To The Moon

Chipmaker Nvidia Corporation ($NVIDIA(NVDA)$) shows – once again – solid results for the third quarter (Q3) of its Fiscal Year (FY) 2026 in its release after markets closed on the 19th of November, leading to the customary uptick in the post-trading session and the pre-trading session the next day estimated to be mostly from Asian and European investors.

American investors – on the other hand – likely wouldn't gloam on to the stock with the same enthusiasm, despite the trends in key line items. 

Trend Analysis

Relative to the trends seen during Q2, there have been a few subtle shifts in product category trends: 

"Compute and Networking" has pulled back from 92% to that seen in FY 2025 while its operating income share remains largely unchanged, possibly indicating some adjustments in unit price and/or reduction in average product costs. 

In the first half (H1) of the current FY, net income per share was trending to close the FY out with a 24% growth in year-on-year (YoY) terms, which was significantly lower than in the past two FYs. This remains largely unchanged in the first nine months (9M) of the current FY.

A similar slowdown in growth is visible which went from a trend of 40% growth over previous FY as of H1 2026 to a little over 17% as of 9M 2026. 

Stock volatility might have contributed to the slight slowdown in the growth in stock-based compensation from 30% to around 25%

In terms of geographical sources of revenue, Singapore had to be separated out of "Other Countries" in the FY 2025 report and made a significant source - bringing with it suspicions (along with investigations that ascribed some likelihood) that it was used as a venue for "re-shipments" of technology proscribed from reaching China under restrictions enacted into effect under the previous U.S. administration and which even currently has bipartisan support. 

As of 9M 2025, Singapore seems to have been folded back into "Other Countries".

Both Taiwan and "China (+ Hong Kong)" now see a 3% and 2% jump from H1 2026 levels while "Other Countries" now command a scant 2% share - a far cry from the 13% share (excluding Singapore) seen in FY 2023.

The company's inventory numbers are also quite interesting: 

"Raw Materials" jump from 12% share of inventory in H1 2026 to 21% while finished goods goes down to 35% from 58% across the same period. Growth in total inventory has slipped from 48% as of H1 2026 to 32% in 9M 2026. 

A belief that a company would replicate strong year-on-year growth within a high-growth sector generally makes it a "high-conviction" stock, which Nvidia has been considered a key element of for several years now. From the face of it, growth trends indicate a net slowdown. 

Of course, AI proponents have been seemingly in overdrive this earnings season, with massive projections. These projections are now being questioned in what could be considered "AI Deal Fatigue" or simply general wariness. 

Nvidia vs the AI Hype

Some reports outline that investing in AI now accounts for at least 40% of the share of US GDP ("Gross Domestic Product") growth this year, and AI companies are estimated to have accounted for 80% of the gains in the U.S. market so far this year. Some analysts estimate that around half of all AI infrastructure capital expenditures now result in the purchase of Nvidia's products and services.

Helping fuel this purchasing frenzy is the "Circular AI Deal Machine" - a massive network of "circular deals" worth several trillions of dollars predominantly around Nvidia and OpenAI – which typically sees a pattern of purchases of products and services following an investment. This serial money machine is now nearly inextricably linked to every major American tech firm and leading AI startup.

An example of the "Circular AI Deal Machine" is the recent announcement by Nvidia's ostensible rival AMD: on the 6th of October, AMD announced a “strategic partnership” with OpenAI wherein the latter will purchase 6 Gigawatts of computing hardware in a deal that AMD states will generate “tens of billions of dollars in revenue” for itself. However, the deal also includes the option for OpenAI to receive 160 million shares of AMD (around 0.1% of its total shares outstanding) structured to vest as specific milestones are achieved, AMD achieving certain share-price targets and OpenAI achieving certain technical and commercial milestones.

Since shares are liquid instruments, vesting of shares essentially can be considered a form of payment. Since these are traded instruments, there is no present fixed value. Therefore, a customer receiving payment while making a payment to receive goods and services raises two questions:

1. What is the true value of the payment received by the chipmaker for goods and services and is this reflected in net income and earnings per share?

2. Is the demand for goods and services being driven by the product's qualities, by actual persistent demand or by the payoff from investor hype?

The driving narrative that essentially makes the entire U.S. economy one big bet on AI is the notion that it would deliver a significant boost to productivity growth by delivering cost savings in the long term by requiring fewer workers. For this to be even close to feasible for routine jobs is reproducibility. The ability for AI to be reproducible has been called into question by Thinking Machines Lab, an AI startup currently valued at around $12 billion founded and led by Mira Murati, a former CTO of OpenAI. 

AI workloads are directed through kernels on Graphical Processing Units (GPUs) that are designed for highly parallel tasks – such as processing large datasets simultaneously – and are executed by numerous threads on the GPU's cores. A blog posted by Thinking Machine Labs in September revealed that current GPU kernels aren't "batch-invariant", i.e. the result produced varies based on the "batch size" or the size/grouping of queries being made to the system. Inference operations tend to vary their computation strategies and introduce shifts that snowball into divergent outputs over iterations. In simple terms, the same query made repeatedly could (or inevitably would) produce different results, which undermines the idea of AI producing "reproducible" results. 

Thinking Machine Labs posits that "batch-invariant" kernels would lead to higher consistency; however, this comes at a cost to speed and the operational efficiency of GPUs that are designed to be power-efficient via advanced architecture. On top of that, while consistency might be achieved this way, accuracy might not be. AI models frequently "hallucinate" or race down the trail to arrive at wrong conclusions (as many users of AI tools might have noticed). 

If batch-invariant kernels solve for one problem, numerous real-world consequences arise: datacenters might need to have even more compute resources - along with power, et al - to simply produce consistent results. At a certain cost point, it is possible that the human worker might simply be cheaper. 

Earlier in November, Bank of America issued research that indicated that leading companies in AI have already hit the limit on funding for datacenter buildouts based on their existing cashflows in September and October.

Consensus estimates show AI capex hitting 94% of operating cash flow minus dividends and share repurchases in 2025 and 2026, up from 76% in 2024. Firms like Meta, Oracle, and others had issued $75 billion in bonds and loans in these two months – more than double the annual average over the past decade. With these companies already fueling most of their growth through debt, it is expected that future issuances will begin to eat into profits even more acutely. 

While media reports are replete with AI causing job losses, there are already significant caution on claims of AI efficiency. Thus, most AI companies are likely closer to – if not already – spending money they don't have on facilities that haven't been built yet for customers they don't have while offering "human replacement" solutions that are liable to neither be consistent nor accurate.

At around the same time as the Bank of America release, JPMorgan also issued research that painted a rather painful picture: at current outlay, the AI industry would need to make $650 billion in annual revenue in perpetuity to deliver a mere 10% return on investments that they are expected to make through 2030. 

The bank drew parallels to the telecom and fiber buildout experience, wherein the revenue curve eventually failed to materialize at a pace that justified continued investment. 

What aided the continual rise of the AI Hype has been announcements of massive deal sizes and the iterative inflation of benefits. A month after AMD announced its "strategic partnership", AMD Lisa Su stated that this deal with OpenAI could generate more than $100 billion in revenue for the company, seemingly adding an extra zero into the projected deal value. 

Earlier in March, it was reported that OpenAI doesn’t expect to be cash-positive until 2029. In this report too, there are signs of possibly implausible value inflation: the source within OpenAI stated that the company generated $3.7 billion in revenue in 2024, expects to make $12.7 billion in revenue in 2025, $29.4 billion in 2026 and projects revenue topping a staggering $125 billion in 2029. In other words, the company claims it will achieve nearly a 3,300% increase in revenue over five years relative to 2024!

Meanwhile, OpenAI has committed to spend $1.4 trillion on computing power and equipment over the next eight years through the "Circular AI Deal Machine" with the likes of Nvidia, AMD and Oracle. Given the absolute lack of profits, it's clear that this would require heavy borrowing at a time when debt market limits are being reached. The sheer depth of its commitment might have led OpenAI CFO Sam Friar to state during the Wall Street Journal’s Tech Live conference in early November that the company is looking for an ecosystem of banks, private equity and "maybe even governmental” support that would help the company raise debt. Specifically, she said that OpenAI could seek a federal “backstop or guarantee” that could allow it to borrow more money at lower rates. 

For many seasoned observers, the echoes of the Great Financial Crisis (GFC) likely resonated strongly at this point, albeit with one key difference: while the banks approached the government after their portfolios imploded, OpenAI – or perhaps the AI industry – seemed to be reaching out before the bubble popped. Naturally, the company hurried to walk back these statements, stating that it isn't seeking government support.

There's a lot about the technicalities of AI models and the industry that retail investors (and indeed many professional investors) who are vocal champions of AI investments likely don't understand. This possible gap weighs particularly heavy in the current market: as per Federal Reserve data published in early 2024, the wealthiest 10% of Americans - likely aided by professionals - held 93% of all U.S. stocks as of the end of 2023 (i.e. right about when the AI Hype began to rise) and currently account for half of all U.S. consumer spending. In this current earnings season, a number of these individuals are making some interesting moves: Masayoshi Son's Softbank sold its entire stake of 32.1 million shares of the company for $5.83 billion, followed by Peter Thiel's hedge fund doing the same as well as billionaire Stanley Druckenmiller (who did the same for both Nvidia and Palantir).

Meanwhile, Klarna chief Sebastian Siemiatkowski — who holds shares in prominent AI companies including OpenAI, Perplexity, xAI and Cerebras through his family office Flat Capital and is an AI evangelist who claimed that Klarna used AI to cut more than half of its workforce in recent years — stated that he's "very nervous about the size of these investments in these data centres".

Too Many “Deals”, Too Many “Experts”

At a time when big-box retailers such as Home Depot, Target and Lowe's have been lowering profit outlooks and layoffs in the U.S. are steadily trending close to GFC levels in 2025:

— with October reportedly estimated as having the largest number of layoffs for that month in 22 years, AI has become a way to "explain away" the situation: director of the administration's National Economic Council Kevin Hassett stated earlier this week, "there could be a little bit of, almost, quiet time in the labor market, because firms are finding that AI is making their workers so productive that they don’t necessarily have to hire the new kids out of college and so on."

While this narrative is well in line with some of the more speculative "predictions" made earlier in the year that full AI replacement of human labour would come about by 2027 and "superintelligence" would make in appearance the next year and possibly fueled the AI Hype among investors, the reality – when considered more rationally – is a little more blase: AI, for all its cool features, might be a support or augmentation to human labour with frequent errors but definitively not a replacement (at least yet). 

The speculative frenzy behind AI also gives a little insight as to why the likes of Nvidia perpetrate the multi-trillion-dollar "Circular AI Deal Machine": compared to datacenters and AI providers, chipmakers have been pretty light in their debt load. Their strong cash flows underpin complex financial deals, which could possibly propel sales and almost definitely boost prices of their stocks due to the AI Hype. This quarter is no different and brought forth yet another circular deal: one day before the Q3 earnings release, Nvidia announced that it would invest up to $10 billion in Microsoft-backed OpenAI competitor Anthropic while Microsoft pledged up to $5 billion. In return, Anthropic will buy 1 gigawatt (GW) worth of Nvidia’s Grace Blackwell and Vera Rubin systems, $30 billion of compute power from Microsoft's Azure cloud network and contract an additional 1 GW of compute capacity.

From tariffs and wars to the AI industry, the year 2025 has had far too many mentions of "deals". At least in the AI industry, these deals are in the realms of technology that is little understood by most investors – making conversations around it rife with speculation, investors fearful of missing out, and "experts" making far too great an attribution of benefits. If/when the bubble pops, there goes the vast bulk of Nvidia's forward outlook: its current financials have removed it far from its consumer, gamer and crypto miner market that originally had a hand in its products gaining notice from the corporate client base that now almost exclusively fuels it.

There is plenty of cause to be wary of the AI industry and Nvidia's persistent deal cycle isn't helping. At the same time, many of those who advocate for it and find themselves at the top of the conversation about it are simultaneously championing it and seemingly selling off their stake in it. As a company, Nvidia sells great products and services that are highly valued to the modern economy. However, the stock seems to be heavily overvalued and prone to volatility as the hype it was capitalizing on seems to be wearing thin.

Professional investors in Europe might consider the +3x Long NVIDIA ETP ($LS 3X NVIDIA(NVD3.UK)$) and the -3X Short NVIDIA ETP ($LS -3X SHORT NVIDIA (NVDA) ETP(NV3S.UK)$) during bullish and bearish trends in NVDA’s price. To potentially capitalize on major tech stocks seemingly driving the market currently, the 5x Long Magnificent 7 ETP (LSE ticker: MAG7) and the -3x Short Magnificent 7 ETP ($-3X SHORT MAGNIFICENT 7 ETP(MAGS.UK)$) are at hand.

Furthermore, the NVIDIA Options ETP (LSE ticker: NVDI) seeks to generate monthly income by buying NVIDIA shares and selling put options on them. Also available is the Magnificent 7 Options ETP (MAGO), which invests in each Magnificent 7 constituent’s respective Options ETP in an equally-weighted manner.

***********

For broader articles that deep-dives into business and culture in Asia, visit asianomics.substack.com. Numerous new articles have been published that fully explain the rationale behind the commentary I’ve made in various media publications in diverse areas. The latest one references my comments about WeRide and Pony.AI’s Hong Kong IPO that was featured in Bloomberg News.