To finance the build-out of AI data centres during the 2020s, an unprecedented amount of capital has been mobilised in the USA in particular amid a broad competition for global technological supremacy in artificial intelligence. By September 2026, analysis showed AI capital spending was larger and faster than during any previous technological boom. Early AI investment was funded primarily with operating cash flow, but by 2026 the bond market came to dominate the financing. The surge in new AI debt increased borrowing costs across other sectors of the economy.
Citigroup forecast in December 2025 that $2.8 trillion of capital would be spent for AI data centre infrastructure in the US by 2030, while McKinsey estimated that almost $7 trillion would be spent globally by that time. In 2026, American hyperscalers such as Microsoft, Amazon and Google's parent Alphabet were expected to spend $764 billion on AI infrastructure. Their Chinese counterparts were second at $102 billion.
Vendor financing has become a prominent feature of the build-out of artificial intelligence (AI) infrastructure in the mid-2020s, as chipmakers and cloud-computing providers have extended capital, guarantees and purchase commitments to the companies that buy their hardware and services. The practice is most closely associated with Nvidia, whose graphics processing units (GPUs) power most hyperscaler training and inference, and with "frontier labs," a term used for AI developers building the most advanced large language models, principally OpenAI and Anthropic. Because hyperscalers and frontier labs are simultaneously one another's investors, suppliers and customers, commentators have described the resulting arrangements as "circular financing" or "circular deals".
Circular financing
A circular deal typically involves one companyoften a chip manufacturer or cloud provider—taking an equity stake in, or extending financing to, a firm that in turn commits to buy that same company's products or services. Nvidia has been described as "particularly aggressive" in this respect, investing in AI startups, funding infrastructure buildouts, and taking stakes in data-centre and "neocloud" companies that rent out access to its chips. Examples cited by Bloomberg include Nvidia's investments in Nscale, Nebius, CoreWeave (in which it also agreed to buy $6.3 billion of cloud services), Safe Superintelligence and Naver, as well as up to $100 billion pledged toward OpenAI (of which it ultimately contributed $30 billion to a funding round) and up to $15 billion pledged jointly with Microsoft toward Anthropic. Similar dynamics have linked Microsoft and OpenAI (which has committed to purchase $250 billion of Microsoft cloud services), Amazon and Google with Anthropic, and AMD with both OpenAI and Anthropic.
Proponents argue such arrangements function as a "virtuous circle" or flywheel: capital from a supplier lets a customer expand, which in turn generates more demand for the supplier's products, at a time when the most advanced AI chips remain scarce and companies seek to lock in supply through financing tied to long-term purchase commitments. Anthropic chief executive Dario Amodei defended the practice on similar grounds, saying that one party has capital and a commercial interest in selling chips while the other expects future revenue but lacks the up-front capital.
Critics counter that circularity can distort incentives: a company with a major shareholder among its own suppliers may be more inclined to keep purchasing that supplier's products regardless of commercial merit, and if a customer's own revenue fails to keep pace with its data-centre and hardware bills, the supplier-investor can lose twice—once as its customer stops buying, and again as the value of its equity stake falls. Bloomberg noted this risk is heightened in AI infrastructure because a relatively small number of buyers account for a large share of the market.
Circular investments
Beyond loan guarantees and lease-financing structures, chipmakers and their customers have exchanged direct equity stakes and warrants on a large scale, producing cross-shareholdings across the sector. The research firm IDC defined the resulting pattern as one "where the same capital flows simultaneously as vendor payment and equity stake," such that a company "funds its own customer's revenue while also supplying that customer's core infrastructure," producing revenue growth that IDC described as real but not cleanly separable from the investing company's own capital deployment.
Nvidia has been the most prolific corporate investor in this sector. Between 2021 and 2025 it participated in 283 funding rounds across 241 companies, roughly 85% of them AI startups, making it the fourth-largest corporate venture investor of 2023 behind Microsoft, SoftBank and Alphabet; all ten of the most heavily funded private companies on the Forbes AI 50 list, including competitors OpenAI and Anthropic, had received Nvidia investment. Its disclosed positions include a roughly $30 billion stake in OpenAI, built alongside agreements for OpenAI to purchase Nvidia systems; a 7% stake in CoreWeave alongside a $6.3 billion cloud-services purchase commitment; and $2 billion positions each in Nebius, Marvell Technology, Synopsys, Coherent and Lumentum, along with smaller stakes of up to $2.1 billion in Iren and up to $3.2 billion in Corning. Nvidia also disclosed $5 billion invested in Safe Superintelligence and $1 billion in Naver, and was a repeat investor in Mistral AI, Cohere and Perplexity AI.
Nvidia's two largest disclosed equity positions outside the AI-software sector are in fellow hardware suppliers. It finalised a roughly $5 billion, 4% stake in Intel in late 2025 (214.7 million shares at $23.28 each), tied to joint development of x86 central processing units incorporating Nvidia's NVLink interconnect and of system-on-chips combining Intel processor cores with Nvidia RTX graphics chiplets; that stake was valued at roughly $19.9 billion as of 19 August 2026 after Intel's share price rose sharply, having peaked at more than $30 billion in late June. Separately, Nvidia disclosed a stake of nearly $21 billion in SpaceX (122.8 million shares at $170.86 each) in August 2026, making it the company's sixth-largest outside investor; the position originated as a $10 billion investment in xAI that converted into SpaceX shares once SpaceX acquired xAI, and had fallen in value to roughly $17.2 billion as of 19 August 2026 as SpaceX's private share price declined. Reporting on the stake noted that SpaceX builds its own AI data centres exclusively with Nvidia chips.
Other chipmakers have taken comparable positions in their own customers. Microsoft had invested more than $13 billion in OpenAI by the mid-2020s, the largest tranche$10 billionin a 2023, alongside OpenAI's commitment to purchase $250 billion of Microsoft cloud services. Amazon and Google committed up to $4 billion and $2 billion respectively to Anthropic in 2023, and Amazon separately agreed in 2026 to invest $15 billion in OpenAI with a further $35 billion pledged conditionally. Microsoft and Nvidia jointly pledged up to $15 billion toward Anthropic in late 2025. In July 2026, AMD announced it would invest up to $5 billion in Anthropic as part of a deal under which Anthropic would deploy 2 gigawatts of AMD Instinct MI450-series GPUs in AMD's Helios rack-scale systems, with the first gigawatt due in the first half of 2027. AMD had separately issued OpenAI a warrant for up to 160 million shares of AMD common stockequivalent to roughly 10% of the companywith vesting tied to both the volume of AMD hardware OpenAI deployed and the performance of AMD's own share price.
These equity exchanges have run alongside, and in some cases overlapped with, a wider set of compute-purchase commitments. Anthropic alone entered a $1.25 billion-a-month agreement through May 2029 for exclusive use of the computing capacity at SpaceX's Colossus 1 data centre in Memphis, Tennessee; a multi-billion-dollar agreement with Amazon; a multi-gigawatt compute agreement with Google and Broadcom; and, according to CNBC, preliminary talks to lease computing power from Meta, while reporting a revenue run-rate of $47 billion by May 2026 (up from about $10 billion for the whole of 2025) and closing a funding round that month worth $965 billion.
Commentators have noted that these overlapping stakes concentrate risk as well as opportunity: because Nvidia and other chipmakers hold equity in many of the same companies that buy their hardware, a slowdown in AI spending could reduce both their product revenue and the value of their investment portfolios simultaneously, deepening rather than diversifying their exposure to a downturn in the sector. Reporting has also noted that, because Nvidia and its peers can build influence over customers and competitors through minority equity stakes rather than outright acquisitions, such arrangements can fall outside regulatory scrutiny that would apply to a full acquisition.
Summary of major disclosed stakes
+ Selected equity stakes and warrants disclosed among AI chipmakers, cloud providers and AI developers Date Investor Investee Stake / amount Quid pro quo (commercial component) Source 2019–2023 (largest tranche 2023) Microsoft OpenAI >$13 billion cumulative (incl. $10 billion in 2023) OpenAI committed to purchase $250 billion of Microsoft cloud services 2023 Amazon Anthropic Up to $4 billion Anthropic committed to use Amazon Web Services for AI training 2023 Google Anthropic Up to $2 billion Anthropic committed to use Google's chips and cloud services 2024 Nvidia Mistral AI Undisclosed (one of three funding rounds) Continued reliance on Nvidia chips 2024–2025 Nvidia OpenAI ~$30 billion contributed (from an up-to-$100 billion pledge) Agreements for OpenAI to purchase Nvidia systems 2025 Nvidia CoreWeave 7% equity stake Nvidia agreed to buy $6.3 billion of CoreWeave cloud services 2025 Nvidia Nscale, Nebius Undisclosed / ~$2 billion (Nebius) Continued use of Nvidia chips in neocloud infrastructure Late 2025 Nvidia Intel ~$5 billion (4% stake; 214.7 million shares at $23.28 each) Joint development of x86 CPUs with NVLink integration and Intel–Nvidia RTX system-on-chips Late 2025 Microsoft and Nvidia (joint) Anthropic Up to $15 billion combined Anthropic to draw on Microsoft and Nvidia chips and cloud capacity n.d. (disclosed July 2026) AMD OpenAI Warrant for up to 160 million shares (~10% of AMD) OpenAI to deploy AMD chips; vesting tied to deployment volume and AMD's share price 2026 Amazon OpenAI $15 billion invested; further $35 billion pledged conditionally Commercial agreement for OpenAI to draw on Amazon computing capacity 2026 Nvidia Safe Superintelligence $5 billion Reliance on Nvidia chips for model training 2026 Nvidia Naver $1 billion Reliance on Nvidia chips July 2026 AMD Anthropic Up to $5 billion Anthropic to deploy 2 gigawatts of AMD Instinct MI450-series GPUs in AMD Helios systems August 2026 (originated as a $10 billion xAI stake) Nvidia SpaceX ~$21 billion (122.8 million shares at $170.86 each) SpaceX builds its AI data centres exclusively with Nvidia chips n.d. (as of August 2026) Nvidia Marvell Technology, Synopsys, Coherent, Lumentum $2 billion each Continued use of Nvidia hardware and platforms n.d. (as of August 2026) Nvidia IREN Up to $2.1 billion Data-centre capacity built around Nvidia hardware n.d. (as of August 2026) Nvidia Corning Up to $3.2 billion Optical connectivity components for Nvidia-based data centres n.d. (as of August 2026) Nvidia Cohere, Perplexity AI Undisclosed (four separate rounds each) Continued reliance on Nvidia chips
Nvidia's asset-backed financing plan