Learning from Nvidia’s “good example,” both Google and Broadcom have started to “follow suit,” beginning a “closed loop” for AI chips.

Nvidia’s AI chip business empire, built with financial guarantees and revolving financing, is now being systematically copied by its strongest rivals. Google and Broadcom are using their respective balance sheets as weapons, leveraging Nvidia’s “playbook” to make a strong breakthrough in the AI computing power market.

Google is directly challenging Nvidia’s dominant status in AI chips by emulating its approach. On June 18, The Wall Street Journal reported that Google is providing financial guarantees for data center projects and using revolving financing to facilitate chip procurement, replicating Nvidia’s customer lock-in strategy. Backed by a planned $85 billion equity financing, Google is aggressively targeting external computing power customers.

Meanwhile, a Wallstreetcn article states that Broadcom is following a similar path—teaming up with Apollo and Blackstone to establish a $35 billion AI computing power financing platform. By offering deficit guarantees for senior bonds through its own credit, Broadcom aims to create a novel closed-loop financing bundle involving chip manufacturers, private credit, and AI computing demand, directly targeting Nvidia’s over 90% market share in AI chips.

The core logic of this challenge is: When computing power scarcity becomes the decisive factor in the AI race, whoever can help clients solve financing issues, can win chip orders. Analysts point out that the market significance of these moves is: Nvidia's longstanding business model of reducing data center financing costs through financial guarantees and leveraging revolving investments for chip procurement is now being systematically transplanted by Google and Broadcom.

This trend not only means that the competitive landscape of the AI chip market is being reshaped, but also signals that the deep integration of private credit and AI infrastructure financing is becoming the new industry norm, posing a substantial challenge to Nvidia’s over 90% market share.

Google copies Nvidia’s playbook: Financial guarantees for chip orders

According to reports, Google is systematically replicating Nvidia’s core business strategy—offering financial guarantees to help data centers access lower-cost debt financing, and arranging “revolving financing” so that part of its investment returns as chip purchases.

The most representative case is the Lake Mariner project on the southern shore of Lake Ontario, New York. Google has provided $3.2 billion in financial guarantees for this AI data center cluster. The project is jointly developed by TeraWulf and FluidStack, which is backed by Google, and its computing power will be leased to AI giant Anthropic. TeraWulf co-founder and CTO Nazar Khan said: “All these well-capitalized companies believe that the market around computing power will create huge value, and they don’t want to be left behind.”

Google’s financial guarantee layout goes far beyond just this. It is reported that, according to insiders, Google has also backed another Anthropic project—the $7 billion River Bend project near Baton Rouge, Louisiana—and has provided an additional $1.4 billion guarantee for an AI computing power leasing project in Colorado City, Texas.

On a broader strategic level, Google recently reached a $5 billion agreement with Blackstone, planning to set up a new cloud service company, directly targeting CoreWeave and Nebius—two cloud service providers exclusively using Nvidia hardware. Bernstein tech analyst Stacy Rasgon commented:

“They are clearly more opportunistic than a few years ago, more proactive in monetizing their assets. But a few years ago, this opportunity simply didn’t exist. Now, all we hear is that there isn’t enough computing power.”

Direct sales of TPU: From internal tool to external competitive weapon

Google’s commercialization path for its self-developed AI chip, the TPU (Tensor Processing Unit), evolved from internal exclusive use to external availability and onto direct sales.

According to The Wall Street Journal, all this started in 2013. Jeff Dean, then a Google AI researcher and now DeepMind’s Chief Scientist, conducted a “thought experiment” while studying speech recognition: To launch a speech model for 100 million users, the required computing power would be double Google’s entire server count at the time. His conclusion: “We need to build dedicated hardware.”

Initially, TPUs were only used internally at Google to support AI functionalities for its search engine and other products. As external computing power demand surged, Google began opening TPUs to external clients via its Cloud platform, spurring rapid growth in cloud business. This May, Google further announced plans to directly sell TPUs to clients and launched the first TPU product specially customized for inference scenarios, expected to directly compete with Nvidia’s new Groq 3 LPU.

Mark Lohmeyer, VP of AI and compute infrastructure at Google Cloud, said that inference-optimized chips, combined with improvements in cross-system chip collaboration, have attracted new clients who had never considered TPUs before. Among them is Citadel Securities, a long-term Google Cloud user, which recently began using TPUs for some research software workloads. CTO Josh Woods said operating costs for key workloads dropped by 30%, and speed improved up to four times.

Broadcom bets on “deficit guarantee”: Trading credit for market share

Meanwhile, a Wallstreetcn article previously noted that Broadcom is trading its own credit for market share in the AI chip field, pioneering a new financing model that bundles chip manufacturers, private credit, and AI computing demand together.

Last week, Broadcom, Apollo, and Blackstone jointly announced the establishment of the “AI XPV platform,” with the first deal totaling $35 billion, financing the expansion of over 1 gigawatt of Anthropic’s compute infrastructure—the largest private credit Special Purpose Vehicle (SPV) deal to date. The core of the deal is an SPV led by Apollo’s Atlas SP Partners, which purchases chips and then leases them to Anthropic, using rental income for debt repayment.

The debt structure has three layers: $600 million A1 notes are sold to banks at 100 basis points above Treasury rates; $24 billion A2 notes are sold to institutional investors at a 5.75% yield; $4.5 billion subordinated notes lack Broadcom backing and carry an 8.5% yield; Atlas SP Partners also provides an $800 million equity tranche. The key to low-cost financing for senior bonds is Broadcom’s “deficit guarantee”—if Anthropic cannot pay and chip disposition proceeds don’t cover principal and interest, Broadcom will cover losses for A1 and A2 tranche investors.

In March, Broadcom CEO Hock E. Tan was cautious about using Broadcom’s balance sheet for such guarantees, but later changed his stance. The practical pressure for this shift was that Nvidia had started similar supplier financing to accelerate chip sales, and failing to follow suit would leave Broadcom behind in AI chip competition. Tan called this partnership “the first of many transactions to come,” and plans to provide over 20 gigawatts of compute financing for frontier AI labs through this platform by 2028, with potential chip procurement scale reaching up to $700 billion.

Nvidia’s moat: CUDA ecosystem and the “Jensen Prison”

Facing encirclement from Google and Broadcom, Nvidia still retains considerable market resilience. Its over 90% AI chip market share is underpinned by a powerful ecosystem of plug-and-play interconnected hardware and the user-friendly CUDA programming libraries.

According to reports, some emerging cloud service providers worry that deviating from Nvidia’s complete hardware stack may risk losing chip quotas. The industry calls this predicament the “Jensen Prison.” Adam Fisher, partner at Bessemer Venture Partners, said:

“Not every Nvidia cloud provider will say this. Some will say Nvidia gives them everything they need, but some are desperately hoping for alternatives, and just cannot get them from other suppliers.”

Regarding competitive challenges, Nvidia CEO Jensen Huang remains publicly indifferent. In an April podcast, he said Nvidia holds a significant lead over Google and other custom chip (ASIC) makers, and questioned the cost advantage of TPUs: “I’d like to hear them prove the TPU’s cost advantage, as it makes no sense to me.” He also emphasized that Anthropic is TPU’s only major external customer.

However, Google’s CTO for AI infrastructure Amin Vahdat has a different view. He said he does not focus on confrontation with Nvidia or any rival—Nvidia is both a competitor and a key partner, as Google's data centers also use Nvidia GPUs. “For me and us, this is not a zero-sum game, the market demand is big enough.”

Trillion-dollar capital spending ushers in new financing landscape

Google and Broadcom’s moves reflect the industry backdrop of rapidly expanding AI infrastructure financing demands.

A Wallstreetcn article states that Morgan Stanley predicts capital market financing in the US AI sector will reach $400 billion, and may exceed $1 trillion by 2028, matching about $1.8 trillion in capital expenditure demand over the next two years. Traditional banks are showing obvious strain in digesting massive AI-related debt, so private credit has become a key alternative channel.

Google announced plans this month to raise $85 billion in equity financing, mainly to support AI infrastructure construction. Reference cases include:

Meta completed a $27.3 billion SPV transaction for the Hyperion data center in Louisiana, provided by private credit firm Blue Owl, arranged by Morgan Stanley, with Meta providing a guarantee-like support; Amazon completed a nearly C$14 billion (about $10 billion) bond issuance in Canada, the largest single issue in Canadian bond market history.

For Anthropic, this arrangement is the clearest signal of its shift toward self-built computing supply, freeing itself from reliance on cloud providers like Google or Amazon. Wallstreetcn mentioned, citing The Information, that Anthropic has arranged for Google to guarantee leases for five data center facilities, securing physical space for chip installation.

These arrangements show that, in the competition for AI computing power, the interests of chip manufacturers, tech giants, and private capital are being intertwined at unprecedented depth and speed, and the “financial guarantees for market share” model pioneered by Nvidia has become the new paradigm that the entire industry is racing to follow.

Risk Warning and DisclaimerThe market has risks; investment must be cautious. This article does not constitute personal investment advice and does not take into account individual users’ unique investment goals, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this article is at one’s own risk.