"Selling GPUs and Guaranteeing Demand" Nvidia Moves to Lock In Customers as Big Tech's In-House Chip Push Intensifies Ecosystem Defense

Nvidia seeks to expand market share and generate recurring revenue through GPU demand guarantees
Circular transaction model spreads across AI infrastructure market, with Google joining Nvidia
"In-House Chips Instead of Costly Nvidia GPUs" Big Tech Accelerates Shift Away from Nvidia

Nvidia has rolled out a new financing model targeting artificial intelligence (AI) cloud providers. Rather than simply selling graphics processing units (GPUs), the company is lowering customers' upfront investment burden by re-leasing idle computing capacity while simultaneously reinforcing its market position. The market is closely watching whether Nvidia's latest strategy can meaningfully slow the growing trend among major technology companies to reduce their dependence on the chipmaker.

Nvidia's New Financing Model

According to a report by The Information on July 2 (local time), Nvidia has recently introduced a new revenue-sharing financing model for cloud providers. The core of the program extends beyond GPU sales, with Nvidia re-leasing unused GPU capacity from cloud operators at a predetermined price to help fill demand gaps. The initiative is designed to enable emerging cloud providers with limited operating histories or weaker credit profiles to expand Nvidia GPU-based infrastructure more rapidly. In return, Nvidia receives a share of the cloud service revenue generated by those GPUs, in addition to its traditional chip sales revenue.

SharonAI and Firmus have been cited as early adopters of the model. SharonAI plans to deploy up to 40,000 Nvidia Grace Blackwell GB300 GPUs under the program. James Manning, SharonAI's co-founder and chief executive officer, said, "This strategic collaboration with Nvidia marks a significant milestone in SharonAI's mission to deliver sovereign, hyperscale AI compute infrastructure." Firmus also plans to deploy up to 170,000 Nvidia GPUs in Batam, Indonesia.

From Nvidia's perspective, the model represents far more than an expansion of GPU sales channels. Once cloud providers build their initial infrastructure around Nvidia's architecture and GPUs, they are more likely to remain tied to Nvidia's networking products, software, and operating environment. Data centers comprise integrated infrastructure combining servers, networking, cooling systems, software stacks, and operational optimization, making ecosystem lock-in a meaningful competitive advantage in the intensifying AI infrastructure race. Some analysts also believe the model could influence the memory semiconductor market, as long-term infrastructure contracts supporting Nvidia's ecosystem would also underpin demand for high-bandwidth memory (HBM) and DRAM.

Capital Circulating Throughout the AI Market

This is not the first time Nvidia has leveraged such a circular structure in the AI infrastructure market. A representative example is its relationship with OpenAI and CoreWeave. OpenAI completed a $6.6 billion funding round in October 2024, with Nvidia participating as one of the investors. Nvidia subsequently announced a strategic partnership with OpenAI in September last year to build at least 10 gigawatts (GW) of data center infrastructure, committing to invest up to $100 billion alongside the phased construction of the system.

That capital does not remain solely within OpenAI. The company sources part of its computing capacity through cloud providers such as CoreWeave, while Nvidia is involved with CoreWeave as both an investor and supplier.

Within this structure, the three companies have effectively formed a circular transaction model. For example, CoreWeave disclosed in September last year that it had signed a long-term cloud capacity agreement worth up to $6.3 billion with Nvidia. Under the agreement, if CoreWeave's data center capacity is not fully utilized by its own customers, Nvidia must purchase the remaining unsold capacity through April 2032. OpenAI procures computing capacity from CoreWeave, CoreWeave finances infrastructure expansion by leveraging Nvidia GPUs acquired with those funds, and Nvidia in turn purchases CoreWeave's idle computing capacity.

Google has recently adopted a similar strategy. According to The Wall Street Journal, Google recently provided $3.2 billion in financial guarantees for the Lake Mariner data center cluster in western New York. The facility will house both Tensor Processing Unit (TPU)-based clusters using Google's proprietary AI accelerators and GPU clusters powered by AMD and Nvidia. The operator plans to lease the TPU-based cluster to Anthropic. Google is also providing financing for Anthropic's planned $7 billion data center project in Louisiana and has extended an additional $1.733 billion in financial guarantees to support AI computing capacity lease agreements in Colorado City, Texas.

Major Big Tech Firms Move to Reduce Dependence on Nvidia

The key question going forward is whether these emerging alliance strategies will materially reshape the competitive landscape. In particular, the market is focused on whether Nvidia can slow the industry's accelerating shift away from its ecosystem. The AI chip market, once dominated by Nvidia's general-purpose GPUs, is increasingly diversifying around application-specific integrated circuits (ASICs) developed by major technology companies.

Google already relies heavily on TPUs as the backbone of its AI infrastructure, while Microsoft unveiled its proprietary Maia 200 inference accelerator in January with the goal of lowering AI token generation costs. Amazon Web Services (AWS) plans to improve training and inference cost efficiency through its 3-nanometer Trainium3 chips, while Meta continues expanding deployment of its in-house MTIA AI processors. OpenAI also unveiled its first proprietary inference AI chip, "Jalapeño," developed with Broadcom, last month and plans to begin deploying it in data centers later this year.

One of the primary drivers behind Big Tech's accelerating investment in proprietary AI chips is the enormous cost of Nvidia GPUs. According to market estimates, Nvidia's H100 GPUs have typically sold for between $25,000 and $40,000 per unit, significantly higher than AMD's MI300X, which generally sells for between $10,000 and $15,000 each. Costs rise even further with next-generation H200, B200, and GB200 systems, where server- and rack-level pricing increases substantially. Nvidia's next-generation Vera Rubin-based NVL72 rack system is estimated to cost between $5 million and $7 million per rack. As a result, dependence on Nvidia increasingly translates directly into higher data center capital expenditures.

Meanwhile, some companies have gone a step further by building cloud data centers entirely without Nvidia products. TensorWave, headquartered in Las Vegas, Nevada, provides cloud services using AMD chips instead. The company has moved away from Nvidia's CUDA-centric ecosystem by adopting AMD's ROCm software platform. TensorWave currently operates 10,000 GPUs across data centers in Pennsylvania, Arizona, and Florida. It has already secured leases for 500 megawatts (MW) of data center capacity and plans to expand that footprint to 2 gigawatts (GW).