Search
In the previous chapter, we uncovered the truth behind tech giants being choked by “existential fear,” driving them to pour trillions of dollars into upstream hardware manufacturers without regard for cost. However, the laws of physics in the capital markets are cold and irreversible: no matter how dazzling Silicon Valley’s product launches are, or how eye-catching Jensen Huang’s leather jacket may be, every cent of Capital Expenditure (CapEx) burned by tech giants today must ultimately be replenished by the cold, hard cash of end consumers or enterprise clients.
There is no such thing as water without a source in the business world. If we drag the current AI industry chain into the daylight, it is undoubtedly the most deformed, severely top-heavy “financial food chain” in human financial history. The flow of funds and the accumulation of profits are trapped in an extremely dangerous fault line. This food chain is drastically fractured into three tiers, each harboring different calculations. Beneath this prosperous facade, a massive cash flow crisis is quietly brewing.
Apex Predators: The “Absolute Taxing Power” in the Gold Rush
At the very top of the food chain are the hardware tycoons providing the underlying computing infrastructure—led by Nvidia, and supported by TSMC and absolute dominators of High Bandwidth Memory (HBM) like SK Hynix. They are the only players in this frenzied gold rush who are genuinely putting gold into their pockets.
It is a cliché that “the first to get rich in a gold rush are always the shovel sellers,” but in this AI frenzy, the profit margins on these “shovels” are so high they defy traditional economic common sense. Nvidia’s gross margin once approached 75%, even hitting 80%. What does this mean in perspective? In traditional hardware manufacturing, even when Apple squeezes its supply chain to the absolute limit, its gross margin hovers only around 45%. A 75% margin is a dream figure usually only attainable by pure software companies with virtually no physical production costs (like Microsoft or Adobe). Yet, Nvidia is selling tangible silicon and metal. As for TSMC, the production capacity of its most advanced nodes is entirely monopolized by tech oligarchs, granting it a virtually unreasonable “absolute taxing power” over the entire global tech industry.
For these apex predators, whether the AI bubble will eventually burst is entirely irrelevant right now. The collective anxiety and existential fear of mid-stream giants like Microsoft, Google, and Meta have perfectly translated into geometrically explosive net profits on Nvidia’s financial statements.
This upstream revelry has easily given Wall Street the illusion that the entire AI industry has already figured out a perfect business model. But this is a highly deceptive accounting mirage. Fundamentally, every cent earned by upstream hardware manufacturers today is “capital expenditure” drawn against the future by mid-stream cloud giants, rather than a genuine increase in consumption from the pockets of the general public. In other words, the entity currently footing the bill for this feast is still the tech giants’ own balance sheets.
The Ticking Countdown for Mid-Stream Bookmakers: The Time Bomb of Depreciation
Moving down the financial food chain, we arrive at the very center of the pressure cooker. This is the realm of cloud bookmakers like Microsoft Azure, Google Cloud, and AWS, as well as top-tier labs like OpenAI and Anthropic, which spend hundreds of billions developing Foundation Models. They are the biggest buyers of upstream chips. On the surface, they appear to be sitting pretty, breaking down massive computing power into smaller pieces and renting it out as APIs or cloud contracts to downstream software developers, essentially collecting tolls.
But behind the glossy financial reports of these mid-stream giants lies a ticking time bomb: Depreciation.
Under corporate accounting standards, purchasing servers and GPU chips cannot be treated as a one-time daily expense; they must be capitalized and amortized over their useful life. The problem is that Moore’s Law for AI chips is still operating ruthlessly. The physical lifespan and technological iteration cycle of this astronomically priced hardware are extremely short, facing technological obsolescence in at most three to five years.
This implies a terrifying mathematical formula: if a tech giant splurges $100 billion on AI infrastructure in 2024, it must forcibly deduct at least $20 billion in depreciation expenses from its net income statement every year for the next five years. If these exorbitantly priced servers cannot squeeze out more than $20 billion in “additional net profit” for the company annually, the entire company’s earnings growth will be dragged down by this massive gravitational pull of depreciation, triggering a fierce sell-off on Wall Street.
The giants are currently gritting their teeth, using the cash flow generated by traditional “cash cows”—search engine ads, Office software subscriptions, and legacy cloud contracts—to stubbornly offset these astronomical depreciation costs. They are making an epic gamble: betting that downstream applications will explode with revolutionary money-making capabilities, funneling an endless stream of cash upwards before the batch of H100 or B200 chips currently in their hands turns into scrap metal.
The Dried-Up Downstream Pond and the $600 Billion Monetization Black Hole
However, when we turn our gaze to the bottom of the food chain—the software application layer directly facing end consumers (B2C) and enterprise clients (B2B)—the cruel reality truly emerges. This was supposed to be the flowing spring of the entire industry chain, the ultimate payer for all capital investments; yet today, it is a bottomless, dried-up pond.
Top Silicon Valley venture capital firm Sequoia Capital once conducted a highly famous macro projection. By stacking the astronomical costs of purchasing upstream GPUs, the massive electricity and cooling expenses of data centers, and the profit margins that mid-stream cloud providers must maintain, they reached a gasp-inducing conclusion: for this massive infrastructure investment to break even, the downstream AI software market must generate at least $600 billion in “purely new AI revenue” annually.
What does this look like in context? Before the AI explosion, the entire global SaaS (Software as a Service) market had developed for over a decade, yet its total size was only around $300 billion. Now, Wall Street is demanding that the AI industry forcefully squeeze double that amount of new money out of the market within a mere two or three years. In the realistic business world, this is a nearly impossible task, as the downstream sector is facing two insurmountable walls.
First is the severe “subscription fatigue” on the B2C consumer end. Aside from ChatGPT Plus, which leveraged its first-mover advantage and brand halo to successfully accumulate a cohort of die-hard users willing to pay $20 a month, the vast majority of consumer-facing AI applications face catastrophic churn rates. The general public, drawn by novelty, might download an AI writing assistant or avatar generator to play with for a month. But once the curiosity fades, how many are willing to turn this expense into an unshakable, fixed monthly cost?
Even more fatal is the “disappearance of moats.” Over the past year, countless startups utilized OpenAI’s API, slapped on a pretty user interface, and created various “wrapper apps” in an attempt to charge fees. But when Meta unleashed its open-source large model Llama, Alibaba rolled out Qwen, or OpenAI itself casually updated a new version with identical features, the pricing power of these startups instantly dropped to zero. When the market is flooded with free and powerful open-source alternatives, downstream applications simply cannot earn enough profit to pay the expensive API tolls to the upstream.
On the B2B enterprise front, the situation is equally pessimistic. The enterprise market was originally seen as the last hope for filling this $600 billion gap. Microsoft gleefully launched Microsoft 365 Copilot at $30 per user per month, aiming to reap profits from its massive enterprise client base. But the mindset of a corporate CFO is extremely cold-blooded and rational. When a multinational company with tens of thousands of employees is pitched this system, the CFO will ask only one question: “How much tangible revenue growth will this multi-million dollar annual extra investment bring to the company?”
Currently, the vast majority of AI tools remain in an auxiliary stage of “improving localized efficiency.” They can indeed help an employee write an external email faster, or summarize a fifty-page meeting transcript in ten seconds. But the cruel question arises: if employees write emails faster, will the company’s total revenue double as a result? Usually, it will not.
Under the premise that total revenue cannot increase significantly because of AI, corporate management has only one logical equation in mind to recoup this $30 monthly subscription fee: use AI to boost employee efficiency, then directly lay off a portion of redundant staff, and use the saved salaries to offset the AI software expenses. If this math doesn’t check out, or if a company is afraid or unwilling to execute layoffs, then purchasing AI services merely adds to the enterprise’s operational burden. This hesitation and reluctance have infinitely elongated the large-scale procurement cycle in the B2B market, meaning the speed of monetization is fundamentally failing to keep pace with the speed of hardware depreciation on the mid-stream cloud giants’ financial reports.
A Game of “Left Hand to Right Hand” Lacking Real Economy Injection
If we dig deeper, the few seemingly bright revenue figures in the downstream AI sector hide a much more unsettling secret: a large portion of these funds has never left the closed greenhouse of the tech bubble.
This is an “internal circulation” lacking external capital injection from the real economy. Its operational model is incredibly simple: Venture Capital (VC) firms inject tens of millions of dollars into an AI application startup; upon receiving the funding, this startup turns around and spends a majority of it on purchasing computing power and model API licenses from Microsoft Azure or Google Cloud; Microsoft and Google, upon receiving this cloud revenue, use it as capital expenditure to buy more GPUs from Nvidia.
Do you see it? This is a “left hand passing to right hand” game driven by primary market VC funding. The prosperity of the upstream and mid-stream is largely built upon frenzied blood transfusions from the venture capital circle into downstream startups, rather than cold, hard cash from traditional retail, manufacturing, or the pockets of everyday consumers. This fundamentally bears the characteristics of a Ponzi structure. Once the macroeconomic environment fluctuates, the Federal Reserve maintains high interest rates, or VCs lose patience with the AI application layer’s prolonged inability to turn a profit and halt their injections, this cycle—which has no real economy footing the bill—will snap in an instant.
The Countdown Waiting for a Killer App
At this very moment, the entire AI industry chain is perched in an extremely unstable, inverted pyramid state. Nvidia is popping champagne at the top, celebrating record profits; cloud giants are anxiously staring at ticking depreciation reports in the middle; and the downstream application layer is struggling in the mud, attempting to squeeze pennies from reluctant consumers and cold-blooded corporate CFOs.
Everyone is holding their breath, waiting for the descent of the next “Killer App”—just as the completion of 4G and 5G network rollouts gave birth to Uber and TikTok, which completely revolutionized human travel and consumption habits.
But what if two or three years pass, and an application capable of making the entire populace willingly pay up, or capable of creating entirely new business models for enterprises, fails to materialize? By then, the mid-stream cloud giants will no longer be able to explain away those massive capital expenditures to Wall Street shareholders. Chopping off the next round of GPU orders will become the only option to stop the bleeding. That exact moment will be when the AI computing bubble undergoes a chain-reaction burst.
However, if we project this from another angle: what if downstream applications actually do find the perfect monetization model in the B2B market? That scenario might be even more chilling. Because, as we just analyzed, under the current business logic, the only pathway for AI software to achieve large-scale monetization on the enterprise end—making CFOs willingly fork over the money—is to launch a ruthless purge against the middle-class white-collar workforce.
And this is exactly the cruel truth we must confront in the next chapter.
(To be continued)
All articles and images on this website are original works by the author and are copyrighted. Without permission, please do not arbitrarily reproduce, copy, or use for commercial purposes; if you wish to reproduce, please [contact this site] in advance to obtain authorization.
Disclaimer: The content of this article is for academic discussion, ideological exchange, and reference purposes only. The philosophical views mentioned herein do not represent the stance of this website or its authors, nor do they constitute any psychological, financial, or life coaching advice. Readers are solely responsible for their own choices and lifestyles. This website and its owners shall not be held liable for any direct or indirect consequences arising from reading this article.
