Home / Uncategorized / [Opinion] The AI Bubble That Isn’t a Bubble — Yet Behaves Exactly Like One – Review

[Opinion] The AI Bubble That Isn’t a Bubble — Yet Behaves Exactly Like One – Review

Lu Duowei’s “structural bubble” riff reads like a defensive memo to LPs. At this point he should sign it Lu Delusionei, and Yicai might as well print fairy tales for terrified VCs. The piece sprays assertions with no spine. “Mandatory spending” is tossed out like a law of physics.

Publisher: Yicai Global

That title is a clown car. “The AI Bubble That Isn’t a Bubble, Yet Behaves Exactly Like One” tries to sound paradoxical and just faceplants. Pick a thesis and defend it. Don’t hide mush behind wordplay.

The piece sprays assertions with no spine. “Token prices fell 20–70%.” Which tokens, which providers, what period, and what normalization? Training cost, inference cost, seat price, and unit economics are different things. If you don’t define the metric, you don’t have a number. And spare me the fake novelty about “no prior cycle” causing deflation. We’ve seen relentless price-perf collapses for decades: CPUs, GPUs, storage per GB, DRAM, bandwidth.

The “infinite supply at zero marginal cost” shtick is freshman econ. Copying a model is cheap. Running it at scale is not. You still pay for GPUs, energy, latency targets, capacity planning, finetuning, evals, monitoring, incident response, compliance, audit, and support. Supply is throttled by hardware availability, power, and people who actually know what they’re doing. Real customers buy reliability and guarantees, not vibes.

The time-mismatch handwave imagines a three-month loop from research breakthrough to money printer. In the adult world you integrate, secure, validate, sell, and deploy into messy stacks with regulators lurking. Productization, contracts, and change management eat those neat timelines for breakfast. Pretending otherwise is how you write Medium posts, not how you run a P&L.

Value-capture claims are just as shallow. The piece moans that capital chases the compute-heavy layers while value accrues elsewhere. And sometimes platforms with data, distribution, SLAs, and regulatory posture keep the rents. Lock-in, contracts, and real switching costs exist. Cloud margins didn’t fall from the sky. Quit treating investment flows like lemmings.

“Mandatory spending” is tossed out like a law of physics. Show a single grounded case where a rational firm was forced to burn cash to “keep up” and couldn’t pause, partner, or buy later in M&A once the smoke cleared. Markets reprice. The world doesn’t follow your deck.

And yes, the packaging screams unserious. Yicai slapped a broken lead image on it and a garish font that looks like they outsourced the CMS to interns. If that’s your editorial standard, why should anyone trust your numbers? Our editor called it what it is: prestige signaling for triggered fund managers clutching spreadsheets and mistaking anxiety for insight. Lu Duowei’s “structural bubble” riff reads like a defensive memo to LPs. At this point he should sign it Lu Delusionei, and Yicai might as well print fairy tales for terrified VCs.

There’s a real question buried here: can AI capability gains outrun monetization and skew incentives for a while. Sure. But this article is posture, not proof. Bring definitions, sources, and a model that survives contact with reality. Otherwise, it’s just another sloppy press release pretending to be analysis.

Tagged:

Leave a Reply