Orange AI argued in an Oct. 31 article that the debate over an AI bubble intensified after Friday’s U.S. market close, as a broad tech sell-off collided with a series of sharp single-stock drops.
The Nasdaq fell 2.86%. On its own, that was not an especially extreme move, the author wrote, but the reaction online was far more dramatic because several major technology names sold off hard despite reporting figures that did not look disastrous on the surface.
Intel fell 28.58% on Friday, its biggest one-day drop in 40 years. The company’s third-quarter EPS came in at 0.52, well above the market expectation of 0.40. A day earlier, Google dropped 12.64% to close at $283.82, its worst trading day since its 2004 listing, wiping out nearly $500 billion in market value in a single session. Yet its revenue missed expectations by only 0.2% to 0.4%, cloud revenue doubled, and search query growth reached its fastest pace in five years. SanDisk posted 23% quarter-on-quarter revenue growth, adjusted EPS of $49.51, up more than 40 times year on year, and held $91.1 billion in contracted orders. Its shares still fell 13.47% after earnings.
The article shifts the focus from stock moves to whether AI is creating new value
For the author, the more important question is not whether AI has a bubble. It does, the piece says, and bubbles eventually burst. The real issue is whether AI is creating incremental value right now.
One of the clearest signs cited in the article is the rise in memory and flash prices this year. Higher hardware costs, the author says, have directly weighed on PC and smartphone sales. Apple’s memory bill now accounts for 40% of iPhone costs, according to the piece, and the standard iPhone model is no longer being released. The new iPad mini sells for 5,499 yuan but comes with only 128 GB of flash storage. On the Android side, many vendors are no longer launching low-end phones in the 1,000-yuan range.
The article cites Intel’s forecast that the PC market will shrink 12% in 2026 and another 10% in 2027, with the downturn set to continue. It traces the root cause of hardware inflation to the rapid buildout of AI cloud infrastructure: AI needs cloud hardware for inference, and consumer hardware prices are being pulled higher along with it.
The piece also references a comment from Sam, who said running local AI on personal computers is a waste of memory. The author argues that this view reflects the priorities of a large-model company, one that sees value mainly through the lens of allocating more resources to AI systems.
AI’s substitution value is clear, but the author says incremental value is still missing
The article does not dispute AI’s capabilities or its economic relevance. In fact, it says AI is powerful enough to replace a large number of white-collar jobs, and that substitution value is beyond doubt.
What remains unclear, in the author’s view, is what new value AI has created beyond replacement. As of Oct. 31, 2026, the answer offered in the piece is blunt: none, or at least none that the author can clearly identify. The only possible exception mentioned is Hongguo short dramas, though the author describes that case as unusual and returns to it later in the article.
AI coding boosts output, but not revenue
The article takes direct aim at one of the strongest arguments made by AI bulls: coding. AI coding is unquestionably powerful, the author says, and software companies across the board have seen productivity rise. Code output is exploding, to the point where one year now feels like ten.
But higher productivity is not the same thing as higher value. The article argues that even if AI coding lifts productivity by 10x, product revenue has not risen by 10x, and cases where revenue even doubles are rare.
To make the point, the author uses an agricultural comparison. Blueberry output in Yunnan rose 39% this year from last year, while prices fell 50%. The official explanation quoted in the article says that a large volume of fresh fruit entered the market at once, changing the supply-demand structure and causing a cliff-like drop in prices. More supply does not guarantee more income. Scarcity is what people pay for.
That logic is then applied to AI. AI makes intelligence less scarce, the article says, and once scarcity disappears, prices fall. The price of tokens that power intelligence is falling. The price of products built on intelligence is falling. The price of jobs tied to intelligent activity is falling as well. To offset that decline, output has to rise: more tokens consumed, more products produced, more labor expended. Even then, the result may only be a rough preservation of value rather than real expansion.
The author says many friends are using AI heavily to write code and carefully managing resets to maximize usage. Products keep launching. But none of them, the article says, has actually made money. AI has made it easy to build things that previously were not worth paying for, yet that still does not amount to clear incremental value.
In conversations with friends, the author asked whether vibe coding saves time or kills time. Most answered that it kills time. That leads to one of the article’s sharper lines: AI is not really 2B or 2C, but 2me, a form of self-oriented consumption.
AI shortens the chain between question and answer, and traffic suffers
The article extends the argument to the structure of the AI economy itself. Its most unusual feature, the author says, is that there is no middleman taking a spread.
Traditional economics runs on transactions. One person earns money through trade, spends it, and that spending becomes someone else’s income. AI works differently in the author’s framing. If AI eventually becomes as cheap as electricity, users can get results directly from a model, with almost no other human participant getting paid along the way. If AI replaces the worker who would otherwise earn and spend, the article asks, who is left to consume?
The same pattern is now showing up in content distribution. Reddit just had its worst day since going public, with the stock down 29%, according to the article. Reddit CEO Steve Huffman said on the earnings call that Google’s AI Mode now gives users direct answers, so they no longer click links. 「AI won’t bring traffic, not to us and not to anyone else.」
The article describes that as a break from the internet’s basic arrangement over the past 25 years: websites supplied content to Google, and Google sent users back to those websites. With AI Mode, the system summarizes the information, the user gets the answer, and the visit never happens. In October, average weekly visits from search engines fell by more than 1 million. The number of users entering websites each day in the U.S. fell 2% over the past year, the first decline since Reddit listed in 2024.
Reddit is only one example, the author says, but an important one. If even one of the internet’s most human communities is under this kind of pressure, the outlook for other sites is unlikely to be better.
Falling model prices are squeezing relay platforms
The article also questions whether intermediary model platforms can build durable profits. From Claude Opus 4.5 to DeepSeek Flash 4.1, the price of tokens needed to complete the same task has fallen by 50x to 100x this year, the author writes, and could fall by at least another 10x next year.
There are now few relay platforms left that can offer low-cost models comparable to Luna, according to the piece. The author cites a friend in that business who said revenue still looks decent on the surface, but profit after costs is thin. Token prices have become so low that they are already below the cost of the relay servers themselves.
Microsoft and ByteDance are both channeling cash into AI infrastructure
The article then turns to corporate spending. This week, Microsoft opened voluntary separation buyouts to 60,000 employees, more than a quarter of its workforce, the author says, describing it as a softer form of layoffs.
The point is not that Microsoft is in trouble. The article says the company is making more money than before, yet is still determined to keep spending on AI. If AI investment budgets are not large enough, labor costs are cut to make room.
On the consumer side, the author points to ByteDance’s closed-loop model around Hongguo short dramas. The article describes a chain that includes Tomato Novel text, Seedance video generation, Juyin Engine smart advertising, and Hongguo’s DAU and user time spent. In the author’s telling, that creates a full loop from producer to model to recommendation to consumer.
But where does the money go in the end? The article says ByteDance made about $50 billion last year and plans to spend as much as $70 billion this year on AI infrastructure. The conclusion drawn is that hardware investors, productivity software companies, and even consumer products with strong monetization loops are all sending large amounts of cash back into AI.
The author describes the current moment as an era of frenzied, unthinking investment in AI, one in which the market is all-in on productivity while paying little attention to production relations.
The author sees the industry as standing near the eve of a bubble unwind
In the final section, the article says the current moment feels a bit like the eve of an AI bubble. It does not claim that Oct. 31 is necessarily a black Friday, and it does not try to call the exact date of a break.
Instead, the author frames the present through the history of technology cycles. In each cycle, a new technology first goes through a bubble, then a sharp decline clears out the excess, and only after that does the technology get applied to places where it creates durable value. Technology changes, the article says, but human nature does not.
From the release of ChatGPT in November 2022 to today, AI is close to completing a full four-year run, according to the piece. Anthropic is set to go public next month, and many people see that as the peak of the AI bubble. The article leaves the question open rather than answering it.
A note at the end says the piece is based only on public information, that the author does not understand investing, and that nothing in it should be taken as investment advice.


