The AI Boom's Money Is Moving in Circles

TSMC, Alphabet and Foxconn disclosures in one week show suppliers, customers and borrowers converging, and how hard that makes reading real demand.

A week when the AI trade showed its plumbing

Four disclosures in the same news cycle only make sense together.

TSMC's board approved up to $1 billion in AI investments: $500 million for OpenAI preferred stock reserved for large investors, and $500 million to buy Anthropic common shares at the company's expected IPO price, according to Reuters. Alphabet's first Hong Kong and Singapore dollar bond sales drew about $5 billion in early orders. Foxconn beat analyst estimates with a 34% rise in quarterly profit. And Nvidia's board held urgent talks about the future of its founder and chief executive.

Read separately, each item is ordinary. Read together, they describe a system in which the supplier, the customer, the borrower and the assembler keep turning out to be the same small group of institutions.

Why it matters

When a company sells a product to another company and also owns a piece of the buyer, the usual signals get harder to read. Revenue at the seller is partly a measure of demand and partly a measure of money the seller itself helped supply. The Wall Street Journal reported that analysts have warned this kind of circular financing among investors, suppliers and customers has made it difficult to evaluate real market demand for AI hardware and services. That is not a claim about whether the demand is real. It is a claim about what the numbers can tell you on their own.

The supplier becomes a shareholder

TSMC's fabs produce the vast majority of Nvidia's AI data center chips and most of Apple's custom chips, along with Apple's planned AI computing server. Most sit in Taiwan, with additional plants in Japan and China, an expanding cluster in Arizona, and one under construction in Germany. Anthropic and OpenAI both buy large quantities of hardware that TSMC manufactures.

The company can afford the bet. Ongoing AI demand has given it six consecutive quarters of record profit, including more than $23 billion in the July to September 2026 quarter, according to Reuters. For scale, Nvidia has invested $30 billion in OpenAI and $10 billion in Anthropic, among tens of billions more across other AI and robotics companies.

The borrowers diversify

Alphabet's debut Hong Kong dollar offering drew orders above HK$24 billion ($3.1 billion), including HK$12.78 billion from the banks arranging the deal. Its four-part Singapore dollar offering drew more than S$2.5 billion ($1.9 billion), including S$790 million from joint lead managers. Initial yield guidance was set at 4.60%, 5.00% and 5.50% for the 2028, 2031 and 2036 Hong Kong dollar tranches, and 2.75%, 3.25%, 3.65% and 4.00% for the 2028, 2031, 2036 and 2056 Singapore dollar tranches, per term sheets seen by Reuters. Those were early indications, not final pricing.

The company had floated minimum sizes of HK$8 billion and S$1.5 billion. The bonds settle on November 19 and are expected to be rated Aa2 by Moody's and AA+ by S&P. This follows €6 billion ($6.7 billion) of euro bonds the previous week, $25 billion of US dollar bonds in August, and A$5.5 billion ($3.8 billion) of Australian dollar bonds the same month. The point of the exercise, as the reporting puts it, is to expand and diversify a geographic debt footprint while financing AI infrastructure.

The assembler

Foxconn, formally Hon Hai Precision Industry, is Nvidia's biggest server maker and Apple's top iPhone assembler. Net profit for July to September was T$77.5 billion ($2.35 billion), against an LSEG consensus of T$69.6 billion and T$57.7 billion a year earlier. It reported a 47% year-on-year jump in third-quarter revenue in October, stuck to a forecast of strong revenue growth this year, and said it expects high double digit growth in 2027 for AI rack shipments.

Most iPhones it makes for Apple are assembled in China, but it now builds the bulk of US-bound units in India, and it is building factories in Mexico and Texas for Nvidia AI servers. Its shares have risen 4% this year, against a 49% gain on the broader Taiwan index.

Governance as a market variable

Nvidia's board, which unusually has no chair, met after reports that Jensen Huang had a romantic relationship with Kimberly Powell, the company's vice-president of healthcare and an 18-year veteran. Huang admitted two days earlier to having an affair, following reporting on links to an executive at a Singapore-based Nvidia customer whom US intelligence officials believe was recruited by China's security services. The board hired a former federal prosecutor to investigate and has so far backed Huang to remain chief executive. President Trump said the FBI was investigating and called Huang a high-quality person.

The company's shares fell 8.6% in after-hours trading, and Nasdaq 100 futures slipped about 1%. Nvidia's annual reports have contained no specific reference to succession planning. For a $4.2 trillion company at the center of AI dealmaking, that absence is itself information.

The analogy worth holding loosely

In the late 1990s, telecom carriers borrowed heavily to lay fibre, and equipment vendors booked the proceeds as revenue. The buildout was real. So was much of the capacity. What was hard to see at the time was how much of the spending depended on the same capital loop continuing. This is not a forecast, and the analogy breaks in obvious places. But it is a reminder that when buyer and seller share a balance sheet, growth figures measure conviction as much as appetite.

What we do not know

Final pricing and deal sizes for Alphabet's bonds were not available when the stories ran. The price TSMC pays for Anthropic shares depends on an IPO that has not happened. Whether Nvidia's board keeps its chief executive, and what the investigations conclude, are open. And the central question, whether demand outside these loops is broadening, is one no single quarter's earnings can settle.

Three things to try

Separate operating revenue from investment flows. When a supplier reports growth, ask who funded the purchase.

Go to primary documents. Term sheets, filings and earnings releases say what summaries round off.

Ask what would have to be true. For AI demand to be durable, someone outside the loop has to be paying for the output.

A calm close

None of this is a verdict. Circular financing is an old pattern, and it has financed genuinely useful infrastructure before. What it does demand is a certain humility about what we can see from the outside. The numbers are real. The question is what they are measuring. Holding both thoughts at once is uncomfortable, and it is probably the most honest position available.

By Nikhil Teltumbade, Director, Godrix AI Solutions.
Working on something similar? Godrix AI Solutions builds AI automation and apps for business workflows. Book a free audit.

← All posts