxin 4F2 breakthrough, Micron CTO: 3D DRAM is "long-term more scalable and cost-effective"
The global storage chip market has recently experienced valuation fluctuations, and debates over technological approaches are intensifying. Scott DeBoer, President of Micron, stated that although the company internally evaluated the 4F² architecture for a long time, it will stay committed to the 6F² roadmap and move directly to 3D DRAM, citing a lack of sustainable scalability for the vertical channel transistors in the 4F² design. In contrast, ChangXin Memory Technologies previously announced plans to launch 4F² DDR5 products by the end of the year, and Samsung is also considering introducing the architecture in 2028.

Excess Supply of Hashrate!?

What did Nvidia say at the Berlin GTC? Here’s Goldman Sachs’ summary
After attending Jensen Huang's GTC keynote, Goldman Sachs released a research report highlighting three key points: Nvidia's supply chain is undergoing significant expansion across multiple segments, suggesting a larger-scale GPU shipment is imminent; DSX POD compresses data center deployment cycles from several months to just two weeks; physical AI revenue reached $11 billion in the first half of FY27, accounting for about 30% more than the entire FY26, with a clear trend toward diversified revenue streams. The report also announces the release of the OpenShell architecture to address enterprise AI security concerns.

