As mass production of Rubin approaches, Nvidia (NVDA.US) tightens its 50% profit-sharing model, using “de-circularization” governance to ease market anxiety

Zhitong Finance APP has learned that media outlets, citing sources on Thursday local time, reported that Nvidia (NVDA.US), the "AI chip superpower" that just released explosive earnings, has suspended part of its new financing plan; the plan was initially intended to provide credit support to cloud computing companies focused on AI compute infrastructure services in exchange for a share of revenue. Sources say the chip giant temporarily pulled back the plan last week, but added that Nvidia may redesign the plan in the future or incorporate it into another initiative.

Nvidia is only suspending part of the revenue-sharing deals within the "AI Compute Partnership" program, not terminating the entire initiative, and this does not mean a decline in demand for Nvidia-related AI GPU clusters such as Vera Rubin. Nvidia has made it clear that the business model aimed at expanding compute accessibility still exists and will continue to be adjusted due to strong demand. What has truly been halted are the most controversial deal combinations—specifically those where Nvidia both sells GPU compute clusters and also guarantees unsold compute, assists in financing for clients, restricts rental customers, and takes a 50% revenue share of the excess above a set threshold.

Nvidia's revenue for the second quarter of fiscal 2027 soared 106% year-on-year to $96.2 billion, with data center business revenue climbing 117% to $89 billion and expects revenue to grow about 70% in the next fiscal year; meanwhile, Anthropic and Nscale signed a six-year, $45 billion compute procurement agreement to deploy the latest Vera Rubin platform. These figures indicate that global demand for AI infrastructure and next-generation compute clusters remains exceptionally robust. Nvidia's suspension of these deals does not signal cooling off in compute orders but rather a risk tightening of its credit support, compute leaseback, and revenue-sharing models, proactively pushing for the "de-circularization" of its most controversial compute contracts.

Financing Boundaries Tightened! Nvidia Halts Some Revenue-Sharing as $500 Billion Credit Expansion Faces Compliance Reevaluation

More precisely, Nvidia is erecting a firewall due to controversies around "circular financing," antitrust scrutiny, and excessive control over customer operations. On one hand, such a plan could promote AI chip purchases among smaller cloud service providers, but on the other, it could allow the hardware supplier to act as financier, backstop lessee, and revenue sharer. As tech leaders such as Jensen Huang, Sam Altman, and Elon Musk are set to appear at the G20 Tech Summit, this move indicates that the AI compute supercycle is shifting from mere capital competition to a new phase focused on financing quality, regulatory boundaries, and real end-user demand.

By suspending deals that combine "GPU sales + credit support + unsold compute leasebacks + excess cloud revenue sharing over 50%," Nvidia is proactively promoting "de-circularization" of compute transactions, effectively reducing the closed loop of "financing clients to buy its own chips and then guaranteeing their revenue." This allows more orders to be scrutinized by independent demand and real cash flows, while avoiding risks related to antitrust, over-controlling client operations, and "AI circular financing" damaging credit ratings.

A Nvidia spokesperson responded to the media's latest report: "This new business model to open up compute access for the rapidly growing AI ecosystem... remains active and is continually evolving due to strong demand for AI compute."

According to media reports, Nvidia took this action less than two months after announcing the plan. The initiative aimed to meet strong financing needs and AI compute resource demands for smaller AI cloud computing companies.

Nvidia's previous plan stipulated that if cloud clients could not fully rent out their compute capacity, the company would lease back the compute itself, providing a backstop for clients and making it easier for AI cloud startups to obtain the large amount of capital needed to purchase Nvidia AI GPU clusters.

Under this model, Nvidia would first book sales from its hardware, and then take a share of revenue generated by clients utilizing Nvidia compute in their cloud businesses.

Nvidia stated during this week's earnings call that, in the medium to long term, this healthy operating model could bring in tens of billions in revenue.

However, in recent months, as Nvidia continued to reinvest capital into the AI ecosystem, investors have become increasingly scrutinous and cautious, leading to rising market concerns over so-called "AI circular financing and deals" and fears that the Nvidia-led AI GPU sector might artificially inflate demand. Such concerns have fueled widespread market commentary on an "AI credit bubble burst," causing short-term selloffs in both stock prices and credit bonds for Nvidia and other leading AI compute supply chain players.

This month, the company (Nvidia) helped its clients secure $500 billion in financing from major U.S. financial institutions, and agreed to provide up to $105 billion in backstop guarantees to help OpenAI lease a large data center.

According to sources cited in the media, some senior Nvidia employees previously expressed concerns to current and potential clients that the plan might trigger antitrust scrutiny. Cloud computing customers also have sensitivities regarding how much Nvidia could dictate their business models, especially among hyperscale cloud giants.

The report states that in the plan's initial weeks, the level of control Nvidia sought to impose already angered some potential large partners.

The report adds that Nvidia told some cloud computing providers that they could only rent Nvidia AI GPUs to company-approved clients; Nvidia also stated it would prefer to allocate these compute resources to a variety of smaller or early-stage AI cloud leasing companies rather than concentrate them with one hyperscale client.

Sources said that in some aforementioned preliminary deals, once cloud service providers' revenue from Nvidia AI GPU usage exceeded a certain operational threshold in the agreement, Nvidia would claim 50% of the excess. Some large cloud companies questioned these preliminary proposals.

AI Super Bull Market Is Still Alive, but Financing Flywheel Now Constrained! Nvidia's AI GPU Compute Chain Enters the "De-circularization" Era

Nvidia's latest move is more akin to a proactive regulatory and credit risk "brake": while short-term revenue-sharing may decrease, it alleviates worries around circular financing, antitrust review, and customer operational autonomy, ultimately improving the future verifiability of its revenues—and for Nvidia's robust fundamental prospects, it is likely a neutral to slightly positive governance repair.

On the demand side, there is no turning point; instead, exponential expansion continues. Nvidia’s Q2 FY27 revenue rose 106% to $96.221 billion, data center revenue soared 117% to $89 billion, and Q3 guidance reached $108 billion (±2%). Vera Rubin demand is strong and has entered mass production. Management expects about 70% revenue growth for FY28, stressing that this figure is already constrained by memory/storage chips and advanced node manufacturing capacity from TSMC.

Simon Leopold, a senior analyst at the renowned Wall Street investment firm Raymond James, dramatically raised Nvidia’s price target from $352 to $550 in a report on Thursday. Leopold's analysis indicates upside of about 130% from current levels, implying a market value of $13 trillion. The chip giant's current market value is around $5.2 trillion.

Meanwhile, Anthropic joined with Nscale in a six-year, $45 billion compute procurement deal, securing about 460MW of power capacity and planning further deployment of the Nvidia Vera Rubin platform. These signals reflect that the global AI infrastructure race is extending beyond chip orders to core data center bottlenecks, power resources, and long-term compute contracts, with capital intensity still accelerating fast.

Market descriptions of Nvidia as an "AI central bank" or offering "Balance-Sheet-as-a-Service" accurately capture its expansion from chip supplier to ecosystem finance intermediary, but it's incorrect to view the entire $530.5 billion in off-balance sheet commitments and guarantees on Nvidia's balance sheet as actual debt.

According to Nvidia's official filings, this number combines $366 billion in future commitments, $56 billion in additional AI cloud/third-party data center commitments, and $108.5 billion as the maximum notional guarantee exposure. Of these, $279 billion is for memory and manufacturing supply commitments (some of which are cancellable, reschedulable, or adjustable), and the $105 billion OpenAI-related guarantee is only gradually realized from 2029 onward over nine construction phases. The real red flag is cash conversion: accounts receivable rose from $38.466 billion to $63.059 billion, with the top five clients accounting for 70%. Q2 operating cash flow was about $24.077 billion, noticeably lower than the $59.688 billion under U.S. GAAP net profit, but the gap was also affected by equity investment returns, inventory increases, and extended payment periods by investment-grade clients, which cannot be directly equated to customer credit deterioration.

For the entire AI compute supply chain, Nvidia's adjustments to compute revenue-sharing agreements and credit guarantees are not an industry downturn but may represent an inflection point toward greater financing discipline, reducing concerns about AI circular financing and antitrust review. The Nvidia capacity chain led by TSMC, high-bandwidth memory dominated by HBM and advanced NAND, advanced packaging, high-speed Ethernet infrastructure, photonic interconnects, data center power chains, and liquid cooling—all driven by hyperscale cloud vendors and top-tier credit clients—remain highly certain. The real pressure will fall on smaller or new cloud service providers relying on Nvidia's credit backing, lacking long-term purchase contracts and stable cash flows, as well as on high-leverage financing chains centered around speculative data center parks.