Skip to content

X5H Info Portal

Bilingual coverage of industry news, policy and practice

繁中EN
Business

The US$60 billion syndicate is not just about “buying chips”

2026-10-03 X5H 編輯部 28 min read

US$60 billion AI chip financing syndicate emerges: Wall Street group reshapes computing power supply chain

According to Bloomberg, a Wall Street banking group related to Broadcom is gathering about US$60 billion in funds to finance AI chips and related key infrastructure. Companies such as Anthropic are expected to use this to obtain chips and computing resources. The news is not long, but the direction is very clear: AI computing power competition is moving from pure technology procurement to a new stage of competition in capital organization capabilities.

In the past, model companies purchased cards and cloud vendors built data centers, mainly relying on their own cash flow, equity financing or cloud service contracts. Nowadays, syndicated loans have begun to directly intervene in AI chips and infrastructure, which means that the computing power supply chain is being financialized, structured, and even securitized. For the communications industry, this is not an isolated financing news, but a signal that the computing network, optical interconnection, data center and chip ecology are being repriced at the same time.

What is the concept of US$60 billion? It is approaching the level of financing for large infrastructure projects. Traditional syndicated loans are common in telecommunications networks, energy power stations, and transportation infrastructure because such projects require large investments, long cycles, and relatively predictable cash flows. AI chip financing has entered a syndicated model, indicating that financial institutions have begun to regard computing power as an asset that can be operated in the long term and generate stable income.

But AI chips are different from power stations and optical fibers. Chips iterate and depreciate quickly, and today’s high-end accelerators may be replaced by new generation products in two or three years. Wall Street is willing to form a group at this moment. The core logic is not the chip itself, but the computing power requirements, model training contracts, cloud service revenue and ecological lock-in behind the chip. On the surface, the financing object is chips, but in essence it is the AI ​​computing power supply capacity in the next few years.

Broadcom’s place in this structure is noteworthy. Broadcom is not just a GPU manufacturer. It has deep accumulation in customized AI chips, switching chips, high-speed interconnection, Ethernet and optical communications. The larger the AI ​​cluster, the more the bottleneck is not just a single chip, but the efficiency of data transmission between chips, between cabinets, and between data centers. Whoever can provide key network and customized chip solutions will be more likely to become the organizer of computing power financing.

Why do Anthropic need syndication?

Model companies like Anthropic face a typical contradiction: computing power demand is growing rapidly, but their own cash flow and asset scale may not be able to fully cover advance purchases. If you only rely on equity financing, there will be great dilution pressure; if you only rely on the credit of cloud vendors, you will easily be restricted by a single platform. The value of syndicated financing lies in bringing multiple financial institutions, chip suppliers, computing power users and infrastructure parties to the same table to share risks and expand financial capabilities.

For model companies, locking in chips and infrastructure in advance is equivalent to locking in the rhythm of training and inference. AI competition is not only a competition of model parameters and algorithms, but also a competition of “who can continue to obtain computing power.” Once there is a gap in the supply of computing power, model iteration, product launch and customer commitment will be affected. If the $60 billion-level financing arrangement can be implemented, it will significantly change the way leading model companies obtain computing power.

More importantly, this financing may not be a simple loan. It may involve chip purchase contracts, cloud service commitments, data center leases, accounts receivable and even future computing power revenue. In other words, AI chips are changing from a capital expenditure item for enterprises to an asset that financial institutions can package, layer, and price.

What are the risks of the financialization of the computing power supply chain?

Financialization can accelerate the deployment of computing power and amplify cyclical fluctuations. First, the risk of technology iteration. AI chips depreciate quickly. If the financing term is longer than the economic life of the chip, the repayment pressure will increase in the future. Second, demand risk. If AI application revenue falls short of expectations, computing power contracts and cloud service cash flow may be under pressure. Third, concentration risk. If syndicates, chip suppliers and computing power users are highly bound, once a problem occurs in a certain link, a chain reaction may occur.

Also pay attention to the boundaries of “revolving financing”. If a complex closed loop of procurement and investment is formed between chip manufacturers, cloud manufacturers, model companies and financial institutions, it can generate huge revenue in the short term, but it may accumulate hidden risks in the long term. Wall Street’s willingness to enter the market shows that it sees the certainty of AI computing power; but where the certainty is stronger, the more we need to be wary of the excessive accumulation of leverage.

From a positive perspective, the syndicated model can reduce the capital pressure of a single company, speed up the deployment of AI infrastructure, and also allow more small and medium-sized AI companies to obtain computing power through leasing, purchasing commitments, etc. This is not necessarily a bad thing for the entire AI application ecosystem. The key lies in whether the financing structure is transparent, whether asset pricing is reasonable, and whether supervision can keep up.

What does it mean for the communications industry?

The more concentrated the AI ​​computing power is, the more important the communication network becomes. Within large-scale AI clusters, east-west traffic far exceeds that of traditional data centers. Training tasks require the collaboration of a large number of chips, and switching chips, optical modules, optical fiber connections, liquid cooling and power systems jointly determine computing efficiency. If the US$60 billion in financing eventually flows to AI chips and critical infrastructure, the beneficiaries will not only be chip companies, but also the high-speed Internet and network equipment industry chain.

Optical communication is the direct beneficiary link. 800G, 1.6T optical modules, silicon photonics, CPO, and coherent sinking are all key technologies for the expansion of AI clusters. The switching chip and Ethernet ecosystem will also benefit simultaneously. Broadcom occupies a special position in AI infrastructure precisely because of its network chips and custom ASIC capabilities, which can expand computing power from single-point chips to cluster systems.

For operators, the role is changing too. Traditional communication networks provide connections. In the future, intelligent computing center interconnection, computing power scheduling, deterministic networks and edge reasoning will allow operators to shift from “pipeline providers” to “computing power network operators”. If the AI ​​chip financing syndicate model matures, similar structured financing may also spill over to intelligent computing centers, submarine optical cables, data center interconnection and 5G-A/6G network upgrades.

Comparison and Enlightenment of the Chinese Market

China’s AI computing power construction path is different from the Wall Street syndicate model. In China, it is more jointly promoted by operators, leading cloud manufacturers, local intelligent computing centers and domestic chip companies. The main financing methods are bank credit, special bonds, industrial funds and financial leasing. Domestic AI chips such as Huawei’s Ascend, Cambrian, and Haiguang continue to iterate, and optical modules, switches, liquid cooling and other links have formed strong global competitiveness.

The gaps are mainly in high-end chip manufacturing, EDA, IP and software ecology, but the advantages lie in policy coordination, project implementation and industrial chain integrity. Chinese communications companies can continue to expand their advantages in optical interconnection, network equipment, liquid cooling, power supply and intelligent computing center operations. At the same time, China is also exploring tools such as data center REITs and computing power asset securitization. If the cash flow transparency and standardization of computing power assets can be improved in the future, there is room for improvement in financing efficiency.

The emergence of Broadcom-related syndicates reminds us that competition in AI computing power is not just a competition in chip performance, but also a competition in supply chain organization and capital mobilization capabilities. Whoever controls key network links and has stable customers and computing power contracts will be more likely to obtain large-scale, long-term financial support.

What to watch next?

First, it depends on whether the syndicate has finally completed the fundraising, and what the participating institutions, interest rates, terms and mortgage arrangements are. Second, look at whether the financing is tied to specific chip purchases, cloud contracts or computing power income. Third, see whether this model is copied by other AI chip manufacturers, cloud manufacturers and model companies. Fourth, look at how regulators assess the systemic risks of large-scale AI infrastructure financing.

Related reading

About Us · 關於我們