If you noticed the circularity of money flows in the above-mentioned deals, you are not alone. In essence, Nvidia is using the strength of its balance sheet to help OpenAI fund its ambitious data centre rollout and massive long-term commitments inked with cloud providers, notably Oracle. (OpenAI is still governed by a not-for-profit parent, is loss-making and burning huge amounts of cash by the day. According to recent news reports, the company’s annual recurring revenue is now running at about US$12 billion. OpenAI had previously indicated that it would not be turning a profit until at least 2029 and has projected losses to the tune of tens of billions of dollars over the next few years.)
As the lead cloud provider in Stargate, Oracle will, in turn, spend hundreds of billions of dollars to design, build, operate and maintain the core AI infrastructure. Nvidia CEO Jensen Huang said in a recent interview on CNBC that 10gw will require between four million and five million GPUs. He had also previously stated that of the US$50 billion to US$60 billion data centre cost per gw capacity, about US$35 billion typically goes to Nvidia chips and systems.
So, Nvidia by investing in OpenAI is indirectly strengthening demand for its chips, equipment and software services, since up to 70% of the capital expenditure for data centres goes back to Nvidia as revenue for chips and equipment.
Here’s the point of our article: The market is aggregating and valuing Nvidia’s capital injection, OpenAI’s cloud commitment and Oracle’s capex for data centres as three independent growth drivers and cash flow streams, at high price-earnings (PE) multiples. But it’s partially the same funds circulating among the three companies. Boosting the demand for its chips cements Nvidia’s stranglehold on advanced chips in the global market and helps it maintain extremely high margins. Reports suggest that its most advanced GPUs can command gross profit margins exceeding 90%. Nvidia’s biggest revenue generator, the compute and networking segment, has an operating margin of 71% in the latest financial year. That means a significant portion of sales is profits (earnings) and cash flow for the company. In short, there’s an element of inflated sales, profits and stock valuations.
To be sure, such “circular financing” — in which a company invests in its biggest customers who then use the money to buy its products — is nothing new. Microsoft was an early investor in OpenAI which, in return, uses and pays for its cloud compute power to train ChatGPT. Notably, Nvidia has been making tens of similar deals, taking stakes in AI infrastructure companies such as CoreWeave and UK-based Nscale. Nvidia has even provided a US$6.3 billion backstop to CoreWeave, in which its stake exceeds 5%, guaranteeing to buy any unsold cloud capacity through 2032. What’s raising eyebrows about these most recent deals is the scale of the money flows and their huge impact on stock prices and the broader market. Without question, the above announcements drove expectations about future earnings — and fed the AI hype.
Oracle’s share price surged 36% on Sept 10, briefly making co-founder and major shareholder Larry Ellison the world’s richest man. Nvidia’s market capitalisation gained some US$170 billion immediately after the US$100 billion investment in OpenAI was announced. Notice that the gain in market capitalisation exceeds the capital commitment; it’s better than getting “free money”. It is now the most valuable public listed company in the world, with a market capitalisation of more than US$4.5 trillion. OpenAI was valued at some US$300 billion in March 2025 when it raised additional funding, including from Japanese conglomerate SoftBank. Barely six months later, the company completed a deal allowing some employee-owned share sales that values the start-up at US$500 billion. That makes OpenAI the most valuable private company in the world, surpassing TikTok’s parent company ByteDance and Elon Musk’s SpaceX. In short: You spend, your value goes up; you receive, your value goes up.
See also: Transparency demands courage, clarity requires discipline — both are needed to build trust
The risks of such tight dependence among Wall Street’s largest companies are sounding the alarm. At the very least, it could give the misleading impression that demand — and, accordingly, prevailing future margins and profits expectations — is stronger than economic reality can sustain. The bigger worry is the actual ability of AI companies to generate the revenue, earnings and cash flow required to justify capital outlays of trillions of dollars. McKinsey & Co estimates a global capital outlay of nearly US$5.2 trillion for data centres in the next five years alone. Yes, the global AI market is expected to expand very rapidly but still far below the trillions in revenue needed to generate a reasonable return on investment, or ROI. Another consulting firm, Bain & Co, estimates that the world will be short US$800 billion of the US$2 trillion in annual revenue needed to fund the computing power required to meet anticipated AI demand by 2030.
Ultimately, the sustainability of the AI hype depends on the pace and breadth of adoption by the end users, primarily the enterprises and, to a lesser extent, individuals. More importantly, the enterprises’ ability to turn spending on AI investments-applications into real profits. A report published in July 2025 by MIT (The GenAI Divide: State of AI in Business 2025) found that despite US$30 billion to US$40 billion in enterprise spending on generative AI (GenAI), 95% of organisations are getting zero return. It notes in the summary: “Just 5% of integrated
AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact.”
If OpenAI’s monetisation underperforms, it could face difficulties fulfilling its financial obligations. Oracle could be left with costly and underutilised data centre assets, and demand and margins for Nvidia chips would surely drop well below prevailing market projections. And this is not taking into account the certainty that there will be innovative new competitors to Nvidia’s chips, especially from China. Without sufficient end-user demand, the Nvidia-Oracle-OpenAI bubble could collapse under its own weight, potentially triggering a domino effect across the entire AI stack — quite likely even the broader stock market, given the high weightage of the “Magnificent Seven” and AI companies like Palantir Technologies and Broadcom in major bellwether indices and the massive investments in passive funds (that track these indices).
And for the likes of OpenAI to meet market expectations of future revenue, profits and cash flows, where would their sales come from? Surely, much of future revenue must come from replacing sales now achieved by search platforms like Google and Uber (that is, GenAI advertising revenue taking the share of advertising revenue from search engines). But we see that the stock prices of dominant players in the search ecosystem are also at all-time highs. For us, something has to give and, therefore, we think current market valuations for some of these companies are “double counting” future earnings.
As we mentioned, “cash flow recycling” is not novel. Back during the heyday of the then Kuala Lumpur Stock Exchange (KLSE) in the 1990s, we saw companies engage in similar activities. In fact, I had more than a front-row seat in one such example — Aokam Perdana, once Malaysia’s largest listed timber company. Aokam had timber extraction agreements with Idris Hydraulic, which owned timber concessions in Sabah. The controlling shareholders of Aokam (Teh Soon Seng) and Idris (Ishak Ismail) were close associates. What transpired then was that Aokam manipulated its sales — including selling timber products to related companies and using their own money to look like legitimate revenue — thus creating fake fat margins and profits (on the pretext of cheap supply of logs from Idris). The inflated profits and growth supported high stock-price valuations. Aokam overvalued its timber assets based on the high PE multiples instead of the discounted cash flow (DCF) valuation that would have reflected the timber concession as a depleting asset. And major stockbrokers happily pushed up stock prices to earn more commissions. It was not so much the retail but large institutional investors that ended up with the losses.
In April 1993, a research report by Phileo Peregrine Securities (the stockbroking outfit I managed and founded) exposed the accounting irregularities and revealed that its shares were grossly overvalued based on DCF analysis — in a lone critical research report on the then stock market darling, a “must-own” stock for foreign fund managers. After Aokam’s stock price collapsed, an empty bullet casing was sent to the Phileo office.
For more stories about where money flows, click here for Capital Section
Teh bought shares in the company, driving prices back up and was able to raise more cash. As long as the company could sell more bonds-shares at higher prices to the next “sucker”, this game was a profitable cycling of cash. Over a few short years, Aokam’s shares rose from 60 sen (18 cents) to a peak of over RM31 and then fell back to 60 sen.
The company became insolvent and was eventually declared bankrupt in 1998 when it couldn’t repay RM33 million in borrowings. Teh was labelled a “fugitive” over allegations of misappropriation of RM55 million in funds belonging to Aokam and was investigated by the Securities Commission Malaysia for short-selling the company’s shares. As spectacular as Aokam’s boom and bust was, the scandals on the KLSE were tiny in comparison to the deals on Wall Street today. “Good ideas” need scale.
The Absolute Returns Portfolio gained 1% for the week ended Oct 1. Total portfolio returns now stand at 44.5% since inception.
The top three gaining stocks were CrowdStrike (+5%), SPDR Gold MiniShares Trust (+3.7%) and ChinaAMC Biotech ETF (+2.5%). On the other hand, Trip.com (-1.5%), Goldman Sachs (-0.9%) and JP Morgan (-0.9%) were the big losers.
The AI Portfolio, meanwhile, was up 0.2% for the week, lifting total portfolio retuns to 9% since inception. The biggest gainers were Datadog (+11.8%), Alibaba (+1.7%) and RoboSense (+1%) while the notable losers include Horizon Robotics (-6.7%), ServiceNow (-2.2%) and Intuit (-2.2%).
The Malaysian Portfolio was also up last week, by 0.5%, led by Kim Loong Resources (+1.3%), Maybank (+1.2%) and United Plantations (+0.8%). We added two stocks to our portfolio, Southern Cable Group (-2.1%) and Hiap Teck Venture (unchanged). Shares for LPI Capital closed 0.3% lower last week. Total portfolio returns now stand at 187.3% since inception. This portfolio is outperforming the benchmark FBM KLCI, which is down 11.4% over the same period, by a long, long way.
Disclaimer: This is a personal portfolio for information purposes only and does not constitute a recommendation or solicitation or expression of views to influence readers to buy/sell stocks, including the particular stocks mentioned herein. It does not take into account an individual investor’s particular financial situation, investment objectives, investment horizon, risk profile and/or risk preference. Our shareholders, directors and employees may have positions in or may be materially interested in any of the stocks. We may also have or have had dealings with or may provide or have provided content services to the companies mentioned in the reports
