On August 23, Alibaba announced a placement of 710 million new shares in Hong Kong to raise approximately 80 billion Hong Kong dollars, equivalent to 10.2 billion US dollars. The company explicitly stated that all net proceeds would go toward artificial intelligence initiatives, including the expansion and upgrade of AI infrastructure.
Two months earlier, Alphabet, Google’s parent company, completed an even larger‑scale financing round. In June, it secured roughly 49.6 billion US dollars in net equity financing through common stock, mandatory convertible preferred shares, and a 10‑billion‑dollar private placement targeted at Berkshire Hathaway, while retaining an at‑the‑market offering capacity of up to 40 billion US dollars. Alphabet confirmed that the raised funds would support the expansion of its AI infrastructure and global computing capacity. The firm also issued around 20.3 billion US dollars worth of senior unsecured bonds in the second quarter.
Meanwhile, both OpenAI and Anthropic have confidentially filed for initial public offerings in the United States. Latest reports indicate Anthropic is targeting a valuation of approximately 2 trillion US dollars and plans to raise as much as 100 billion US dollars, surpassing SpaceX which went public recently.
It is clear that the artificial intelligence industry has officially entered a true capital‑intensive phase.
Market projections for AI‑related capital expenditures are also being revised upward rapidly.
Goldman Sachs forecasts that major global AI infrastructure providers will record around 755 billion US dollars in capital spending in 2026, with the figure potentially climbing to 920 billion US dollars in 2027. Leading Chinese internet firms alone are expected to invest more than 70 billion US dollars in AI‑related projects throughout 2026.
America’s five major hyperscalers, Alphabet, Microsoft, Amazon, Meta and Oracle, are pushing capital expenditure levels to previously unimaginable heights.
|**Company**|**Key Metric for 2026 AI Infrastructure Spending**|
|:-|:-|
|Alphabet|FY2026 CAPEX guidance: US$195B–205B|
|Amazon|CAPEX of approximately US$200B|
|Microsoft|Calendar-year 2026 CAPEX of approximately US$190B|
|Meta|CAPEX guidance: US$130B–145B|
|Alibaba|Three-year AI infrastructure plan: RMB 380B|
|Tencent|Q2 standalone CAPEX: RMB 52.8B|
A critical question therefore emerges. Where will this massive pool of funding come from? This will become a core issue shaping the next stage of US‑China AI competition.
Over the past two decades, internet companies were widely favored by capital markets largely because of their asset‑light business models. Services such as Google Search, Facebook, WeChat, Taobao and various SaaS platforms required ongoing research and development, yet incurred low marginal costs when serving larger user bases. Successful internet firms accordingly delivered high gross margins, strong free cash flow and relatively modest capital expenditure requirements.
Artificial intelligence is reshaping this fundamental structure. Training large models demands GPUs, and running inference workloads requires even more of these chips. GPUs in turn rely on data centers, power supplies, network infrastructure, liquid cooling facilities and land. While top‑tier engineering talent remains essential, sustained access to tens or even hundreds of billions of dollars in capital has become equally vital.
Google, Amazon, Microsoft and Meta initially fund their expansion using substantial operating cash flow generated internally. When additional capital is needed, they can issue new equity and investment‑grade bonds. Higher‑risk projects can draw financing from high‑yield debt instruments, private credit providers, infrastructure funds, real estate funds and structured project finance arrangements.
Model developers such as OpenAI and Anthropic secure equity capital from venture capital firms, sovereign wealth funds, strategic corporate backers and eventual public‑market IPOs. Upstream suppliers including Nvidia also participate through direct investments, long‑term procurement agreements and credit support packages.
Large‑scale AI projects no longer need to be fully financed by a single company. If OpenAI requires 10 billion US dollars to build a computing cluster, it can issue a long‑term capacity purchase commitment, operators such as Oracle can construct and run the corresponding data center, banks can provide senior debt financing, private credit funds can absorb higher‑risk tranches, and infrastructure funds can hold ownership of the underlying project assets. OpenAI then repays investors over time through long‑term lease contracts or computing‑capacity agreements.
A single project is thereby split into multiple risk tiers, each borne by a different class of investor. This represents one of the most significant strengths of the United States capital market, its ability to decompose and distribute risk across diverse participants.
China also possesses an enormous pool of domestic capital. By 2026, assets under management by Chinese private equity funds stand at roughly 23 trillion renminbi, with PE and VC investors having deployed approximately 4.7 trillion renminbi into new‑economy ventures cumulatively. Additional sources of funding include China’s large banking sector, insurance capital, local government industrial funds, state‑owned capital and corporate investment arms belonging to major technology conglomerates.
The real divergence lies in capital structure and risk appetite. Frontier‑stage artificial intelligence projects carry distinctive characteristics, prolonged operating losses, fast technological iteration cycles, limited collateral value, uncertain cash flow streams and the requirement for massive upfront computing‑related investment. Companies working on such technologies need abundant equity investors willing to accept long‑term risk exposure.
KPMG data shows US venture capital investment reached 267.2 billion US dollars in the first quarter of 2026. Even after excluding OpenAI’s single 122‑billion‑dollar financing round, investment volumes remain near historical highs. During the same period, China’s VC funding totalled only around 20 billion US dollars.
This gap explains recent policy adjustments in China’s capital markets designed to bring public‑market funding into technology firms at an earlier growth stage. Starting in 2025, the STAR Market broadened the scope of its fifth set of listing criteria to include frontier‑technology enterprises focused on artificial intelligence, commercial aerospace and low‑altitude economy sectors. In June 2026, the Shanghai Stock Exchange issued further clarification stating that large‑model AI companies lacking substantial revenue but holding leading‑edge technology and proven large‑scale applications qualify for this listing pathway.
This reform addresses a pressing practical challenge for AI businesses. Firms may still be unprofitable yet require billions of US dollars to sustain research, development and computing‑capacity spending. Domestic GPU developers including MooreThreads and Muxi have begun accessing China’s public capital markets, while Enflame plans to raise funds on the STAR Market to support next‑generation AI chip and software‑hardware research. Separately, several Chinese AI companies are tapping overseas capital markets to diversify financing channels. For instance, NASDAQ‑listed MAAS is pursuing a new financing round of up to 500 million US dollars, alongside its planned Stars Distributed Intelligent Computing Center project with total potential investment reaching 5 billion renminbi.
Unlike the United States, which relies on market‑driven capital to divide and spread risk, China channels sustained funding toward semiconductors, computing resources and core technologies through industrial policies, state‑backed capital and capital‑market reforms. The contrast between these two systems has grown increasingly distinct.
America’s greatest competitive advantage stems from its deep, sophisticated global capital ecosystem. A 1000‑billion‑dollar AI initiative can be separated into multiple risk layers, with costs shared by technology corporations, bondholders, private‑credit investors, real estate funds, infrastructure funds, pension schemes and sovereign wealth funds.
China’s strength lies in its capacity to mobilise large‑scale resources toward strategically important priorities via industrial policy, national and local government investment, and coordinated infrastructure planning. The two approaches can be summarised concisely. The United States excels at risk dispersion, while China excels at resource concentration. Each model carries distinct advantages and inherent costs.
Rising leverage levels and growing financing complexity for AI infrastructure represent risks for the United States. If future computing‑capacity demand falls short of forecasts, long‑term supply contracts, outstanding debt and data‑centre assets could come under significant financial pressure.
China faces challenges related to capital allocation efficiency. Simultaneous investment by industrial funds, local governments and public markets into a narrow set of strategic sectors can trigger duplicated spending, valuation bubbles and misallocated resources.
For these reasons, future competition in artificial intelligence cannot be judged solely by the technical sophistication of large models. Models run on GPUs, GPUs depend on semiconductor fabs, and data centres require power grids, land and debt financing. Every layer of this stack ultimately demands capital backing.
As AI investment scales from hundreds of billions into trillions of US dollars, the true contest between the United States and China will centre on which side can convert tomorrow’s AI revenue streams into today’s chips, computing power, electricity supply and research expenditure, at a lower cost of capital, across longer investment horizons and with superior resource‑allocation efficiency. The deeper‑level competition will determine which of the two systems, covering capital markets, industrial policy and resource allocation, is better equipped to support an AI investment cycle lasting more than a decade.