NVIDIA Joins Forces with Wall Street to Unlock $500 Billion: AI Compute Redefined as a "New Asset Class"

In the second week of August 2026, the AI computing sector was shaken by a piece of news that reverberated across both the global technology and financial industries. On August 10, local time, NVIDIA announced that it had signed memoranda of understanding with six of the world’s top financial institutions – Apollo Global Management, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR – to establish independent computing power financing platforms. The plan aims to gradually mobilise over $500 billion (more than RMB 3.37 trillion) in third‑party capital for AI infrastructure construction. This unprecedented collaboration was hailed by NVIDIA CEO Jensen Huang as a "big idea."

Following the announcement, NVIDIA’s share price fell more than 3% intraday and eventually closed down 2.86%, wiping over $70 billion off its market capitalisation in a single day; the Philadelphia Semiconductor Index also dropped 2.94%. The next day, NVIDIA’s stock edged down 0.02%, essentially stabilising. The market’s reaction was telling – in the face of such "good news," Wall Street was scrutinising the grand AI narrative through a very different lens.

From Chip Supplier to "Capital Organiser"

At the heart of this partnership is a fundamental shift in NVIDIA’s identity.

Under the proposed framework, the six institutions will set up dedicated capital pools of considerable size, offering financing at competitive interest rates to frontier AI labs, enterprises, and AI cloud service providers within the NVIDIA ecosystem. With third‑party funding, clients can procure hardware and build "AI factories" based on the DSX architecture, without having to heavily tap their own balance sheets. In return, the capital providers receive long‑term returns linked to the actual utilisation of computing power, turning AI compute projects into investable infrastructure assets akin to public utilities, suitable for global institutional capital allocation.

"NVIDIA has reached an important milestone. We started out focused on chip design, and now we are helping to create a new class of productive, investable infrastructure – AI factories," said Jensen Huang, founder and CEO of NVIDIA. In the announcement, he wrote: "We have moved from an era where companies buy chips and build data centres one by one, to an era where AI factories can be financed as productive infrastructure."

In his view, "in the AI era, compute is revenue." NVIDIA’s computing power is not only widely adopted and adaptable to various models and workloads, but can also be continuously upgraded via CUDA software to extend its useful life and improve economic returns, backed by a global ecosystem of developers, customers, and downstream demand. "That is why we are bringing together the world’s leading long‑term capital institutions to independently provide credit support for AI infrastructure projects. These financing platforms will help clients access scarce computing power at scale, build DSX AI factories, and empower every industry and every country in the AI era."

This means that NVIDIA is transforming from a hardware vendor into a capital organiser for the AI infrastructure ecosystem. Once clients secure financing through these platforms, their data centres become more deeply integrated with NVIDIA’s chip, software, and operational standards. The financing platforms not only unlock new procurement demand but may also strengthen NVIDIA’s influence over industry standards and its clients’ technology ecosystems.

Goldman Sachs CEO David Solomon revealed that Jensen Huang had proactively approached these Wall Street giants with the financing proposal. BlackRock CEO Larry Fink said he believes the project represents the beginning of a "new chapter in financial engineering," comparable in significance to the birth of mortgage‑backed securities in the 1970s.

Compute as a "New Asset Class"

The most noteworthy aspect of this collaboration is the redefinition of assets.

For a long time, GPUs have been viewed as rapidly depreciating electronic hardware, making them difficult to use as high‑quality collateral for long‑term, low‑cost credit. NVIDIA is now attempting a cognitive overhaul – repositioning the entire CUDA‑based compute cluster as an investable asset that can generate stable cash flows and be transferred across clients.

In an interview with foreign media, Huang stated bluntly: "This really is the first time that a technology chip has become an asset class that is investable. They are now assets that generate revenue – highly productive, long‑lived, fungible, and extremely flexible." He further explained that because NVIDIA’s hardware is widely adopted and can be seamlessly migrated between different clients, lenders can reliably underwrite compute power as an asset. "Fundamentally, what is different about this industry and this form of computing is that computers have now become part of infrastructure, just like electricity and the internet – so you have to look at it the way you look at infrastructure."

From the perspective of the participating financial institutions, AI computing is shifting from a technological product to an infrastructure asset that can be held long‑term by global capital. Apollo President Jim Zelter said that modern compute has become "a scarce and critical core asset with highly attractive investment characteristics, poised to drive long‑term economic growth and significant productivity gains." Goldman Sachs Chairman David Solomon suggested that financial institutions could build a new credit market around NVIDIA’s computing power.

Blackstone President Jonathan Gray said in a CNBC interview that among their portfolio companies, demand for large language models had grown seven‑fold over the past six months, but supply – whether in data centres, power, or chips – was far from keeping pace. BlackRock’s Larry Fink remarked that data centre financing "feels like when I first entered the mortgage‑backed securities market in the 1970s," and he believes this will become the next frontier of financial engineering.

The core logic of this financing structure is to treat AI data centres and GPU clusters as comparable to commercial real estate, telecom towers, energy pipelines, and toll roads – massive upfront investment, but if utilisation can be sustained, they can generate cash flows through long‑term leases and service revenues, thus becoming appraisable and financeable infrastructure assets.

Analysts point out that the key to this arrangement is: the success or failure of an AI company does not affect the value of the compute assets themselves – even if a company goes under, the equipment can be taken over by others. Because these chips have longer lives and retain more value, financial institutions can securitise them, turning them into tradable, financeable assets.

Controversies and Concerns: The "Debt Spectre" of Circular Financing

After the $500 billion plan was unveiled, market questions quickly turned to "circular financing." This refers to NVIDIA investing in, guaranteeing, or facilitating financing for its clients, who then use those funds to buy NVIDIA chips – the money ultimately flows back to NVIDIA, creating a closed loop of "NVIDIA‑backed financing → client chip purchases → money back to NVIDIA."

This model has triggered deep concerns on Wall Street. Some analysts have pointed out that when NVIDIA needs to bring in six global capital giants to build financing channels for its clients, the market sees not only strong demand but also asks: has AI construction become so expensive that tech companies alone can no longer sustain it with their own cash flows? A simple and unsettling question has been repeatedly raised: if AI compute is truly in such short supply, why does the chip giant need to help clients arrange financing?

Market participants worry that if AI demand, financing conditions, or partner repayment capacity fall short of expectations, risks could propagate among chip suppliers, clients, and capital providers. With "NVIDIA’s client base highly concentrated, with top cloud vendors and AI labs contributing the lion’s share of chip orders," such a tightly coupled borrowing‑and‑procurement model could amplify industry cyclical risks.

The capital market has cast its vote of no confidence in real money. After the announcement, NVIDIA’s five‑year credit default swap (CDS) spreads jumped nearly 6 basis points in a single day, doubling since late May and standing just a step away from the record high set in late July. Rising CDS prices indicate that institutional investors perceive higher contingent guarantee risks for NVIDIA and greater debt default concerns.

The "dot‑com bubble" spectre has also been repeatedly invoked. Many market observers worry that NVIDIA’s financing model is essentially "Wall Street lending to AI clients, who then spend all the money on NVIDIA chips" – artificially creating chip demand through external leverage rather than genuine organic demand. Under this scenario, if downstream AI companies’ revenues fall short and they cannot repay their huge loans, the entire industry chain could experience a domino‑like debt meltdown, replicating the vendor‑financing chaos before the 2000 dot‑com crash.

Michael Burry, the investor famous for "The Big Short," openly warned that NVIDIA’s expanding financing operations have pushed circular spending to unprecedented levels. Together with another well‑known short‑seller, Jim Chanos, he went further, targeting the financial leverage and structured design behind the plan – Burry argued that the unnatural credit structuring in the late stages of a bull market, aimed at sustaining momentum, is what is truly worrying; Chanos even directly likened the plan to the financial engineering that contributed to the 2008 subprime crisis. Financial blog Zerohedge also warned that a large number of off‑balance‑sheet SPV special‑purpose vehicles hide invisible debts, with concentrated industry risks that could shock the entire US tech sector once funding tightens.

Nigel Green, founder and CEO of financial advisory firm deVere Group, noted: "Previously, chips were never seen as a financeable long‑term asset, because they depreciate so quickly and lose value as soon as a new generation appears. Converting them into collateral for institutional loans presupposes that the underlying asset can truly retain its value over time." Jack Ablin, founding partner of Cresset Capital, put it even more bluntly: "If you are a debt investor taking compute as collateral, historically, the shelf life of such assets is about as short as lettuce."

Jensen Huang’s response came swiftly. Facing a barrage of criticism, he responded through official statements, media interviews, and a long post on social media platform X, seeking to refute the "circular financing" allegations. He pointed out that NVIDIA only provides "up to 25% of the project size" in financing support, while the decision‑making power and independent due diligence are borne by the financial institutions. He wrote on X: "Our role is to help unlock a large pool of independent capital while maintaining disciplined risk exposure."

Huang emphasised that real demand comes from frontier AI labs, AI‑native cloud service providers, and sovereign AI projects. He pointed out that previous‑generation chips like the A100 are still widely pre‑ordered by clients years after their release, proving that software upgrades (such as CUDA) can extend the economic life of hardware, making compute assets persistently profitable. Morgan Stanley also analysed in a research report that NVIDIA’s provision of a residual value support mechanism of no more than 25% can, in theory, effectively alleviate circular financing concerns, while opening up an almost 100% gross margin annuity‑style revenue‑sharing channel for NVIDIA, potentially bringing over $10 billion in shared revenue by fiscal 2029.

From a broader perspective, a data centre built today may take more than a decade to recoup its investment; the chips installed inside, however, may face competition from new generations in just a few years – NVIDIA now updates its chip architecture annually rather than every two years, meaning each new generation quickly erodes the resale value of older chips. Even if demand for compute continues to grow, factors such as compute pricing, utilisation rates, energy costs, and technological substitution will constantly rewrite the initial return projections.

Hu Yanping, a distinguished professor at Shanghai University of Finance and Economics, told The Paper that the sharp market fluctuations precisely reflect rising investor concerns about the long‑term sustainability of this leveraged capital model and the potential credit risks behind the massive financing. Although circular financing is a common practice in capital markets, against the backdrop of huge debt and uncertain commercial returns, risk vigilance has been raised to its highest level across the board.

China’s Computing Industry Also Heats Up

While capital manoeuvres were unfolding across the Pacific, China’s AI computing industry chain also experienced a dual explosion of performance and technology this week.

Earnings releases were dense. Cambricon, a leading AI chip company, reported in its 2026 semi‑annual results that revenue reached RMB 5.996 billion, up 108.13% year‑on‑year; net profit was RMB 2.311 billion, up 122.61%, marking two consecutive quarters of billion‑yuan‑level profitability. Foxconn Industrial Internet (FII), a leading AI server maker, posted revenue of RMB 557.86 billion for the first half, up 54.6%, and net profit of RMB 23.74 billion, up 96%. Within that, cloud‑service AI server revenue grew 2.3 times year‑on‑year, and GPU AI rack shipments increased 3.2 times. The memory sector saw explosive growth, with Jiangbolong (Longsys) reporting a net profit increase of over 700 times.

On the technology front, moves were equally frequent. On August 10, Alibaba Cloud announced that based on its fully modular design architecture CUBE 5.0, it had shortened the delivery cycle of large‑scale AIDC data centres to 100 days. According to Alibaba Cloud, the 100‑day delivery cycle includes 30 days of factory production and site preparation, 50 days of on‑site installation, and 20 days of commissioning and testing. By comparison, traditional delivery cycles in China range from 6 to 12 months, and in the US from 12 to 18 months; construction costs were reduced by over 10% compared to the previous generation. Alibaba Cloud plans to use this to more than double its global data centre capacity. On August 12, Alibaba Cloud’s Lingzhen Zhenwu M890 super‑node instance officially went live, becoming the first super‑node computing form in China capable of successfully running large models with over 2 trillion parameters.

On the evening of August 12, DeepSeek updated its API documentation with the DeepSeek‑V4‑Pro‑0813 model version. This release marked another step forward for domestic large‑model capabilities, and the exponential increase in token calls directly ignited the sentiment of the A‑share semiconductor and computing chain. On the morning of August 13, the computing‑power leasing sector saw a surge of limit‑up stocks, with Liyang Technology hitting the 20% daily limit and Chengdi Xiangjiang posting its third consecutive limit‑up.

The Next Stop for Compute

From NVIDIA joining forces with Wall Street to unlock $500 billion, to the comprehensive performance boom in China’s computing industry chain, this week in August 2026 has made it clear: AI compute is no longer a purely technological issue, but is rapidly evolving into a core infrastructure asset contested by global capital.

Jensen Huang said: "Fundamentally, for our industry and this computing model, the difference is that compute has now become infrastructure, just like electricity and the internet. So we must look at it and lay it out as infrastructure. In the future, every company will be driven by it, and every country will build its own computing infrastructure. This will be an unprecedented and monumental wave of infrastructure construction."

However, the leap from "technology product" to "infrastructure asset" will not be smooth. According to Morgan Stanley estimates, between 2026 and 2028, hyperscale cloud providers will spend as much as $3.5 trillion on AI infrastructure, with the long‑term total potentially exceeding $8 trillion. Financing intentions do not mean project approvals; project approvals do not mean data centres are built; and data centres being built do not mean compute is fully utilised and generates cash returns. The above cooperation still requires formal agreements to be signed by all parties before it can be implemented. The grand financing vision does not mean industry risks disappear; subsequent project approvals, interest rate pricing, default handling, and hardware asset valuation all need to be refined through multiple rounds of negotiation.

Danny Hewson, financial analysis director at UK investment platform AJ Bell, added that this financing model is essentially "debt on top of debt." The biggest market doubt is whether the massive capital injection will ultimately yield sustainable returns, or whether it will turn into unproductive "white elephants." Xiang Ligang, Director of the China Information Consumption Alliance, pointed out that if cloud giants like Google and Microsoft were the biggest beneficiaries of real compute shortages, their stock prices should have risen; instead, NVIDIA’s push for financing solutions raises a red flag, suggesting that potential buyers may be facing balance‑sheet difficulties.

As Wall Street begins to scrutinise AI investments through the dual microscope of "finance and physics," the real test for the computing industry has only just begun.